Authors: Steven Vander Eeckt, Hugo Van hamme
Abstract: Catastrophic forgetting remains a major challenge when neural networks learn tasks sequentially. Elastic Weight Consolidation (EWC) attempts to address this problem by introducing a Bayesian-inspired regularization loss to preserve knowledge of previously learned tasks. However, EWC relies on a Laplace approximation where the Hessian is simplified to the diagonal of the Fisher information matrix, assuming uncorrelated model parameters. This overly simplistic assumption often leads to poor Hessian estimates, limiting its effectiveness. To overcome this limitation, we introduce Continual Learning with Sampled Quasi-Newton (CSQN), which leverages Quasi-Newton methods to compute more accurate Hessian approximations. CSQN captures parameter interactions beyond the diagonal without requiring architecture-specific modifications, making it applicable across diverse tasks and architectures. Experimental results across four benchmarks demonstrate that CSQN consistently outperforms EWC and other state-of-the-art baselines, including rehearsal-based methods. CSQN reduces EWC's forgetting by 50 percent and improves its performance by 8 percent on average. Notably, CSQN achieves superior results on three out of four benchmarks, including the most challenging scenarios, highlighting its potential as a robust solution for continual learning.
Authors: Han Chen, Zicong Jiang, Zining Zhang, Bingsheng He, Pingyi Luo, Mian Lu, Yuqiang Chen
Abstract: We introduce LogQuant, a groundbreaking 2-bit quantization technique for KV Cache in large language model (LLM) inference, delivering substantial memory savings while preserving superior performance. Previous methods either assume that later tokens are more important or attempt to predict important tokens based on earlier attention patterns. Both approaches, however, can result in performance bottlenecks or frequent mispredictions. LogQuant takes a different approach. By applying a log-based filtering mechanism, it selectively compresses the KV Cache across the entire context, achieving better performance with the same or even reduced memory footprint compared to existing methods. In benchmark tests, it enhances throughput by 25% and boosts batch size by 60% without increasing memory consumption. For challenging tasks such as Math and Code Completion, LogQuant improves accuracy by 40% to 200% at the same compression ratio, outperforming comparable techniques.LogQuant integrates effortlessly with popular inference frameworks like Python's transformers library. Implementation can be available in https://github.com/Concyclics/LogQuantKV.
Authors: Bohan Zhai, Canwen Xu, Yuxiong He, Zhewei Yao
Abstract: Text-to-SQL demands precise reasoning to convert natural language questions into structured queries. While large language models (LLMs) excel in many reasoning tasks, their ability to leverage Chain-of-Thought (CoT) reasoning for text-to-SQL remains underexplored. We identify critical limitations: zero-shot CoT offers minimal gains, and Direct Preference Optimization (DPO) applied without CoT yields marginal improvements. We propose ExCoT, a novel framework that iteratively optimizes open-source LLMs by combining CoT reasoning with off-policy and on-policy DPO, relying solely on execution accuracy as feedback. This approach eliminates the need for reward models or human-annotated preferences. Our experimental results demonstrate significant performance gains: ExCoT improves execution accuracy on BIRD dev set from 57.37% to 68.51% and on Spider test set from 78.81% to 86.59% for LLaMA-3 70B, with Qwen-2.5-Coder demonstrating similar improvements. Our best model achieves state-of-the-art performance in the single-model setting on both BIRD and Spider datasets, notably achieving 68.53% on the BIRD test set.
Authors: Jiuqi Wang, Rohan Chandra, Shangtong Zhang
Abstract: Loss of plasticity is one of the main challenges in continual learning with deep neural networks, where neural networks trained via backpropagation gradually lose their ability to adapt to new tasks and perform significantly worse than their freshly initialized counterparts. The main contribution of this paper is to propose a new hypothesis that experience replay addresses the loss of plasticity in continual learning. Here, experience replay is a form of memory. We provide supporting evidence for this hypothesis. In particular, we demonstrate in multiple different tasks, including regression, classification, and policy evaluation, that by simply adding an experience replay and processing the data in the experience replay with Transformers, the loss of plasticity disappears. Notably, we do not alter any standard components of deep learning. For example, we do not change backpropagation. We do not modify the activation functions. And we do not use any regularization. We conjecture that experience replay and Transformers can address the loss of plasticity because of the in-context learning phenomenon.
Authors: Gabriel Bo, Justin Gu, Christopher Sun
Abstract: We present a foundation modeling framework that leverages deep learning to uncover latent genetic signatures across the hematopoietic hierarchy. Our approach trains a fully connected autoencoder on multipotent progenitor cells, reducing over 20,000 gene features to a 256-dimensional latent space that captures predictive information for both progenitor and downstream differentiated cells such as monocytes and lymphocytes. We validate the quality of these embeddings by training feed-forward, transformer, and graph convolutional architectures for blood disease diagnosis tasks. We also explore zero-shot prediction using a progenitor disease state classification model to classify downstream cell conditions. Our models achieve greater than 95% accuracy for multi-class classification, and in the zero-shot setting, we achieve greater than 0.7 F1-score on the binary classification task. Future work should improve embeddings further to increase robustness on lymphocyte classification specifically.
Authors: Volkan Ustun, Soham Hans, Rajay Kumar, Yunzhe Wang
Abstract: Multi-agent reinforcement learning (MARL) is increasingly ubiquitous in training dynamic and adaptive synthetic characters for interactive simulations on geo-specific terrains. Frameworks such as Unity's ML-Agents help to make such reinforcement learning experiments more accessible to the simulation community. Military training simulations also benefit from advances in MARL, but they have immense computational requirements due to their complex, continuous, stochastic, partially observable, non-stationary, and doctrine-based nature. Furthermore, these simulations require geo-specific terrains, further exacerbating the computational resources problem. In our research, we leverage Unity's waypoints to automatically generate multi-layered representation abstractions of the geo-specific terrains to scale up reinforcement learning while still allowing the transfer of learned policies between different representations. Our early exploratory results on a novel MARL scenario, where each side has differing objectives, indicate that waypoint-based navigation enables faster and more efficient learning while producing trajectories similar to those taken by expert human players in CSGO gaming environments. This research points out the potential of waypoint-based navigation for reducing the computational costs of developing and training MARL models for military training simulations, where geo-specific terrains and differing objectives are crucial.
Authors: Dmitry B. Rokhlin, Olga V. Gurtovaya
Abstract: We introduce a novel multi-kernel learning algorithm, VAW$^2$, for online least squares regression in reproducing kernel Hilbert spaces (RKHS). VAW$^2$ leverages random Fourier feature-based functional approximation and the Vovk-Azoury-Warmuth (VAW) method in a two-level procedure: VAW is used to construct expert strategies from random features generated for each kernel at the first level, and then again to combine their predictions at the second level. A theoretical analysis yields a regret bound of $O(T^{1/2}\ln T)$ in expectation with respect to artificial randomness, when the number of random features scales as $T^{1/2}$. Empirical results on some benchmark datasets demonstrate that VAW$^2$ achieves superior performance compared to the existing online multi-kernel learning algorithms: Raker and OMKL-GF, and to other theoretically grounded method methods involving convex combination of expert predictions at the second level.
Authors: Somnath Basu Roy Chowdhury, Avinava Dubey, Ahmad Beirami, Rahul Kidambi, Nicholas Monath, Amr Ahmed, Snigdha Chaturvedi
Abstract: Concept erasure is the task of erasing information about a concept (e.g., gender or race) from a representation set while retaining the maximum possible utility -- information from original representations. Concept erasure is useful in several applications, such as removing sensitive concepts to achieve fairness and interpreting the impact of specific concepts on a model's performance. Previous concept erasure techniques have prioritized robustly erasing concepts over retaining the utility of the resultant representations. However, there seems to be an inherent tradeoff between erasure and retaining utility, making it unclear how to achieve perfect concept erasure while maintaining high utility. In this paper, we offer a fresh perspective toward solving this problem by quantifying the fundamental limits of concept erasure through an information-theoretic lens. Using these results, we investigate constraints on the data distribution and the erasure functions required to achieve the limits of perfect concept erasure. Empirically, we show that the derived erasure functions achieve the optimal theoretical bounds. Additionally, we show that our approach outperforms existing methods on a range of synthetic and real-world datasets using GPT-4 representations.
Authors: Chang Chen, Hany Hamed, Doojin Baek, Taegu Kang, Yoshua Bengio, Sungjin Ahn
Abstract: This paper tackles a novel problem, extendable long-horizon planning-enabling agents to plan trajectories longer than those in training data without compounding errors. To tackle this, we propose the Hierarchical Multiscale Diffuser (HM-Diffuser) and Progressive Trajectory Extension (PTE), an augmentation method that iteratively generates longer trajectories by stitching shorter ones. HM-Diffuser trains on these extended trajectories using a hierarchical structure, efficiently handling tasks across multiple temporal scales. Additionally, we introduce Adaptive Plan Pondering and the Recursive HM-Diffuser, which consolidate hierarchical layers into a single model to process temporal scales recursively. Experimental results demonstrate the effectiveness of our approach, advancing diffusion-based planners for scalable long-horizon planning.
Authors: Xiaobo Ma, Hyunsoo Noh, Ryan Hatch, James Tokishi, Zepu Wang
Abstract: Urban transportation networks are vital for the efficient movement of people and goods, necessitating effective traffic management and planning. An integral part of traffic management is understanding the turning movement counts (TMCs) at intersections, Accurate TMCs at intersections are crucial for traffic signal control, congestion mitigation, and road safety. In general, TMCs are obtained using physical sensors installed at intersections, but this approach can be cost-prohibitive and technically challenging, especially for cities with extensive road networks. Recent advancements in machine learning and data-driven approaches have offered promising alternatives for estimating TMCs. Traffic patterns can vary significantly across different intersections due to factors such as road geometry, traffic signal settings, and local driver behaviors. This domain discrepancy limits the generalizability and accuracy of machine learning models when applied to new or unseen intersections. In response to these limitations, this research proposes a novel framework leveraging domain adaptation (DA) to estimate TMCs at intersections by using traffic controller event-based data, road infrastructure data, and point-of-interest (POI) data. Evaluated on 30 intersections in Tucson, Arizona, the performance of the proposed DA framework was compared with state-of-the-art models and achieved the lowest values in terms of Mean Absolute Error and Root Mean Square Error.
Authors: Bicheng Ying, Zhe Li, Haibo Yang
Abstract: This work introduces a novel decentralized framework to interpret federated learning (FL) and, consequently, correct the biases introduced by arbitrary client participation and data heterogeneity, which are two typical traits in practical FL. Specifically, we first reformulate the core processes of FedAvg - client participation, local updating, and model aggregation - as stochastic matrix multiplications. This reformulation allows us to interpret FedAvg as a decentralized algorithm. Leveraging the decentralized optimization framework, we are able to provide a concise analysis to quantify the impact of arbitrary client participation and data heterogeneity on FedAvg's convergence point. This insight motivates the development of Federated Optimization with Exact Convergence via Push-pull Strategy (FOCUS), a novel algorithm inspired by the decentralized algorithm that eliminates these biases and achieves exact convergence without requiring the bounded heterogeneity assumption. Furthermore, we theoretically prove that FOCUS exhibits linear convergence (exponential decay) for both strongly convex and non-convex functions satisfying the Polyak-Lojasiewicz condition, regardless of the arbitrary nature of client participation.
Authors: Zhenkai Qin, Weibao Zhong, Caifeng Gao
Abstract: With the acceleration of urbanization, the spatiotemporal characteristics of criminal activities have become increasingly complex. Accurate prediction of crime distribution is crucial for optimizing the allocation of police resources and preventing crime. This paper proposes LGSTime, a crime spatiotemporal prediction model that integrates Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and the Multi-head Sparse Self-attention mechanism. LSTM and GRU capture long-term dependencies in crime time series, such as seasonality and periodicity, through their unique gating mechanisms. The Multi-head Sparse Self-attention mechanism, on the other hand, focuses on both temporal and spatial features of criminal events simultaneously through parallel processing and sparsification techniques, significantly improving computational efficiency and prediction accuracy. The integrated model leverages the strengths of each technique to better handle complex spatiotemporal data. Experimental findings demonstrate that the model attains optimal performance across four real - world crime datasets. In comparison to the CNN model, it exhibits performance enhancements of 2.8\%, 1.9\%, and 1.4\% in the Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) metrics respectively. These results offer a valuable reference for tackling the challenges in crime prediction.
Authors: Junyi Zhu, Ruicong Yao, Taha Ceritli, Savas Ozkan, Matthew B. Blaschko, Eunchung Noh, Jeongwon Min, Cho Jung Min, Mete Ozay
Abstract: Artificial Intelligence (AI) technologies have revolutionized numerous fields, yet their applications often rely on costly and time-consuming data collection processes. Federated Learning (FL) offers a promising alternative by enabling AI models to be trained on decentralized data where data is scattered across clients (distributed nodes). However, existing FL approaches struggle to match the performance of centralized training due to challenges such as heterogeneous data distribution and communication delays, limiting their potential for breakthroughs. We observe that many real-world use cases involve hybrid data regimes, in which a server (center node) has access to some data while a large amount of data is distributed across associated clients. To improve the utilization of decentralized data under this regime, address data heterogeneity issue, and facilitate asynchronous communication between the server and clients, we propose a dual learning approach that leverages centralized data at the server to guide the merging of model updates from clients. Our method accommodates scenarios where server data is out-of-domain relative to decentralized client data, making it applicable to a wide range of use cases. We provide theoretical analysis demonstrating the faster convergence of our method compared to existing methods. Furthermore, experimental results across various scenarios show that our approach significantly outperforms existing technologies, highlighting its potential to unlock the value of large amounts of decentralized data.
Authors: Yongshuai Liu, Xin Liu
Abstract: Model-based reinforcement learning (MBRL) has demonstrated superior sample efficiency compared to model-free reinforcement learning (MFRL). However, the presence of inaccurate models can introduce biases during policy learning, resulting in misleading trajectories. The challenge lies in obtaining accurate models due to limited diverse training data, particularly in regions with limited visits (uncertain regions). Existing approaches passively quantify uncertainty after sample generation, failing to actively collect uncertain samples that could enhance state coverage and improve model accuracy. Moreover, MBRL often faces difficulties in making accurate multi-step predictions, thereby impacting overall performance. To address these limitations, we propose a novel framework for uncertainty-aware policy optimization with model-based exploratory planning. In the model-based planning phase, we introduce an uncertainty-aware k-step lookahead planning approach to guide action selection at each step. This process involves a trade-off analysis between model uncertainty and value function approximation error, effectively enhancing policy performance. In the policy optimization phase, we leverage an uncertainty-driven exploratory policy to actively collect diverse training samples, resulting in improved model accuracy and overall performance of the RL agent. Our approach offers flexibility and applicability to tasks with varying state/action spaces and reward structures. We validate its effectiveness through experiments on challenging robotic manipulation tasks and Atari games, surpassing state-of-the-art methods with fewer interactions, thereby leading to significant performance improvements.
Authors: Seyedeh Azadeh Fallah Mortezanejad (International Science Consulting and Training), Ruochen Wang (International Science Consulting and Training), Ali Mohammad-Djafari (International Science Consulting and Training)
Abstract: A significant advancement in Neural Network (NN) research is the integration of domain-specific knowledge through custom loss functions. This approach addresses a crucial challenge: how can models utilize physics or mathematical principles to enhance predictions when dealing with sparse, noisy, or incomplete data? Physics-Informed Neural Networks (PINNs) put this idea into practice by incorporating physical equations, such as Partial Differential Equations (PDEs), as soft constraints. This guidance helps the networks find solutions that align with established laws. Recently, researchers have expanded this framework to include Bayesian NNs (BNNs), which allow for uncertainty quantification while still adhering to physical principles. But what happens when the governing equations of a system are not known? In this work, we introduce methods to automatically extract PDEs from historical data. We then integrate these learned equations into three different modeling approaches: PINNs, Bayesian-PINNs (B-PINNs), and Bayesian Linear Regression (BLR). To assess these frameworks, we evaluate them on a real-world Multivariate Time Series (MTS) dataset. We compare their effectiveness in forecasting future states under different scenarios: with and without PDE constraints and accuracy considerations. This research aims to bridge the gap between data-driven discovery and physics-guided learning, providing valuable insights for practical applications.
Authors: Seyedeh Azadeh Fallah Mortezanejad (School of Automotive,Traffic Engineering, Jiangsu University, Zhenjiang, Jiangsu, China), Ruochen Wang (School of Automotive,Traffic Engineering, Jiangsu University, Zhenjiang, Jiangsu, China)
Abstract: The explosion of Time Series (TS) data, driven by advancements in technology, necessitates sophisticated analytical methods. Modern management systems increasingly rely on analyzing this data, highlighting the importance of effcient processing techniques. State-of-the-art Machine Learning (ML) approaches for TS analysis and forecasting are becoming prevalent. This paper briefly describes and compiles suitable algorithms for TS regression task. We compare these algorithms against each other and the classic ARIMA method using diverse datasets: complete data, data with outliers, and data with missing values. The focus is on forecasting accuracy, particularly for long-term predictions. This research aids in selecting the most appropriate algorithm based on forecasting needs and data characteristics.
Authors: Bahareh Golchin, Noushin Riahi
Abstract: One of the most significant issues as attended a lot in recent years is that of recognizing the sentiments and emotions in social media texts. The analysis of sentiments and emotions is intended to recognize the conceptual information such as the opinions, feelings, attitudes and emotions of people towards the products, services, organizations, people, topics, events and features in the written text. These indicate the greatness of the problem space. In the real world, businesses and organizations are always looking for tools to gather ideas, emotions, and directions of people about their products, services, or events related to their own. This article uses the Twitter social network, one of the most popular social networks with about 420 million active users, to extract data. Using this social network, users can share their information and opinions about personal issues, policies, products, events, etc. It can be used with appropriate classification of emotional states due to the availability of its data. In this study, supervised learning and deep neural network algorithms are used to classify the emotional states of Twitter users. The use of deep learning methods to increase the learning capacity of the model is an advantage due to the large amount of available data. Tweets collected on various topics are classified into four classes using a combination of two Bidirectional Long Short Term Memory network and a Convolutional network. The results obtained from this study with an average accuracy of 93%, show good results extracted from the proposed framework and improved accuracy compared to previous work.
Authors: Xianke Qiang, Zheng Chang, Ying-Chang Liang
Abstract: Federated learning (FL) can fully leverage large-scale terminal data while ensuring privacy and security, and is considered as a distributed alternative for the centralized machine learning. However, the issue of data heterogeneity poses limitations on FL's performance. To address this challenge, artificial intelligence-generated content (AIGC) which is an innovative data synthesis technique emerges as one potential solution. In this article, we first provide an overview of the system architecture, performance metrics, and challenges associated with AIGC-assistant FL system design. We then propose the Generative federated learning (GenFL) architecture and present its workflow, including the design of aggregation and weight policy. Finally, using the CIFAR10 and CIFAR100 datasets, we employ diffusion models to generate dataset and improve FL performance. Experiments conducted under various non-independent and identically distributed (non-IID) data distributions demonstrate the effectiveness of GenFL on overcoming the bottlenecks in FL caused by data heterogeneity. Open research directions in the research of AIGC-assisted FL are also discussed.
Authors: RuiXi Qiao, Jie Cheng, Xingyuan Dai, Yonglin Tian, Yisheng Lv
Abstract: Skills have been introduced to offline reinforcement learning (RL) as temporal abstractions to tackle complex, long-horizon tasks, promoting consistent behavior and enabling meaningful exploration. While skills in offline RL are predominantly modeled within a continuous latent space, the potential of discrete skill spaces remains largely underexplored. In this paper, we propose a compact discrete skill space for offline RL tasks supported by state-of-the-art transformer-based encoder and diffusion-based decoder. Coupled with a high-level policy trained via offline RL techniques, our method establishes a hierarchical RL framework where the trained diffusion decoder plays a pivotal role. Empirical evaluations show that the proposed algorithm, Discrete Diffusion Skill (DDS), is a powerful offline RL method. DDS performs competitively on Locomotion and Kitchen tasks and excels on long-horizon tasks, achieving at least a 12 percent improvement on AntMaze-v2 benchmarks compared to existing offline RL approaches. Furthermore, DDS offers improved interpretability, training stability, and online exploration compared to previous skill-based methods.
Authors: Pirzada Suhail, Amit Sethi
Abstract: In machine learning, especially with vision classifiers, generating inputs that are confidently classified by the model is essential for understanding its decision boundaries and behavior. However, creating such samples that are confidently classified yet distinct from the training data distribution is a challenge. Traditional methods often modify existing inputs, but they don't always ensure confident classification. In this work, we extend network inversion techniques to generate Confidently Classified Counterfeits-synthetic samples that are confidently classified by the model despite being significantly different from the training data. We achieve this by modifying the generator's conditioning mechanism from soft vector conditioning to one-hot vector conditioning and applying Kullback-Leibler divergence (KLD) between the one-hot vectors and the classifier's output distribution. This encourages the generator to produce samples that are both plausible and confidently classified. Generating Confidently Classified Counterfeits is crucial for ensuring the safety and reliability of machine learning systems, particularly in safety-critical applications where models must exhibit confidence only on data within the training distribution. By generating such counterfeits, we challenge the assumption that high-confidence predictions are always indicative of in-distribution data, providing deeper insights into the model's limitations and decision-making process.
Authors: Srihas Yarlagadda, Amey Agrawal, Elton Pinto, Hakesh Darapaneni, Mitali Meratwal, Shivam Mittal, Pranavi Bajjuri, Srinivas Sridharan, Alexey Tumanov
Abstract: Training large foundation models costs hundreds of millions of dollars, making deployment optimization critical. Current approaches require machine learning engineers to manually craft training recipes through error-prone trial-and-error on expensive compute clusters. To enable efficient exploration of training configurations, researchers have developed performance modeling systems. However, these systems force users to translate their workloads into custom specification languages, introducing a fundamental semantic gap between the actual workload and its representation. This gap creates an inherent tradeoff: systems must either support a narrow set of workloads to maintain usability, require complex specifications that limit practical adoption, or compromise prediction accuracy with simplified models. We present Maya, a performance modeling system that eliminates these tradeoffs through transparent device emulation. By operating at the narrow interface between training frameworks and accelerator devices, Maya can capture complete workload behavior without requiring code modifications or translations. Maya intercepts device API calls from unmodified training code to directly observe low-level operations, enabling accurate performance prediction while maintaining both ease of use and generality. Our evaluation shows Maya achieves less than 5% prediction error across diverse models and optimization strategies, identifying configurations that reduce training costs by up to 56% compared to existing approaches.
Authors: Salaheddin Alzubi, Creston Brooks, Purva Chiniya, Edoardo Contente, Chiara von Gerlach, Lucas Irwin, Yihan Jiang, Arda Kaz, Windsor Nguyen, Sewoong Oh, Himanshu Tyagi, Pramod Viswanath
Abstract: We introduce Open Deep Search (ODS) to close the increasing gap between the proprietary search AI solutions, such as Perplexity's Sonar Reasoning Pro and OpenAI's GPT-4o Search Preview, and their open-source counterparts. The main innovation introduced in ODS is to augment the reasoning capabilities of the latest open-source LLMs with reasoning agents that can judiciously use web search tools to answer queries. Concretely, ODS consists of two components that work with a base LLM chosen by the user: Open Search Tool and Open Reasoning Agent. Open Reasoning Agent interprets the given task and completes it by orchestrating a sequence of actions that includes calling tools, one of which is the Open Search Tool. Open Search Tool is a novel web search tool that outperforms proprietary counterparts. Together with powerful open-source reasoning LLMs, such as DeepSeek-R1, ODS nearly matches and sometimes surpasses the existing state-of-the-art baselines on two benchmarks: SimpleQA and FRAMES. For example, on the FRAMES evaluation benchmark, ODS improves the best existing baseline of the recently released GPT-4o Search Preview by 9.7% in accuracy. ODS is a general framework for seamlessly augmenting any LLMs -- for example, DeepSeek-R1 that achieves 82.4% on SimpleQA and 30.1% on FRAMES -- with search and reasoning capabilities to achieve state-of-the-art performance: 88.3% on SimpleQA and 75.3% on FRAMES.
Authors: Xiao-Cheng Liao, Yi Mei, Mengjie Zhang, Xiang-Ling Chen
Abstract: Appropriate traffic state representation is crucial for learning traffic signal control policies. However, most of the current traffic state representations are heuristically designed, with insufficient theoretical support. In this paper, we (1) develop a flexible, efficient, and theoretically grounded method, namely generalized phase pressure (G2P) control, which takes only simple lane features into consideration to decide which phase to be actuated; 2) extend the pressure control theory to a general form for multi-homogeneous-lane road networks based on queueing theory; (3) design a new traffic state representation based on the generalized phase state features from G2P control; and 4) develop a reinforcement learning (RL)-based algorithm template named G2P-XLight, and two RL algorithms, G2P-MPLight and G2P-CoLight, by combining the generalized phase state representation with MPLight and CoLight, two well-performed RL methods for learning traffic signal control policies. Extensive experiments conducted on multiple real-world datasets demonstrate that G2P control outperforms the state-of-the-art (SOTA) heuristic method in the transportation field and other recent human-designed heuristic methods; and that the newly proposed G2P-XLight significantly outperforms SOTA learning-based approaches. Our code is available online.
Authors: D. Veerababu, Prasanta K. Ghosh
Abstract: Physics-informed neural networks offered an alternate way to solve several differential equations that govern complicated physics. However, their success in predicting the acoustic field is limited by the vanishing-gradient problem that occurs when solving the Helmholtz equation. In this paper, a formulation is presented that addresses this difficulty. The problem of solving the two-dimensional Helmholtz equation with the prescribed boundary conditions is posed as an unconstrained optimization problem using trial solution method. According to this method, a trial neural network that satisfies the given boundary conditions prior to the training process is constructed using the technique of transfinite interpolation and the theory of R-functions. This ansatz is initially applied to the rectangular domain and later extended to the circular and elliptical domains. The acoustic field predicted from the proposed formulation is compared with that obtained from the two-dimensional finite element methods. Good agreement is observed in all three domains considered. Minor limitations associated with the proposed formulation and their remedies are also discussed.
Authors: Xiao Lin, Manoj Acharya, Anirban Roy, Susmit Jha
Abstract: Mitigating Trojans in Large Language Models (LLMs) is one of many tasks where alignment data is LLM specific, as different LLMs have different Trojan triggers and trigger behaviors to be removed. In this paper, we introduce TeleLoRA (Teleporting Low-Rank Adaptation), a novel framework that synergizes model-specific alignment data across multiple LLMs to enable zero-shot Trojan mitigation on unseen LLMs without alignment data. TeleLoRA learns a unified generator of LoRA adapter weights by leveraging local activation information across multiple LLMs. This generator is designed to be permutation symmetric to generalize across models with different architectures and sizes. We optimize the model design for memory efficiency, making it feasible to learn with large-scale LLMs with minimal computational resources. Experiments on LLM Trojan mitigation benchmarks demonstrate that TeleLoRA effectively reduces attack success rates while preserving the benign performance of the models.
Authors: Yunrui Zhang, Gustavo Batista, Salil S. Kanhere
Abstract: Time series classification is usually regarded as a distinct task from tabular data classification due to the importance of temporal information. However, in this paper, by performing permutation tests that disrupt temporal information on the UCR time series classification archive, the most widely used benchmark for time series classification, we identify a significant proportion of datasets where temporal information has little to no impact on classification. Many of these datasets are tabular in nature or rely mainly on tabular features, leading to potentially biased evaluations of time series classifiers focused on temporal information. To address this, we propose UCR Augmented, a benchmark based on the UCR time series classification archive designed to evaluate classifiers' ability to extract and utilize temporal information. Testing classifiers from seven categories on this benchmark revealed notable shifts in performance rankings. Some previously overlooked approaches perform well, while others see their performance decline significantly when temporal information is crucial. UCR Augmented provides a more robust framework for assessing time series classifiers, ensuring fairer evaluations. Our code is available at https://github.com/YunruiZhang/Revisit-Time-Series-Classification-Benchmark.
URLs: https://github.com/YunruiZhang/Revisit-Time-Series-Classification-Benchmark.
Authors: Hongye Cao, Fan Feng, Jing Huo, Shangdong Yang, Meng Fang, Tianpei Yang, Yang Gao
Abstract: Model-based offline Reinforcement Learning (RL) constructs environment models from offline datasets to perform conservative policy optimization. Existing approaches focus on learning state transitions through ensemble models, rollouting conservative estimation to mitigate extrapolation errors. However, the static data makes it challenging to develop a robust policy, and offline agents cannot access the environment to gather new data. To address these challenges, we introduce Model-based Offline Reinforcement learning with AdversariaL data augmentation (MORAL). In MORAL, we replace the fixed horizon rollout by employing adversaria data augmentation to execute alternating sampling with ensemble models to enrich training data. Specifically, this adversarial process dynamically selects ensemble models against policy for biased sampling, mitigating the optimistic estimation of fixed models, thus robustly expanding the training data for policy optimization. Moreover, a differential factor is integrated into the adversarial process for regularization, ensuring error minimization in extrapolations. This data-augmented optimization adapts to diverse offline tasks without rollout horizon tuning, showing remarkable applicability. Extensive experiments on D4RL benchmark demonstrate that MORAL outperforms other model-based offline RL methods in terms of policy learning and sample efficiency.
Authors: Francesco Micheli, Efe C. Balta, Anastasios Tsiamis, John Lygeros
Abstract: We address the challenge of sequential data-driven decision-making under context distributional uncertainty. This problem arises in numerous real-world scenarios where the learner optimizes black-box objective functions in the presence of uncontrollable contextual variables. We consider the setting where the context distribution is uncertain but known to lie within an ambiguity set defined as a ball in the Wasserstein distance. We propose a novel algorithm for Wasserstein Distributionally Robust Bayesian Optimization that can handle continuous context distributions while maintaining computational tractability. Our theoretical analysis combines recent results in self-normalized concentration in Hilbert spaces and finite-sample bounds for distributionally robust optimization to establish sublinear regret bounds that match state-of-the-art results. Through extensive comparisons with existing approaches on both synthetic and real-world problems, we demonstrate the simplicity, effectiveness, and practical applicability of our proposed method.
Authors: Yulu Han, Ziye Jia, Sijie He, Yu Zhang, Qihui Wu
Abstract: The unmanned aerial vehicle (UAV) network has gained significant attentions in recent years due to its various applications. However, the traffic security becomes the key threatening public safety issue in an emergency rescue system due to the increasing vulnerability of UAVs to cyber attacks in environments with high heterogeneities. Hence, in this paper, we propose a novel anomaly traffic detection architecture for UAV networks based on the software-defined networking (SDN) framework and blockchain technology. Specifically, SDN separates the control and data plane to enhance the network manageability and security. Meanwhile, the blockchain provides decentralized identity authentication and data security records. Beisdes, a complete security architecture requires an effective mechanism to detect the time-series based abnormal traffic. Thus, an integrated algorithm combining convolutional neural networks (CNNs) and Transformer (CNN+Transformer) for anomaly traffic detection is developed, which is called CTranATD. Finally, the simulation results show that the proposed CTranATD algorithm is effective and outperforms the individual CNN, Transformer, and LSTM algorithms for detecting anomaly traffic.
Authors: Tianqi He, Xiaohan Huang, Yi Du, Qingqing Long, Ziyue Qiao, Min Wu, Yanjie Fu, Yuanchun Zhou, Meng Xiao
Abstract: Feature Transformation is crucial for classic machine learning that aims to generate feature combinations to enhance the performance of downstream tasks from a data-centric perspective. Current methodologies, such as manual expert-driven processes, iterative-feedback techniques, and exploration-generative tactics, have shown promise in automating such data engineering workflow by minimizing human involvement. However, three challenges remain in those frameworks: (1) It predominantly depends on downstream task performance metrics, as assessment is time-consuming, especially for large datasets. (2) The diversity of feature combinations will hardly be guaranteed after random exploration ends. (3) Rare significant transformations lead to sparse valuable feedback that hinders the learning processes or leads to less effective results. In response to these challenges, we introduce FastFT, an innovative framework that leverages a trio of advanced strategies.We first decouple the feature transformation evaluation from the outcomes of the generated datasets via the performance predictor. To address the issue of reward sparsity, we developed a method to evaluate the novelty of generated transformation sequences. Incorporating this novelty into the reward function accelerates the model's exploration of effective transformations, thereby improving the search productivity. Additionally, we combine novelty and performance to create a prioritized memory buffer, ensuring that essential experiences are effectively revisited during exploration. Our extensive experimental evaluations validate the performance, efficiency, and traceability of our proposed framework, showcasing its superiority in handling complex feature transformation tasks.
Authors: Mahbod Nouri, David Rotermund, Alberto Garcia-Ortiz, Klaus R. Pawelzik
Abstract: The brain uses positive signals as a means of signaling. Forward interactions in the early visual cortex are also positive, realized by excitatory synapses. Only local interactions also include inhibition. Non-negative matrix factorization (NMF) captures the biological constraint of positive long-range interactions and can be implemented with stochastic spikes. While NMF can serve as an abstract formalization of early neural processing in the visual system, the performance of deep convolutional networks with NMF modules does not match that of CNNs of similar size. However, when the local NMF modules are each followed by a module that mixes the NMF's positive activities, the performances on the benchmark data exceed that of vanilla deep convolutional networks of similar size. This setting can be considered a biologically more plausible emulation of the processing in cortical (hyper-)columns with the potential to improve the performance of deep networks.
Authors: Rita T. Sousa, Heiko Paulheim
Abstract: Gene expression datasets offer insights into gene regulation mechanisms, biochemical pathways, and cellular functions. Additionally, comparing gene expression profiles between disease and control patients can deepen the understanding of disease pathology. Therefore, machine learning has been used to process gene expression data, with patient diagnosis emerging as one of the most popular applications. Although gene expression data can provide valuable insights, challenges arise because the number of patients in expression datasets is usually limited, and the data from different datasets with different gene expressions cannot be easily combined. This work proposes a novel methodology to address these challenges by integrating multiple gene expression datasets and domain-specific knowledge using knowledge graphs, a unique tool for biomedical data integration. Then, vector representations are produced using knowledge graph embedding techniques, which are used as inputs for a graph neural network and a multi-layer perceptron. We evaluate the efficacy of our methodology in three settings: single-dataset learning, multi-dataset learning, and transfer learning. The experimental results show that combining gene expression datasets and domain-specific knowledge improves patient diagnosis in all three settings.
Authors: Antonio Maratea, Rita Perna
Abstract: Adequate sampling space coverage is the keystone to effectively train trustworthy Machine Learning models. Unfortunately, real data do carry several inherent risks due to the many potential biases they exhibit when gathered without a proper random sampling over the reference population, and most of the times this is way too expensive or time consuming to be a viable option. Depending on how training data have been gathered, unmitigated biases can lead to harmful or discriminatory consequences that ultimately hinders large scale applicability of pre-trained models and undermine their truthfulness or fairness expectations. In this paper, a mixed active sampling and data generation strategy -- called samplation -- is proposed as a mean to compensate during fine-tuning of a pre-trained classifer the unfair classifications it produces, assuming that the training data come from a non-probabilistic sampling schema. Given a pre-trained classifier, first a fairness metric is evaluated on a test set, then new reservoirs of labeled data are generated and finally a number of reversely-biased artificial samples are generated for the fine-tuning of the model. Using as case study Deep Models for visual semantic role labeling, the proposed method has been able to fully cure a simulated gender bias starting from a 90/10 imbalance, with only a small percentage of new data and with a minor effect on accuracy.
Authors: Yangqi Feng, Shing-Ho J. Lin, Baoyuan Gao, Xian Wei
Abstract: Recent research has revealed that high compression of Deep Neural Networks (DNNs), e.g., massive pruning of the weight matrix of a DNN, leads to a severe drop in accuracy and susceptibility to adversarial attacks. Integration of network pruning into an adversarial training framework has been proposed to promote adversarial robustness. It has been observed that a highly pruned weight matrix tends to be ill-conditioned, i.e., increasing the condition number of the weight matrix. This phenomenon aggravates the vulnerability of a DNN to input noise. Although a highly pruned weight matrix is considered to be able to lower the upper bound of the local Lipschitz constant to tolerate large distortion, the ill-conditionedness of such a weight matrix results in a non-robust DNN model. To overcome this challenge, this work develops novel joint constraints to adjust the weight distribution of networks, namely, the Transformed Sparse Constraint joint with Condition Number Constraint (TSCNC), which copes with smoothing distribution and differentiable constraint functions to reduce condition number and thus avoid the ill-conditionedness of weight matrices. Furthermore, our theoretical analyses unveil the relevance between the condition number and the local Lipschitz constant of the weight matrix, namely, the sharply increasing condition number becomes the dominant factor that restricts the robustness of over-sparsified models. Extensive experiments are conducted on several public datasets, and the results show that the proposed constraints significantly improve the robustness of a DNN with high pruning rates.
Authors: Jinghui Yuan, Fangyuan Xie, Feiping Nie, Xuelong Li
Abstract: The indicator matrix plays an important role in machine learning, but optimizing it is an NP-hard problem. We propose a new relaxation of the indicator matrix and prove that this relaxation forms a manifold, which we call the Relaxed Indicator Matrix Manifold (RIM manifold). Based on Riemannian geometry, we develop a Riemannian toolbox for optimization on the RIM manifold. Specifically, we provide several methods of Retraction, including a fast Retraction method to obtain geodesics. We point out that the RIM manifold is a generalization of the double stochastic manifold, and it is much faster than existing methods on the double stochastic manifold, which has a complexity of \( \mathcal{O}(n^3) \), while RIM manifold optimization is \( \mathcal{O}(n) \) and often yields better results. We conducted extensive experiments, including image denoising, with millions of variables to support our conclusion, and applied the RIM manifold to Ratio Cut, achieving clustering results that outperform the state-of-the-art methods. Our Code in \href{https://github.com/Yuan-Jinghui/Riemannian-Optimization-on-Relaxed-Indicator-Matrix-Manifold}{https://github.com/Yuan-Jinghui/Riemannian-Optimization-on-Relaxed-Indicator-Matrix-Manifold}.
URLs: https://github.com/Yuan-Jinghui/Riemannian-Optimization-on-Relaxed-Indicator-Matrix-Manifold, https://github.com/Yuan-Jinghui/Riemannian-Optimization-on-Relaxed-Indicator-Matrix-Manifold
Authors: Ryumei Nakada, Wenlong Ji, Tianxi Cai, James Zou, Linjun Zhang
Abstract: Prompt engineering has emerged as a powerful technique for guiding large language models (LLMs) toward desired responses, significantly enhancing their performance across diverse tasks. Beyond their role as static predictors, LLMs increasingly function as intelligent agents, capable of reasoning, decision-making, and adapting dynamically to complex environments. However, the theoretical underpinnings of prompt engineering remain largely unexplored. In this paper, we introduce a formal framework demonstrating that transformer models, when provided with carefully designed prompts, can act as a configurable computational system by emulating a ``virtual'' neural network during inference. Specifically, input prompts effectively translate into the corresponding network configuration, enabling LLMs to adjust their internal computations dynamically. Building on this construction, we establish an approximation theory for $\beta$-times differentiable functions, proving that transformers can approximate such functions with arbitrary precision when guided by appropriately structured prompts. Moreover, our framework provides theoretical justification for several empirically successful prompt engineering techniques, including the use of longer, structured prompts, filtering irrelevant information, enhancing prompt token diversity, and leveraging multi-agent interactions. By framing LLMs as adaptable agents rather than static models, our findings underscore their potential for autonomous reasoning and problem-solving, paving the way for more robust and theoretically grounded advancements in prompt engineering and AI agent design.
Authors: Ran Wang, Xinlei Zhou, Rihao Li, Meng Hu, Wenhui Wu, Yuheng Jia
Abstract: Despite the remarkable success of deep neural networks (DNNs), the security threat of adversarial attacks poses a significant challenge to the reliability of DNNs. By introducing randomness into different parts of DNNs, stochastic methods can enable the model to learn some uncertainty, thereby improving model robustness efficiently. In this paper, we theoretically discover a universal phenomenon that adversarial attacks will shift the distributions of feature statistics. Motivated by this theoretical finding, we propose a robustness enhancement module called Feature Statistics with Uncertainty (FSU). It resamples channel-wise feature means and standard deviations of examples from multivariate Gaussian distributions, which helps to reconstruct the attacked examples and calibrate the shifted distributions. The calibration recovers some domain characteristics of the data for classification, thereby mitigating the influence of perturbations and weakening the ability of attacks to deceive models. The proposed FSU module has universal applicability in training, attacking, predicting and fine-tuning, demonstrating impressive robustness enhancement ability at trivial additional time cost. For example, against powerful optimization-based CW attacks, by incorporating FSU into attacking and predicting phases, it endows many collapsed state-of-the-art models with 50%-80% robust accuracy on CIFAR10, CIFAR100 and SVHN.
Authors: Trung Duc Ha, Sidney Bender
Abstract: Counterfactual explanations have been successfully applied to create human interpretable explanations for various black-box models. They are handy for tasks in the image domain, where the quality of the explanations benefits from recent advances in generative models. Although counterfactual explanations have been widely applied to classification models, their application to regression tasks remains underexplored. We present two methods to create counterfactual explanations for image regression tasks using diffusion-based generative models to address challenges in sparsity and quality: 1) one based on a Denoising Diffusion Probabilistic Model that operates directly in pixel-space and 2) another based on a Diffusion Autoencoder operating in latent space. Both produce realistic, semantic, and smooth counterfactuals on CelebA-HQ and a synthetic data set, providing easily interpretable insights into the decision-making process of the regression model and reveal spurious correlations. We find that for regression counterfactuals, changes in features depend on the region of the predicted value. Large semantic changes are needed for significant changes in predicted values, making it harder to find sparse counterfactuals than with classifiers. Moreover, pixel space counterfactuals are more sparse while latent space counterfactuals are of higher quality and allow bigger semantic changes.
Authors: Zongyuan Zhang, Tianyang Duan, Zheng Lin, Dong Huang, Zihan Fang, Zekai Sun, Ling Xiong, Hongbin Liang, Heming Cui, Yong Cui
Abstract: Recently, deep reinforcement learning (DRL) has emerged as a promising approach for robotic control. However, the deployment of DRL in real-world robots is hindered by its sensitivity to environmental perturbations. While existing whitebox adversarial attacks rely on local gradient information and apply uniform perturbations across all states to evaluate DRL robustness, they fail to account for temporal dynamics and state-specific vulnerabilities. To combat the above challenge, we first conduct a theoretical analysis of white-box attacks in DRL by establishing the adversarial victim-dynamics Markov decision process (AVD-MDP), to derive the necessary and sufficient conditions for a successful attack. Based on this, we propose a selective state-aware reinforcement adversarial attack method, named STAR, to optimize perturbation stealthiness and state visitation dispersion. STAR first employs a soft mask-based state-targeting mechanism to minimize redundant perturbations, enhancing stealthiness and attack effectiveness. Then, it incorporates an information-theoretic optimization objective to maximize mutual information between perturbations, environmental states, and victim actions, ensuring a dispersed state-visitation distribution that steers the victim agent into vulnerable states for maximum return reduction. Extensive experiments demonstrate that STAR outperforms state-of-the-art benchmarks.
Authors: Davide Domini, Gianluca Aguzzi, Mirko Viroli
Abstract: In recent years, cro:flFederated learning (FL) has gained significant attention within the machine learning community. Although various FL algorithms have been proposed in the literature, their performance often degrades when data across clients is non-independently and identically distributed (non-IID). This skewness in data distribution often emerges from geographic patterns, with notable examples including regional linguistic variations in text data or localized traffic patterns in urban environments. Such scenarios result in IID data within specific regions but non-IID data across regions. However, existing FL algorithms are typically evaluated by randomly splitting non-IID data across devices, disregarding their spatial distribution. To address this gap, we introduce ProFed, a benchmark that simulates data splits with varying degrees of skewness across different regions. We incorporate several skewness methods from the literature and apply them to well-known datasets, including MNIST, FashionMNIST, CIFAR-10, and CIFAR-100. Our goal is to provide researchers with a standardized framework to evaluate FL algorithms more effectively and consistently against established baselines.
Authors: Haci Ismail Aslan, Philipp Wiesner, Ping Xiong, Odej Kao
Abstract: Graph Neural Networks (GNNs) are playing an increasingly important role in the efficient operation and security of computing systems, with applications in workload scheduling, anomaly detection, and resource management. However, their vulnerability to network perturbations poses a significant challenge. We propose $\beta$-GNN, a model enhancing GNN robustness without sacrificing clean data performance. $\beta$-GNN uses a weighted ensemble, combining any GNN with a multi-layer perceptron. A learned dynamic weight, $\beta$, modulates the GNN's contribution. This $\beta$ not only weights GNN influence but also indicates data perturbation levels, enabling proactive mitigation. Experimental results on diverse datasets show $\beta$-GNN's superior adversarial accuracy and attack severity quantification. Crucially, $\beta$-GNN avoids perturbation assumptions, preserving clean data structure and performance.
Authors: Sashuai Zhou, Hai Huang, Yan Xia
Abstract: Multi-modal models excel in cross-modal tasks but are computationally expensive due to their billions of parameters. Parameter-efficient fine-tuning (PEFT) offers a solution by adding small trainable components while freezing pre-trained parameters. However, existing methods primarily focus on uni-modal processing, overlooking the critical modal fusion needed for multi-modal tasks. To fill this gap, we propose heterogeneous mixture of experts adapters that extend the traditional PEFT framework to support multi-modal expert combinations and improve information interaction. Additionally, our approach modifies the affine linear expert design to enable efficient modal fusion in a low-rank space, achieving competitive performance with only 5-8\% of the parameters fine-tuned. Experiments across eight downstream tasks, including visual-audio and text-visual, demonstrate the superior performance of the approach.
Authors: Hozefa Jesawada, Antonio Acernese, Giovanni Russo, Carmen Del Vecchiob
Abstract: Ensuring robustness against epistemic, possibly adversarial, perturbations is essential for reliable real-world decision-making. While the Probabilistic Ensembles with Trajectory Sampling (PETS) algorithm inherently handles uncertainty via ensemble-based probabilistic models, it lacks guarantees against structured adversarial or worst-case uncertainty distributions. To address this, we propose DR-PETS, a distributionally robust extension of PETS that certifies robustness against adversarial perturbations. We formalize uncertainty via a p-Wasserstein ambiguity set, enabling worst-case-aware planning through a min-max optimization framework. While PETS passively accounts for stochasticity, DR-PETS actively optimizes robustness via a tractable convex approximation integrated into PETS planning loop. Experiments on pendulum stabilization and cart-pole balancing show that DR-PETS certifies robustness against adversarial parameter perturbations, achieving consistent performance in worst-case scenarios where PETS deteriorates.
Authors: Hansaka Aluvihare, Sivakumar Sivasankar, Xianqi Li, Arjuna Madanayake, Sirani M. Perera
Abstract: True-time-delay (TTD) beamformers can produce wideband, squint-free beams in both analog and digital signal domains, unlike frequency-dependent FFT beams. Our previous work showed that TTD beamformers can be efficiently realized using the elements of delay Vandermonde matrix (DVM), answering the longstanding beam-squint problem. Thus, building on our work on classical algorithms based on DVM, we propose neural network (NN) architecture to realize wideband multi-beam beamformers using structure-imposed weight matrices and submatrices. The structure and sparsity of the weight matrices and submatrices are shown to reduce the space and computational complexities of the NN greatly. The proposed network architecture has O(pLM logM) complexity compared to a conventional fully connected L-layers network with O(M2L) complexity, where M is the number of nodes in each layer of the network, p is the number of submatrices per layer, and M >> p. We will show numerical simulations in the 24 GHz to 32 GHz range to demonstrate the numerical feasibility of realizing wideband multi-beam beamformers using the proposed neural architecture. We also show the complexity reduction of the proposed NN and compare that with fully connected NNs, to show the efficiency of the proposed architecture without sacrificing accuracy. The accuracy of the proposed NN architecture was shown using the mean squared error, which is based on an objective function of the weight matrices and beamformed signals of antenna arrays, while also normalizing nodes. The proposed NN architecture shows a low-complexity NN realizing wideband multi-beam beamformers in real-time for low-complexity intelligent systems.
Authors: Yankai Chen, Taotao Wang, Yixiang Fang, Yunyu Xiao
Abstract: Node importance estimation, a classical problem in network analysis, underpins various web applications. Previous methods either exploit intrinsic topological characteristics, e.g., graph centrality, or leverage additional information, e.g., data heterogeneity, for node feature enhancement. However, these methods follow the supervised learning setting, overlooking the fact that ground-truth node-importance data are usually partially labeled in practice. In this work, we propose the first semi-supervised node importance estimation framework, i.e., EASING, to improve learning quality for unlabeled data in heterogeneous graphs. Different from previous approaches, EASING explicitly captures uncertainty to reflect the confidence of model predictions. To jointly estimate the importance values and uncertainties, EASING incorporates DJE, a deep encoder-decoder neural architecture. DJE introduces distribution modeling for graph nodes, where the distribution representations derive both importance and uncertainty estimates. Additionally, DJE facilitates effective pseudo-label generation for the unlabeled data to enrich the training samples. Based on labeled and pseudo-labeled data, EASING develops effective semi-supervised heteroscedastic learning with varying node uncertainty regularization. Extensive experiments on three real-world datasets highlight the superior performance of EASING compared to competing methods. Codes are available via https://github.com/yankai-chen/EASING.
Authors: Shiv Shankar, Tomas Geffner
Abstract: Flow matching models typically use linear interpolants to define the forward/noise addition process. This, together with the independent coupling between noise and target distributions, yields a vector field which is often non-straight. Such curved fields lead to a slow inference/generation process. In this work, we propose to learn flexible (potentially curved) interpolants in order to learn straight vector fields to enable faster generation. We formulate this via a multi-level optimization problem and propose an efficient approximate procedure to solve it. Our framework provides an end-to-end and simulation-free optimization procedure, which can be leveraged to learn straight line generative trajectories.
Authors: Juan Javier Diaz-Mejia, Elias Williams, Octavian Focsa, Dylan Mendonca, Swechha Singh, Brendan Innes, Sam Cooper
Abstract: Many methods have been proposed for removing batch effects and aligning single-cell RNA (scRNA) datasets. However, performance is typically evaluated based on multiple parameters and few datasets, creating challenges in assessing which method is best for aligning data at scale. Here, we introduce the K-Neighbors Intersection (KNI) score, a single score that both penalizes batch effects and measures accuracy at cross-dataset cell-type label prediction alongside carefully curated small (scMARK) and large (scREF) benchmarks comprising 11 and 46 human scRNA studies respectively, where we have standardized author labels. Using the KNI score, we evaluate and optimize approaches for cross-dataset single-cell RNA integration. We introduce Batch Adversarial single-cell Variational Inference (BA-scVI), as a new variant of scVI that uses adversarial training to penalize batch-effects in the encoder and decoder, and show this approach outperforms other methods. In the resulting aligned space, we find that the granularity of cell-type groupings is conserved, supporting the notion that whole-organism cell-type maps can be created by a single model without loss of information.
Authors: Masane Fuchi, Tomohiro Takagi
Abstract: Score-based or diffusion models generate high-quality tabular data, surpassing GAN-based and VAE-based models. However, these methods require substantial training time. In this paper, we introduce RecTable, which uses the rectified flow modeling, applied in such as text-to-image generation and text-to-video generation. RecTable features a simple architecture consisting of a few stacked gated linear unit blocks. Additionally, our training strategies are also simple, incorporating a mixed-type noise distribution and a logit-normal timestep distribution. Our experiments demonstrate that RecTable achieves competitive performance compared to the several state-of-the-art diffusion and score-based models while reducing the required training time. Our code is available at https://github.com/fmp453/rectable.
Authors: Z. Zarezadeh, N. Zarezadeh
Abstract: Despite the many challenges in exploratory data analysis, artificial neural networks have motivated strong interests in scientists and researchers both in theoretical as well as practical applications. Among sources of such popularity of artificial neural networks the ability of modeling non-linear dynamical systems, generalization, and adaptation possibilities should be mentioned. Despite this, there is still significant debate about the role of various underlying stochastic processes in stabilizing a unique structure for data learning and prediction. One of such obstacles to the theoretical and numerical study of machine intelligent systems is the curse of dimensionality and the sampling from high-dimensional probability distributions. In general, this curse prevents efficient description of states, providing a significant complexity barrier for the system to be efficiently described and studied. In this strand of research, direct treatment and description of such abstract notions of learning theory in terms of quantum information be one of the most favorable candidates. Hence, the subject matter of these articles is devoted to problems of design, adaptation and the formulations of computationally hard problems in terms of quantum mechanical systems. In order to characterize the microscopic description of such dynamics in the language of inferential statistics, covariance matrix estimation of d-dimensional Gaussian densities and Bayesian interpretation of eigenvalue problem for dynamical systems is assessed.
Authors: Soham Sane
Abstract: This paper introduces a theoretical framework for a Transformer-augmented, sectional Mixture-of-Experts (MoE) architecture that aims to enhance computational efficiency while preserving model scalability. Unlike conventional MoE models, which route entire token embeddings to selected experts, our approach portions the embedding dimension itself -- assigning segments of each token's representation to dedicated experts. To combat losses in token representation, we utilize a pre-expert transformer layer to recompute attention across tokens and reduce the sequence length dimensionality. We extend our theory by deriving optimal scaling laws that a non-linear relationship between the number of experts and factors such as model dimensionality, sequence length, and system overhead. These formulations yield closed-form and numerically-solvable expressions for identifying the optimal expert count under given architectural and hardware constraints. As a result, our framework not only provides theoretical bounds for computing efficiency with varying frameworks but also guides practical design choices for scaling large models effectively. While empirical validation is pending, we present a comprehensive experimental road map to evaluate the framework's efficiency, scalability, and practicality in future work.
Authors: Shakiba Rahimiaghdam, Hande Alemdar
Abstract: Ensuring transparency in machine learning decisions is critically important, especially in sensitive sectors such as healthcare, finance, and justice. Despite this, some popular explainable algorithms, such as Local Interpretable Model-agnostic Explanations (LIME), often produce unstable explanations due to the random generation of perturbed samples. Random perturbation introduces small changes or noise to modified instances of the original data, leading to inconsistent explanations. Even slight variations in the generated samples significantly affect the explanations provided by such models, undermining trust and hindering the adoption of interpretable models. To address this challenge, we propose MindfulLIME, a novel algorithm that intelligently generates purposive samples using a graph-based pruning algorithm and uncertainty sampling. MindfulLIME substantially improves the consistency of visual explanations compared to random sampling approaches. Our experimental evaluation, conducted on a widely recognized chest X-ray dataset, confirms MindfulLIME's stability with a 100% success rate in delivering reliable explanations under identical conditions. Additionally, MindfulLIME improves the localization precision of visual explanations by reducing the distance between the generated explanations and the actual local annotations compared to LIME. We also performed comprehensive experiments considering various segmentation algorithms and sample numbers, focusing on stability, quality, and efficiency. The results demonstrate the outstanding performance of MindfulLIME across different segmentation settings, generating fewer high-quality samples within a reasonable processing time. By addressing the stability limitations of LIME in image data, MindfulLIME enhances the trustworthiness and interpretability of machine learning models in specific medical imaging applications, a critical domain.
Authors: Kang An, Yuxing Liu, Rui Pan, Shiqian Ma, Donald Goldfarb, Tong Zhang
Abstract: Training deep neural networks (DNNs) is a structured optimization problem, because the parameters are naturally represented by matrices and tensors rather than simple vectors. Under this structural representation, it has been widely observed that gradients are low-rank and Hessians are approximately block-wise diagonal. These structured properties are crucial for designing efficient optimization algorithms but may not be utilized by current popular optimizers like Adam. In this paper, we present a novel optimization algorithm ASGO that capitalizes on these properties by employing a preconditioner that is adaptively updated using structured gradients. By fine-grained theoretical analysis, ASGO is proven to achieve superior convergence rates compared to existing structured gradient methods. Based on the convergence theory, we further demonstrate that ASGO can benefit from the low-rank and block-wise diagonal properties. We also discuss practical modifications of ASGO and empirically verify the effectiveness of the algorithm on language model tasks.
Authors: Clara Fannjiang, Ji Won Park
Abstract: Algorithms for machine learning-guided design, or design algorithms, use machine learning-based predictions to propose novel objects with desired property values. Given a new design task -- for example, to design novel proteins with high binding affinity to a therapeutic target -- one must choose a design algorithm and specify any hyperparameters and predictive and/or generative models involved. How can these decisions be made such that the resulting designs are successful? This paper proposes a method for design algorithm selection, which aims to select design algorithms that will produce a distribution of design labels satisfying a user-specified success criterion -- for example, that at least ten percent of designs' labels exceed a threshold. It does so by combining designs' predicted property values with held-out labeled data to reliably forecast characteristics of the label distributions produced by different design algorithms, building upon techniques from prediction-powered inference. The method is guaranteed with high probability to return design algorithms that yield successful label distributions (or the null set if none exist), if the density ratios between the design and labeled data distributions are known. We demonstrate the method's effectiveness in simulated protein and RNA design tasks, in settings with either known or estimated density ratios.
Authors: Haotian Yang, Zhuoran Wang, Benson Chou, Sophie Xu, Hao Wang, Jingxian Wang, Qizhen Zhang
Abstract: Federated Learning (FL) enables distributed ML model training on private user data at the global scale. Despite the potential of FL demonstrated in many domains, an in-depth view of its impact on model accuracy remains unclear. In this paper, we investigate, systematically, how this learning paradigm can affect the accuracy of state-of-the-art ML models for a variety of ML tasks. We present an empirical study that involves various data types: text, image, audio, and video, and FL configuration knobs: data distribution, FL scale, client sampling, and local and global computations. Our experiments are conducted in a unified FL framework to achieve high fidelity, with substantial human efforts and resource investments. Based on the results, we perform a quantitative analysis of the impact of FL, and highlight challenging scenarios where applying FL degrades the accuracy of the model drastically and identify cases where the impact is negligible. The detailed and extensive findings can benefit practical deployments and future development of FL.
Authors: Zichen Liu, Changyu Chen, Wenjun Li, Penghui Qi, Tianyu Pang, Chao Du, Wee Sun Lee, Min Lin
Abstract: DeepSeek-R1-Zero has shown that reinforcement learning (RL) at scale can directly enhance the reasoning capabilities of LLMs without supervised fine-tuning. In this work, we critically examine R1-Zero-like training by analyzing its two core components: base models and RL. We investigate a wide range of base models, including DeepSeek-V3-Base, to understand how pretraining characteristics influence RL performance. Our analysis reveals that DeepSeek-V3-Base already exhibit ''Aha moment'', while Qwen2.5 base models demonstrate strong reasoning capabilities even without prompt templates, suggesting potential pretraining biases. Additionally, we identify an optimization bias in Group Relative Policy Optimization (GRPO), which artificially increases response length (especially for incorrect outputs) during training. To address this, we introduce Dr. GRPO, an unbiased optimization method that improves token efficiency while maintaining reasoning performance. Leveraging these insights, we present a minimalist R1-Zero recipe that achieves 43.3% accuracy on AIME 2024 with a 7B base model, establishing a new state-of-the-art. Our code is available at https://github.com/sail-sg/understand-r1-zero.
Authors: Daniele Lotito
Abstract: Associative networks theory is increasingly providing tools to interpret update rules of artificial neural networks. At the same time, deriving neural learning rules from a solid theory remains a fundamental challenge. We make some steps in this direction by considering general energy-based associative networks of continuous neurons and synapses that evolve in multiple time scales. We use the separation of these timescales to recover a limit in which the activation of the neurons, the energy of the system and the neural dynamics can all be recovered from a generating function. By allowing the generating function to depend on memories, we recover the conventional Hebbian modeling choice for the interaction strength between neurons. Finally, we propose and discuss a dynamics of memories that enables us to include learning in this framework.
Authors: Mengqi Lou, Kabir Aladin Verchand, Sara Fridovich-Keil, Ashwin Pananjady
Abstract: We consider the problem of signal reconstruction for computed tomography (CT) under a nonlinear forward model that accounts for exponential signal attenuation, a polychromatic X-ray source, general measurement noise (e.g. Poisson shot noise), and observations acquired over multiple wavelength windows. We develop a simple iterative algorithm for single-material reconstruction, which we call EXACT (EXtragradient Algorithm for Computed Tomography), based on formulating our estimate as the fixed point of a monotone variational inequality. We prove guarantees on the statistical and computational performance of EXACT under practical assumptions on the measurement process. We also consider a recently introduced variant of this model with Gaussian measurements, and present sample and iteration complexity bounds for EXACT that improve upon those of existing algorithms. We apply our EXACT algorithm to a CT phantom image recovery task and show that it often requires fewer X-ray projection exposures, lower source intensity, and less computation time to achieve similar reconstruction quality to existing methods.
Authors: Suchanuch Piriyasatit, Chaohao Yuan, Ercan Engin Kuruoglu
Abstract: Dynamic network embedding methods transform nodes in a dynamic network into low-dimensional vectors while preserving network characteristics, facilitating tasks such as node classification and community detection. Several embedding methods have been proposed to capture structural proximity among nodes in a network, where densely connected communities are preserved, while others have been proposed to preserve structural equivalence among nodes, capturing their structural roles regardless of their relative distance in the network. However, most existing methods that aim to preserve both network characteristics mainly focus on static networks and those designed for dynamic networks do not explicitly account for inter-snapshot structural properties. This paper proposes a novel unifying dynamic network embedding method that simultaneously preserves both structural proximity and equivalence while considering inter-snapshot structural relationships in a dynamic network. Specifically, to define structural equivalence in a dynamic network, we use temporal subgraphs, known as dynamic graphlets, to capture how a node's neighborhood structure evolves over time. We then introduce a temporal-structural random walk to flexibly sample time-respecting sequences of nodes, considering both their temporal proximity and similarity in evolving structures. The proposed method is evaluated using five real-world networks on node classification where it outperforms benchmark methods, showing its effectiveness and flexibility in capturing various aspects of a network.
Authors: Pinhao Song
Abstract: Object detection aims to obtain the location and the category of specific objects in a given image, which includes two tasks: classification and location. In recent years, researchers tend to apply object detection to underwater robots equipped with vision systems to complete tasks including seafood fishing, fish farming, biodiversity monitoring and so on. However, the diversity and complexity of underwater environments bring new challenges to object detection. First, aquatic organisms tend to live together, which leads to severe occlusion. Second, theaquatic organisms are good at hiding themselves, which have a similar color to the background. Third, the various water quality and changeable and extreme lighting conditions lead to the distorted, low contrast, blue or green images obtained by the underwater camera, resulting in domain shift. And the deep model is generally vulnerable to facing domain shift. Fourth, the movement of the underwater robot leads to the blur of the captured image and makes the water muddy, which results in low visibility of the water. This paper investigates the problems brought by the underwater environment mentioned above, and aims to design a high-performance and robust underwater object detector.
Authors: Qiusheng Huang, Xiaohui Zhong, Xu Fan, Lei Chen, Hao Li
Abstract: Similar to conventional video generation, current deep learning-based weather prediction frameworks often lack explicit physical constraints, leading to unphysical outputs that limit their reliability for operational forecasting. Among various physical processes requiring proper representation, radiation plays a fundamental role as it drives Earth's weather and climate systems. However, accurate simulation of radiative transfer processes remains challenging for traditional numerical weather prediction (NWP) models due to their inherent complexity and high computational costs. Here, we propose FuXi-RTM, a hybrid physics-guided deep learning framework designed to enhance weather forecast accuracy while enforcing physical consistency. FuXi-RTM integrates a primary forecasting model (FuXi) with a fixed deep learning-based radiative transfer model (DLRTM) surrogate that efficiently replaces conventional radiation parameterization schemes. This represents the first deep learning-based weather forecasting framework to explicitly incorporate physical process modeling. Evaluated over a comprehensive 5-year dataset, FuXi-RTM outperforms its unconstrained counterpart in 88.51% of 3320 variable and lead time combinations, with improvements in radiative flux predictions. By incorporating additional physical processes, FuXi-RTM paves the way for next-generation weather forecasting systems that are both accurate and physically consistent.
Authors: Em\'eric Gbaguidi
Abstract: Stochastic coordinate descent algorithms are efficient methods in which each iterate is obtained by fixing most coordinates at their values from the current iteration, and approximately minimizing the objective with respect to the remaining coordinates. However, this approach is usually restricted to canonical basis vectors of $\mathbb{R}^d$. In this paper, we develop a new class of stochastic gradient descent algorithms with random search directions which uses the directional derivative of the gradient estimate following more general random vectors. We establish the almost sure convergence of these algorithms with decreasing step. We further investigate their central limit theorem and pay particular attention to analyze the impact of the search distributions on the asymptotic covariance matrix. We also provide the non-asymptotic $\mathbb{L}^p$ rates of convergence.
Authors: Sakshi Dhankhar, Stefan Wittek, Hamidreza Eivazi, Andreas Rausch
Abstract: Study Region: Goslar and G\"ottingen, Lower Saxony, Germany. Study Focus: In July 2017, the cities of Goslar and G\"ottingen experienced severe flood events characterized by short warning time of only 20 minutes, resulting in extensive regional flooding and significant damage. This highlights the critical need for a more reliable and timely flood forecasting system. This paper presents a comprehensive study on the impact of radar-based precipitation data on forecasting river water levels in Goslar. Additionally, the study examines how precipitation influences water level forecasts in G\"ottingen. The analysis integrates radar-derived spatiotemporal precipitation patterns with hydrological sensor data obtained from ground stations to evaluate the effectiveness of this approach in improving flood prediction capabilities. New Hydrological Insights for the Region: A key innovation in this paper is the use of residual-based modeling to address the non-linearity between precipitation images and water levels, leading to a Spatiotemporal Radar-based Precipitation Model with residuals (STRPMr). Unlike traditional hydrological models, our approach does not rely on upstream data, making it independent of additional hydrological inputs. This independence enhances its adaptability and allows for broader applicability in other regions with RADOLAN precipitation. The deep learning architecture integrates (2+1)D convolutional neural networks for spatial and temporal feature extraction with LSTM for timeseries forecasting. The results demonstrate the potential of the STRPMr for capturing extreme events and more accurate flood forecasting.
Authors: Valerii A. Zuev, Elena G. Salmagambetova, Stepan N. Djakov, Lev V. Utkin
Abstract: Epilepsy is typically diagnosed through electroencephalography (EEG) and long-term video-EEG (vEEG) monitoring. The manual analysis of vEEG recordings is time-consuming, necessitating automated tools for seizure detection. Recent advancements in machine learning have shown promise in real-time seizure detection and prediction using EEG and video data. However, diversity of seizure symptoms, markup ambiguities, and limited availability of multimodal datasets hinder progress. This paper reviews the latest developments in automated video-EEG analysis and discusses the integration of multimodal data. We also propose a novel pipeline for treatment effect estimation from vEEG data using concept-based learning, offering a pathway for future research in this domain.
Authors: Navami Kairanda, Marc Habermann, Shanthika Naik, Christian Theobalt, Vladislav Golyanik
Abstract: 3D reconstruction of highly deformable surfaces (e.g. cloths) from monocular RGB videos is a challenging problem, and no solution provides a consistent and accurate recovery of fine-grained surface details. To account for the ill-posed nature of the setting, existing methods use deformation models with statistical, neural, or physical priors. They also predominantly rely on nonadaptive discrete surface representations (e.g. polygonal meshes), perform frame-by-frame optimisation leading to error propagation, and suffer from poor gradients of the mesh-based differentiable renderers. Consequently, fine surface details such as cloth wrinkles are often not recovered with the desired accuracy. In response to these limitations, we propose ThinShell-SfT, a new method for non-rigid 3D tracking that represents a surface as an implicit and continuous spatiotemporal neural field. We incorporate continuous thin shell physics prior based on the Kirchhoff-Love model for spatial regularisation, which starkly contrasts the discretised alternatives of earlier works. Lastly, we leverage 3D Gaussian splatting to differentiably render the surface into image space and optimise the deformations based on analysis-bysynthesis principles. Our Thin-Shell-SfT outperforms prior works qualitatively and quantitatively thanks to our continuous surface formulation in conjunction with a specially tailored simulation prior and surface-induced 3D Gaussians. See our project page at https://4dqv.mpiinf.mpg.de/ThinShellSfT.
Authors: Jonathan Sauder, Viktor Domazetoski, Guilhem Banc-Prandi, Gabriela Perna, Anders Meibom, Devis Tuia
Abstract: Coral reefs are declining worldwide due to climate change and local stressors. To inform effective conservation or restoration, monitoring at the highest possible spatial and temporal resolution is necessary. Conventional coral reef surveying methods are limited in scalability due to their reliance on expert labor time, motivating the use of computer vision tools to automate the identification and abundance estimation of live corals from images. However, the design and evaluation of such tools has been impeded by the lack of large high quality datasets. We release the Coralscapes dataset, the first general-purpose dense semantic segmentation dataset for coral reefs, covering 2075 images, 39 benthic classes, and 174k segmentation masks annotated by experts. Coralscapes has a similar scope and the same structure as the widely used Cityscapes dataset for urban scene segmentation, allowing benchmarking of semantic segmentation models in a new challenging domain which requires expert knowledge to annotate. We benchmark a wide range of semantic segmentation models, and find that transfer learning from Coralscapes to existing smaller datasets consistently leads to state-of-the-art performance. Coralscapes will catalyze research on efficient, scalable, and standardized coral reef surveying methods based on computer vision, and holds the potential to streamline the development of underwater ecological robotics.
Authors: Yimeng Min, Carla P. Gomes
Abstract: We introduce PLUME search, a data-driven framework that enhances search efficiency in combinatorial optimization through unsupervised learning. Unlike supervised or reinforcement learning, PLUME search learns directly from problem instances using a permutation-based loss with a non-autoregressive approach. We evaluate its performance on the quadratic assignment problem, a fundamental NP-hard problem that encompasses various combinatorial optimization problems. Experimental results demonstrate that PLUME search consistently improves solution quality. Furthermore, we study the generalization behavior and show that the learned model generalizes across different densities and sizes.
Authors: Maksim Borisov, Zhanibek Kozhirbayev, Valentin Malykh
Abstract: Machine translation for low resource language pairs is a challenging task. This task could become extremely difficult once a speaker uses code switching. We propose a method to build a machine translation model for code-switched Kazakh-Russian language pair with no labeled data. Our method is basing on generation of synthetic data. Additionally, we present the first codeswitching Kazakh-Russian parallel corpus and the evaluation results, which include a model achieving 16.48 BLEU almost reaching an existing commercial system and beating it by human evaluation.
Authors: Paul Bertin, Joseph D. Viviano, Alejandro Tejada-Lapuerta, Weixu Wang, Stefan Bauer, Fabian J. Theis, Yoshua Bengio
Abstract: Single-cell data provide high-dimensional measurements of the transcriptional states of cells, but extracting insights into the regulatory functions of genes, particularly identifying transcriptional mechanisms affected by biological perturbations, remains a challenge. Many perturbations induce compensatory cellular responses, making it difficult to distinguish direct from indirect effects on gene regulation. Modeling how gene regulatory functions shape the temporal dynamics of these responses is key to improving our understanding of biological perturbations. Dynamical models based on differential equations offer a principled way to capture transcriptional dynamics, but their application to single-cell data has been hindered by computational constraints, stochasticity, sparsity, and noise. Existing methods either rely on low-dimensional representations or make strong simplifying assumptions, limiting their ability to model transcriptional dynamics at scale. We introduce a Functional and Learnable model of Cell dynamicS, FLeCS, that incorporates gene network structure into coupled differential equations to model gene regulatory functions. Given (pseudo)time-series single-cell data, FLeCS accurately infers cell dynamics at scale, provides improved functional insights into transcriptional mechanisms perturbed by gene knockouts, both in myeloid differentiation and K562 Perturb-seq experiments, and simulates single-cell trajectories of A549 cells following small-molecule perturbations.
Authors: Ajitesh Srivastava, Aryan Shetty, Eric Rice
Abstract: Studying peer relationships is crucial in solving complex challenges underserved communities face and designing interventions. The effectiveness of such peer-based interventions relies on accurate network data regarding individual attributes and social influences. However, these datasets are often collected through self-reported surveys, introducing ambiguities in network construction. These ambiguities make it challenging to fully utilize the network data to understand the issues and to design the best interventions. We propose and solve two variations of link ambiguities in such network data -- (i) which among the two candidate links exists, and (ii) if a candidate link exists. We design a Graph Attention Network (GAT) that accounts for personal attributes and network relationships on real-world data with real and simulated ambiguities. We also demonstrate that by resolving these ambiguities, we improve network accuracy, and in turn, improve suicide risk prediction. We also uncover patterns using GNNExplainer to provide additional insights into vital features and relationships. This research demonstrates the potential of Graph Neural Networks (GNN) to advance real-world network data analysis facilitating more effective peer interventions across various fields.
Authors: Pin-Jie Lin, Rishab Balasubramanian, Fengyuan Liu, Nikhil Kandpal, Tu Vu
Abstract: Modern LLMs struggle with efficient updates, as each new pretrained model version requires repeating expensive alignment processes. This challenge also applies to domain- or language-specific models, where fine-tuning on specialized data must be redone for every new base model release. In this paper, we explore the transfer of fine-tuning updates between model versions. Specifically, we derive the diff vector from one source model version, which represents the weight changes from fine-tuning, and apply it to the base model of a different target version. Through empirical evaluations on various open-weight model versions, we show that transferring diff vectors can significantly improve the target base model, often achieving performance comparable to its fine-tuned counterpart. For example, reusing the fine-tuning updates from Llama 3.0 8B leads to an absolute accuracy improvement of 10.7% on GPQA over the base Llama 3.1 8B without additional training, surpassing Llama 3.1 8B Instruct. In a multilingual model development setting, we show that this approach can significantly increase performance on target-language tasks without retraining, achieving an absolute improvement of 4.7% and 15.5% on Global MMLU for Malagasy and Turkish, respectively, compared to Llama 3.1 8B Instruct. Our controlled experiments reveal that fine-tuning transfer is most effective when the source and target models are linearly connected in the parameter space. Additionally, we demonstrate that fine-tuning transfer offers a stronger and more computationally efficient starting point for further fine-tuning. Finally, we propose an iterative recycling-then-finetuning approach for continuous model development, which improves both efficiency and effectiveness. Our findings suggest that fine-tuning transfer is a viable strategy to reduce training costs while maintaining model performance.
Authors: Hongwei Wen, Annika Betken, Wouter Koolen
Abstract: Robust regression aims to develop methods for estimating an unknown regression function in the presence of outliers, heavy-tailed distributions, or contaminated data, which can severely impact performance. Most existing theoretical results in robust regression assume that the noise has a finite absolute mean, an assumption violated by certain distributions, such as Cauchy and some Pareto noise. In this paper, we introduce a generalized Cauchy noise framework that accommodates all noise distributions with finite moments of any order, even when the absolute mean is infinite. Within this framework, we study the \textit{kernel Cauchy ridge regressor} (\textit{KCRR}), which minimizes a regularized empirical Cauchy risk to achieve robustness. To derive the $L_2$-risk bound for KCRR, we establish a connection between the excess Cauchy risk and $L_2$-risk for sufficiently large scale parameters of the Cauchy loss, which reveals that these two risks are equivalent. Furthermore, under the assumption that the regression function satisfies H\"older smoothness, we derive excess Cauchy risk bounds for KCRR, showing improved performance as the scale parameter decreases. By considering the twofold effect of the scale parameter on the excess Cauchy risk and its equivalence with the $L_2$-risk, we establish the almost minimax-optimal convergence rate for KCRR in terms of $L_2$-risk, highlighting the robustness of the Cauchy loss in handling various types of noise. Finally, we validate the effectiveness of KCRR through experiments on both synthetic and real-world datasets under diverse noise corruption scenarios.
Authors: Oren Kraus, Federico Comitani, John Urbanik, Kian Kenyon-Dean, Lakshmanan Arumugam, Saber Saberian, Cas Wognum, Safiye Celik, Imran S. Haque
Abstract: High Content Screening (HCS) microscopy datasets have transformed the ability to profile cellular responses to genetic and chemical perturbations, enabling cell-based inference of drug-target interactions (DTI). However, the adoption of representation learning methods for HCS data has been hindered by the lack of accessible datasets and robust benchmarks. To address this gap, we present RxRx3-core, a curated and compressed subset of the RxRx3 dataset, and an associated DTI benchmarking task. At just 18GB, RxRx3-core significantly reduces the size barrier associated with large-scale HCS datasets while preserving critical data necessary for benchmarking representation learning models against a zero-shot DTI prediction task. RxRx3-core includes 222,601 microscopy images spanning 736 CRISPR knockouts and 1,674 compounds at 8 concentrations. RxRx3-core is available on HuggingFace and Polaris, along with pre-trained embeddings and benchmarking code, ensuring accessibility for the research community. By providing a compact dataset and robust benchmarks, we aim to accelerate innovation in representation learning methods for HCS data and support the discovery of novel biological insights.
Authors: M\'elisande Teng, Arthur Ouaknine, Etienne Lalibert\'e, Yoshua Bengio, David Rolnick, Hugo Larochelle
Abstract: The potential of tree planting as a natural climate solution is often undermined by inadequate monitoring of tree planting projects. Current monitoring methods involve measuring trees by hand for each species, requiring extensive cost, time, and labour. Advances in drone remote sensing and computer vision offer great potential for mapping and characterizing trees from aerial imagery, and large pre-trained vision models, such as the Segment Anything Model (SAM), may be a particularly compelling choice given limited labeled data. In this work, we compare SAM methods for the task of automatic tree crown instance segmentation in high resolution drone imagery of young tree plantations. We explore the potential of SAM for this task, and find that methods using SAM out-of-the-box do not outperform a custom Mask R-CNN, even with well-designed prompts, but that there is potential for methods which tune SAM further. We also show that predictions can be improved by adding Digital Surface Model (DSM) information as an input.
Authors: Liangzhi Shi, Yulin Liu, Lingqi Zeng, Bo Ai, Zhengdong Hong, Hao Su
Abstract: How can robots learn dexterous grasping skills efficiently and apply them adaptively based on user instructions? This work tackles two key challenges: efficient skill acquisition from limited human demonstrations and context-driven skill selection. We introduce AdaDexGrasp, a framework that learns a library of grasping skills from a single human demonstration per skill and selects the most suitable one using a vision-language model (VLM). To improve sample efficiency, we propose a trajectory following reward that guides reinforcement learning (RL) toward states close to a human demonstration while allowing flexibility in exploration. To learn beyond the single demonstration, we employ curriculum learning, progressively increasing object pose variations to enhance robustness. At deployment, a VLM retrieves the appropriate skill based on user instructions, bridging low-level learned skills with high-level intent. We evaluate AdaDexGrasp in both simulation and real-world settings, showing that our approach significantly improves RL efficiency and enables learning human-like grasp strategies across varied object configurations. Finally, we demonstrate zero-shot transfer of our learned policies to a real-world PSYONIC Ability Hand, with a 90% success rate across objects, significantly outperforming the baseline.
Authors: Ugochukwu Ejike Akpudo, Yongsheng Gao, Jun Zhou, Andrew Lewis
Abstract: Convolutional neural networks (CNNs) have succeeded remarkably in various computer vision tasks. However, they are not intrinsically explainable. While the feature-level understanding of CNNs reveals where the models looked, concept-based explainability methods provide insights into what the models saw. However, their assumption of linear reconstructability of image activations fails to capture the intricate relationships within these activations. Their Fidelity-only approach to evaluating global explanations also presents a new concern. For the first time, we address these limitations with the novel Transformative Nonlinear Concept Explainer (TraNCE) for CNNs. Unlike linear reconstruction assumptions made by existing methods, TraNCE captures the intricate relationships within the activations. This study presents three original contributions to the CNN explainability literature: (i) An automatic concept discovery mechanism based on variational autoencoders (VAEs). This transformative concept discovery process enhances the identification of meaningful concepts from image activations. (ii) A visualization module that leverages the Bessel function to create a smooth transition between prototypical image pixels, revealing not only what the CNN saw but also what the CNN avoided, thereby mitigating the challenges of concept duplication as documented in previous works. (iii) A new metric, the Faith score, integrates both Coherence and Fidelity for a comprehensive evaluation of explainer faithfulness and consistency.
Authors: Mingfu Liang, Jiahuan Zhou, Xu Zou, Ying Wu
Abstract: Existing progress in object keypoint estimation primarily benefits from the conventional supervised learning paradigm based on numerous data labeled with pre-defined keypoints. However, these well-trained models can hardly detect the undefined new keypoints in test time, which largely hinders their feasibility for diverse downstream tasks. To handle this, various solutions are explored but still suffer from either limited generalizability or transferability. Therefore, in this paper, we explore a novel keypoint learning paradigm in that we only annotate new keypoints in the new data and incrementally train the model, without retaining any old data, called Incremental object Keypoint Learning (IKL). A two-stage learning scheme as a novel baseline tailored to IKL is developed. In the first Knowledge Association stage, given the data labeled with only new keypoints, an auxiliary KA-Net is trained to automatically associate the old keypoints to these new ones based on their spatial and intrinsic anatomical relations. In the second Mutual Promotion stage, based on a keypoint-oriented spatial distillation loss, we jointly leverage the auxiliary KA-Net and the old model for knowledge consolidation to mutually promote the estimation of all old and new keypoints. Owing to the investigation of the correlations between new and old keypoints, our proposed method can not just effectively mitigate the catastrophic forgetting of old keypoints, but may even further improve the estimation of the old ones and achieve a positive transfer beyond anti-forgetting. Such an observation has been solidly verified by extensive experiments on different keypoint datasets, where our method exhibits superiority in alleviating the forgetting issue and boosting performance while enjoying labeling efficiency even under the low-shot data regime.
Authors: Hideaki Ishibashi, Kota Matsui, Kentaro Kutsukake, Hideitsu Hino
Abstract: The level set estimation problem seeks to identify regions within a set of candidate points where an unknown and costly to evaluate function's value exceeds a specified threshold, providing an efficient alternative to exhaustive evaluations of function values. Traditional methods often use sequential optimization strategies to find $\epsilon$-accurate solutions, which permit a margin around the threshold contour but frequently lack effective stopping criteria, leading to excessive exploration and inefficiencies. This paper introduces an acquisition strategy for level set estimation that incorporates a stopping criterion, ensuring the algorithm halts when further exploration is unlikely to yield improvements, thereby reducing unnecessary function evaluations. We theoretically prove that our method satisfies $\epsilon$-accuracy with a confidence level of $1 - \delta$, addressing a key gap in existing approaches. Furthermore, we show that this also leads to guarantees on the lower bounds of performance metrics such as F-score. Numerical experiments demonstrate that the proposed acquisition function achieves comparable precision to existing methods while confirming that the stopping criterion effectively terminates the algorithm once adequate exploration is completed.
Authors: William Gilpin
Abstract: Single-cell sequencing technology maps cells to a high-dimensional space encoding their internal activity. This process mirrors parallel developments in machine learning, where large language models ingest unstructured text by converting words into discrete tokens embedded within a high-dimensional vector space. This perspective explores how advances in understanding the structure of language embeddings can inform ongoing efforts to analyze and visualize single cell datasets. We discuss how the context of tokens influences the geometry of embedding space, and the role of low-dimensional manifolds in shaping this space's robustness and interpretability. We highlight new developments in language modeling, such as interpretability probes and in-context reasoning, that can inform future efforts to construct and consolidate cell atlases.
Authors: Chenwei Zhang, Anne Condon, Khanh Dao Duc
Abstract: Enhancing cryogenic electron microscopy (cryo-EM) 3D density maps at intermediate resolution (4-8 {\AA}) is crucial in protein structure determination. Recent advances in deep learning have led to the development of automated approaches for enhancing experimental cryo-EM density maps. Yet, these methods are not optimized for intermediate-resolution maps and rely on map density features alone. To address this, we propose CryoSAMU, a novel method designed to enhance 3D cryo-EM density maps of protein structures using structure-aware multimodal U-Nets and trained on curated intermediate-resolution density maps. We comprehensively evaluate CryoSAMU across various metrics and demonstrate its competitive performance compared to state-of-the-art methods. Notably, CryoSAMU achieves significantly faster processing speed, showing promise for future practical applications. Our code is available at https://github.com/chenwei-zhang/CryoSAMU.
Authors: Tao Wu, Tie Luo
Abstract: Adversarial attacks in black-box settings are highly practical, with transfer-based attacks being the most effective at generating adversarial examples (AEs) that transfer from surrogate models to unseen target models. However, their performance significantly degrades when transferring across heterogeneous architectures -- such as CNNs, MLPs, and Vision Transformers (ViTs) -- due to fundamental architectural differences. To address this, we propose Feature Permutation Attack (FPA), a zero-FLOP, parameter-free method that enhances adversarial transferability across diverse architectures. FPA introduces a novel feature permutation (FP) operation, which rearranges pixel values in selected feature maps to simulate long-range dependencies, effectively making CNNs behave more like ViTs and MLPs. This enhances feature diversity and improves transferability both across heterogeneous architectures and within homogeneous CNNs. Extensive evaluations on 14 state-of-the-art architectures show that FPA achieves maximum absolute gains in attack success rates of 7.68% on CNNs, 14.57% on ViTs, and 14.48% on MLPs, outperforming existing black-box attacks. Additionally, FPA is highly generalizable and can seamlessly integrate with other transfer-based attacks to further boost their performance. Our findings establish FPA as a robust, efficient, and computationally lightweight strategy for enhancing adversarial transferability across heterogeneous architectures.
Authors: Felix Vogel, Walid Bousselham, Anna Kukleva, Nina Shvetsova, Hilde Kuehne
Abstract: Vision-language foundation models have shown impressive capabilities across various zero-shot tasks, including training-free localization and grounding, primarily focusing on localizing objects in images. However, leveraging those capabilities to localize actions and events in videos is challenging, as actions have less physical outline and are usually described by higher-level concepts. In this work, we propose VideoGEM, the first training-free spatial action grounding method based on pretrained image- and video-language backbones. Namely, we adapt the self-self attention formulation of GEM to spatial activity grounding. We observe that high-level semantic concepts, such as actions, usually emerge in the higher layers of the image- and video-language models. We, therefore, propose a layer weighting in the self-attention path to prioritize higher layers. Additionally, we introduce a dynamic weighting method to automatically tune layer weights to capture each layer`s relevance to a specific prompt. Finally, we introduce a prompt decomposition, processing action, verb, and object prompts separately, resulting in a better spatial localization of actions. We evaluate the proposed approach on three image- and video-language backbones, CLIP, OpenCLIP, and ViCLIP, and on four video grounding datasets, V-HICO, DALY, YouCook-Interactions, and GroundingYouTube, showing that the proposed training-free approach is able to outperform current trained state-of-the-art approaches for spatial video grounding.
Authors: Ke Ma, Jiaqi Tang, Bin Guo, Fan Dang, Sicong Liu, Zhui Zhu, Lei Wu, Cheng Fang, Ying-Cong Chen, Zhiwen Yu, Yunhao Liu
Abstract: Despite the growing integration of deep models into mobile terminals, the accuracy of these models declines significantly due to various deployment interferences. Test-time adaptation (TTA) has emerged to improve the performance of deep models by adapting them to unlabeled target data online. Yet, the significant memory cost, particularly in resource-constrained terminals, impedes the effective deployment of most backward-propagation-based TTA methods. To tackle memory constraints, we introduce SURGEON, a method that substantially reduces memory cost while preserving comparable accuracy improvements during fully test-time adaptation (FTTA) without relying on specific network architectures or modifications to the original training procedure. Specifically, we propose a novel dynamic activation sparsity strategy that directly prunes activations at layer-specific dynamic ratios during adaptation, allowing for flexible control of learning ability and memory cost in a data-sensitive manner. Among this, two metrics, Gradient Importance and Layer Activation Memory, are considered to determine the layer-wise pruning ratios, reflecting accuracy contribution and memory efficiency, respectively. Experimentally, our method surpasses the baselines by not only reducing memory usage but also achieving superior accuracy, delivering SOTA performance across diverse datasets, architectures, and tasks.
Authors: Adrian Villalobos, Iban Barrutia, Rafael Pena-Alzola, Tomislav Dragicevic, Jose I. Aizpurua
Abstract: Semiconductor devices, especially MOSFETs (Metal-oxide-semiconductor field-effect transistor), are crucial in power electronics, but their reliability is affected by aging processes influenced by cycling and temperature. The primary aging mechanism in discrete semiconductors and power modules is the bond wire lift-off, caused by crack growth due to thermal fatigue. The process is empirically characterized by exponential growth and an abrupt end of life, making long-term aging forecasts challenging. This research presents a comprehensive comparative assessment of different forecasting methods for MOSFET failure forecasting applications. Classical tracking, statistical forecasting and Neural Network (NN) based forecasting models are implemented along with novel Temporal Fusion Transformers (TFTs). A comprehensive comparison is performed assessing their MOSFET ageing forecasting ability for different forecasting horizons. For short-term predictions, all algorithms result in acceptable results, with the best results produced by classical NN forecasting models at the expense of higher computations. For long-term forecasting, only the TFT is able to produce valid outcomes owing to the ability to integrate covariates from the expected future conditions. Additionally, TFT attention points identify key ageing turning points, which indicate new failure modes or accelerated ageing phases.
Authors: Akylas Stratigakos, Panagiotis Andrianesis
Abstract: Short-term forecasting models typically assume the availability of input data (features) when they are deployed and in use. However, equipment failures, disruptions, cyberattacks, may lead to missing features when such models are used operationally, which could negatively affect forecast accuracy, and result in suboptimal operational decisions. In this paper, we use adaptive robust optimization and adversarial machine learning to develop forecasting models that seamlessly handle missing data operationally. We propose linear- and neural network-based forecasting models with parameters that adapt to available features, combining linear adaptation with a novel algorithm for learning data-driven uncertainty set partitions. The proposed adaptive models do not rely on identifying historical missing data patterns and are suitable for real-time operations under stringent time constraints. Extensive numerical experiments on short-term wind power forecasting considering horizons from 15 minutes to 4 hours ahead illustrate that our proposed adaptive models are on par with imputation when data are missing for very short periods (e.g., when only the latest measurement is missing) whereas they significantly outperform imputation when data are missing for longer periods. We further provide insights by showcasing how linear adaptation and data-driven partitions (even with a few subsets) approach the performance of the optimal, yet impractical, method of retraining for every possible realization of missing data.
Authors: Tom Kempton, Stuart Burrell, Connor Cheverall
Abstract: Existing methods for the zero-shot detection of machine-generated text are dominated by three statistical quantities: log-likelihood, log-rank, and entropy. As language models mimic the distribution of human text ever closer, this will limit our ability to build effective detection algorithms. To combat this, we introduce a method for detecting machine-generated text that is entirely agnostic of the generating language model. This is achieved by targeting a defect in the way that decoding strategies, such as temperature or top-k sampling, normalize conditional probability measures. This method can be rigorously theoretically justified, is easily explainable, and is conceptually distinct from existing methods for detecting machine-generated text. We evaluate our detector in the white and black box settings across various language models, datasets, and passage lengths. We also study the effect of paraphrasing attacks on our detector and the extent to which it is biased against non-native speakers. In each of these settings, the performance of our test is at least comparable to that of other state-of-the-art text detectors, and in some cases, we strongly outperform these baselines.
Authors: Georg Sch\"afer, Tatjana Krau, Jakob Rehrl, Stefan Huber, Simon Hirlaender
Abstract: Reinforcement Learning (RL) offers promising solutions for control tasks in industrial cyber-physical systems (ICPSs), yet its real-world adoption remains limited. This paper demonstrates how seemingly small but well-designed modifications to the RL problem formulation can substantially improve performance, stability, and sample efficiency. We identify and investigate key elements of RL problem formulation and show that these enhance both learning speed and final policy quality. Our experiments use a one-degree-of-freedom (1-DoF) helicopter testbed, the Quanser Aero~2, which features non-linear dynamics representative of many industrial settings. In simulation, the proposed problem design principles yield more reliable and efficient training, and we further validate these results by training the agent directly on physical hardware. The encouraging real-world outcomes highlight the potential of RL for ICPS, especially when careful attention is paid to the design principles of problem formulation. Overall, our study underscores the crucial role of thoughtful problem formulation in bridging the gap between RL research and the demands of real-world industrial systems.
Authors: Llu\'is Palma, Alejandro Peraza, David Civantos, Amanda Duarte, Stefano Materia, \'Angel G. Mu\~noz, Jes\'us Pe\~na, Laia Romero, Albert Soret, Markus G. Donat
Abstract: Most operational climate services providers base their seasonal predictions on initialised general circulation models (GCMs) or statistical techniques that fit past observations. GCMs require substantial computational resources, which limits their capacity. In contrast, statistical methods often lack robustness due to short historical records. Recent works propose machine learning methods trained on climate model output, leveraging larger sample sizes and simulated scenarios. Yet, many of these studies focus on prediction tasks that might be restricted in spatial extent or temporal coverage, opening a gap with existing operational predictions. Thus, the present study evaluates the effectiveness of a methodology that combines variational inference with transformer models to predict fields of seasonal anomalies. The predictions cover all four seasons and are initialised one month before the start of each season. The model was trained on climate model output from CMIP6 and tested using ERA5 reanalysis data. We analyse the method's performance in predicting interannual anomalies beyond the climate change-induced trend. We also test the proposed methodology in a regional context with a use case focused on Europe. While climate change trends dominate the skill of temperature predictions, the method presents additional skill over the climatological forecast in regions influenced by known teleconnections. We reach similar conclusions based on the validation of precipitation predictions. Despite underperforming SEAS5 in most tropics, our model offers added value in numerous extratropical inland regions. This work demonstrates the effectiveness of training generative models on climate model output for seasonal predictions, providing skilful predictions beyond the induced climate change trend at time scales and lead times relevant for user applications.
Authors: Yingdong Shi, Changming Li, Yifan Wang, Yongxiang Zhao, Anqi Pang, Sibei Yang, Jingyi Yu, Kan Ren
Abstract: Diffusion models have demonstrated impressive capabilities in synthesizing diverse content. However, despite their high-quality outputs, these models often perpetuate social biases, including those related to gender and race. These biases can potentially contribute to harmful real-world consequences, reinforcing stereotypes and exacerbating inequalities in various social contexts. While existing research on diffusion bias mitigation has predominantly focused on guiding content generation, it often neglects the intrinsic mechanisms within diffusion models that causally drive biased outputs. In this paper, we investigate the internal processes of diffusion models, identifying specific decision-making mechanisms, termed bias features, embedded within the model architecture. By directly manipulating these features, our method precisely isolates and adjusts the elements responsible for bias generation, permitting granular control over the bias levels in the generated content. Through experiments on both unconditional and conditional diffusion models across various social bias attributes, we demonstrate our method's efficacy in managing generation distribution while preserving image quality. We also dissect the discovered model mechanism, revealing different intrinsic features controlling fine-grained aspects of generation, boosting further research on mechanistic interpretability of diffusion models.
Authors: Haoran Zheng, Renchi Yang, Jianliang Xu
Abstract: Given a graph $G$ and a seed node $v_s$, the objective of local graph clustering (LGC) is to identify a subgraph $C_s \in G$ (a.k.a. local cluster) surrounding $v_s$ in time roughly linear with the size of $C_s$. This approach yields personalized clusters without needing to access the entire graph, which makes it highly suitable for numerous applications involving large graphs. However, most existing solutions merely rely on the topological connectivity between nodes in $G$, rendering them vulnerable to missing or noisy links that are commonly present in real-world graphs. To address this issue, this paper resorts to leveraging the complementary nature of graph topology and node attributes to enhance local clustering quality. To effectively exploit the attribute information, we first formulate the LGC as an estimation of the bidirectional diffusion distribution (BDD), which is specialized for capturing the multi-hop affinity between nodes in the presence of attributes. Furthermore, we propose LACA, an efficient and effective approach for LGC that achieves superb empirical performance on multiple real datasets while maintaining strong locality. The core components of LACA include (i) a fast and theoretically-grounded preprocessing technique for node attributes, (ii) an adaptive algorithm for diffusing any vectors over $G$ with rigorous theoretical guarantees and expedited convergence, and (iii) an effective three-step scheme for BDD approximation. Extensive experiments, comparing 17 competitors on 8 real datasets, show that LACA outperforms all competitors in terms of result quality measured against ground truth local clusters, while also being up to orders of magnitude faster. The code is available at https://github.com/HaoranZ99/alac.
Authors: Jiale Cheng, Ruiliang Lyu, Xiaotao Gu, Xiao Liu, Jiazheng Xu, Yida Lu, Jiayan Teng, Zhuoyi Yang, Yuxiao Dong, Jie Tang, Hongning Wang, Minlie Huang
Abstract: Video generation models have achieved remarkable progress in text-to-video tasks. These models are typically trained on text-video pairs with highly detailed and carefully crafted descriptions, while real-world user inputs during inference are often concise, vague, or poorly structured. This gap makes prompt optimization crucial for generating high-quality videos. Current methods often rely on large language models (LLMs) to refine prompts through in-context learning, but suffer from several limitations: they may distort user intent, omit critical details, or introduce safety risks. Moreover, they optimize prompts without considering the impact on the final video quality, which can lead to suboptimal results. To address these issues, we introduce VPO, a principled framework that optimizes prompts based on three core principles: harmlessness, accuracy, and helpfulness. The generated prompts faithfully preserve user intents and, more importantly, enhance the safety and quality of generated videos. To achieve this, VPO employs a two-stage optimization approach. First, we construct and refine a supervised fine-tuning (SFT) dataset based on principles of safety and alignment. Second, we introduce both text-level and video-level feedback to further optimize the SFT model with preference learning. Our extensive experiments demonstrate that VPO significantly improves safety, alignment, and video quality compared to baseline methods. Moreover, VPO shows strong generalization across video generation models. Furthermore, we demonstrate that VPO could outperform and be combined with RLHF methods on video generation models, underscoring the effectiveness of VPO in aligning video generation models. Our code and data are publicly available at https://github.com/thu-coai/VPO.
Authors: Rakesh Nadig, Vamanan Arulchelvan, Rahul Bera, Taha Shahroodi, Gagandeep Singh, Mohammad Sadrosadati, Jisung Park, Onur Mutlu
Abstract: Hybrid storage systems (HSS) combine multiple storage devices with diverse characteristics to achieve high performance and capacity at low cost. The performance of an HSS highly depends on the effectiveness of two key policies: (1) the data-placement policy, which determines the best-fit storage device for incoming data, and (2) the data-migration policy, which rearranges stored data across the devices to sustain high HSS performance. Prior works focus on improving only data placement or only data migration in HSS, which leads to sub-optimal HSS performance. Unfortunately, no prior work tries to optimize both policies together. Our goal is to design a holistic data-management technique for HSS that optimizes both data-placement and data-migration policies to fully exploit the potential of an HSS. We propose Harmonia, a multi-agent reinforcement learning (RL)-based data-management technique that employs two light-weight autonomous RL agents, a data-placement agent and a data-migration agent, which adapt their policies for the current workload and HSS configuration, and coordinate with each other to improve overall HSS performance. We evaluate Harmonia on a real HSS with up to four heterogeneous storage devices with diverse characteristics. Our evaluation using 17 data-intensive workloads on performance-optimized (cost-optimized) HSS with two storage devices shows that, on average, Harmonia (1) outperforms the best-performing prior approach by 49.5% (31.7%), (2) bridges the performance gap between the best-performing prior work and Oracle by 64.2% (64.3%). On an HSS with three (four) devices, Harmonia outperforms the best-performing prior work by 37.0% (42.0%). Harmonia's performance benefits come with low latency (240ns for inference) and storage overheads (206 KiB for both RL agents together). We plan to open-source Harmonia's implementation to aid future research on HSS.
Authors: Henrik Christiansen, Takashi Maruyama, Federico Errica, Viktor Zaverkin, Makoto Takamoto, Francesco Alesiani
Abstract: We present an end-to-end differentiable molecular simulation framework (DIMOS) for molecular dynamics and Monte Carlo simulations. DIMOS easily integrates machine-learning-based interatomic potentials and implements classical force fields including particle-mesh Ewald electrostatics. Thanks to its modularity, both classical and machine-learning-based approaches can be easily combined into a hybrid description of the system (ML/MM). By supporting key molecular dynamics features such as efficient neighborlists and constraint algorithms for larger time steps, the framework bridges the gap between hand-optimized simulation engines and the flexibility of a PyTorch implementation. The superior performance and the high versatility is probed in different benchmarks and applications, with speed-up factors of up to $170\times$. The advantage of differentiability is demonstrated by an end-to-end optimization of the proposal distribution in a Markov Chain Monte Carlo simulation based on Hamiltonian Monte Carlo. Using these optimized simulation parameters a $3\times$ acceleration is observed in comparison to ad-hoc chosen simulation parameters. The code is available at https://github.com/nec-research/DIMOS.
Authors: Marlies Hafer, Alexander Marx
Abstract: We consider the problem of estimating the expected causal effect $E[Y|do(X)]$ for a target variable $Y$ when treatment $X$ is set by intervention, focusing on continuous random variables. In settings without selection bias or confounding, $E[Y|do(X)] = E[Y|X]$, which can be estimated using standard regression methods. However, regression fails when systematic missingness induced by selection bias, or confounding distorts the data. Boeken et al. [2023] show that when training data is subject to selection, proxy variables unaffected by this process can, under certain constraints, be used to correct for selection bias to estimate $E[Y|X]$, and hence $E[Y|do(X)]$, reliably. When data is additionally affected by confounding, however, this equality is no longer valid. Building on these results, we consider a more general setting and propose a framework that incorporates both selection bias and confounding. Specifically, we derive theoretical conditions ensuring identifiability and recoverability of causal effects under access to external data and proxy variables. We further introduce a two-step regression estimator (TSR), capable of exploiting proxy variables to adjust for selection bias while accounting for confounding. We show that TSR coincides with prior work if confounding is absent, but achieves a lower variance. Extensive simulation studies validate TSR's correctness for scenarios which may include both selection bias and confounding with proxy variables.
Authors: Yunkai Liang, Zhangyu Chen, Pengfei Zuo, Zhi Zhou, Xu Chen, Zhou Yu
Abstract: In large language model (LLM) serving systems, executing each request consists of two phases: the compute-intensive prefill phase and the memory-intensive decoding phase. To prevent performance interference between the two phases, current LLM serving systems typically adopt prefill-decoding disaggregation, where the two phases are split across separate machines. However, we observe this approach leads to significant resource underutilization. Specifically, prefill instances that are compute-intensive suffer from low memory utilization, while decoding instances that are memory-intensive experience low compute utilization. To address this problem, this paper proposes Adrenaline, an attention disaggregation and offloading mechanism designed to enhance resource utilization and performance in LLM serving systems. Adrenaline's key innovation lies in disaggregating part of the attention computation in the decoding phase and offloading them to prefill instances. The memory-bound nature of decoding-phase attention computation inherently enables an effective offloading strategy, yielding two complementary advantages: 1) improved memory capacity and bandwidth utilization in prefill instances, and 2) increased decoding batch sizes that enhance compute utilization in decoding instances, collectively boosting overall system performance. Adrenaline achieves these gains through three key techniques: low-latency decoding synchronization, resource-efficient prefill colocation, and load-aware offloading scheduling. Experimental results show that Adrenaline achieves 2.28x higher memory capacity and 2.07x better memory bandwidth utilization in prefill instances, up to 1.67x improvements in compute utilization for decoding instances, and 1.68x higher overall inference throughput compared to state-of-the-art systems.
Authors: Carlos Gomes, Benedikt Blumenstiel, Joao Lucas de Sousa Almeida, Pedro Henrique de Oliveira, Paolo Fraccaro, Francesc Marti Escofet, Daniela Szwarcman, Naomi Simumba, Romeo Kienzler, Bianca Zadrozny
Abstract: TerraTorch is a fine-tuning and benchmarking toolkit for Geospatial Foundation Models built on PyTorch Lightning and tailored for satellite, weather, and climate data. It integrates domain-specific data modules, pre-defined tasks, and a modular model factory that pairs any backbone with diverse decoder heads. These components allow researchers and practitioners to fine-tune supported models in a no-code fashion by simply editing a training configuration. By consolidating best practices for model development and incorporating the automated hyperparameter optimization extension Iterate, TerraTorch reduces the expertise and time required to fine-tune or benchmark models on new Earth Observation use cases. Furthermore, TerraTorch directly integrates with GEO-Bench, allowing for systematic and reproducible benchmarking of Geospatial Foundation Models. TerraTorch is open sourced under Apache 2.0, available at https://github.com/IBM/terratorch, and can be installed via pip install terratorch.
Authors: Siyuan Guo, Huiwu Liu, Xiaolong Chen, Yuming Xie, Liang Zhang, Tao Han, Hechang Chen, Yi Chang, Jun Wang
Abstract: In this work, we explore the potential of large language models (LLMs) for generating functional test scripts, which necessitates understanding the dynamically evolving code structure of the target software. To achieve this, we propose a case-based reasoning (CBR) system utilizing a 4R cycle (i.e., retrieve, reuse, revise, and retain), which maintains and leverages a case bank of test intent descriptions and corresponding test scripts to facilitate LLMs for test script generation. To improve user experience further, we introduce Re4, an optimization method for the CBR system, comprising reranking-based retrieval finetuning and reinforced reuse finetuning. Specifically, we first identify positive examples with high semantic and script similarity, providing reliable pseudo-labels for finetuning the retriever model without costly labeling. Then, we apply supervised finetuning, followed by a reinforcement learning finetuning stage, to align LLMs with our production scenarios, ensuring the faithful reuse of retrieved cases. Extensive experimental results on two product development units from Huawei Datacom demonstrate the superiority of the proposed CBR+Re4. Notably, we also show that the proposed Re4 method can help alleviate the repetitive generation issues with LLMs.
Authors: Jeffery L Painter, Gregory E Powell, Andrew Bate
Abstract: Reliable drug safety reference databases are essential for pharmacovigilance, yet existing resources like SIDER are outdated and static. We introduce PVLens, an automated system that extracts labeled safety information from FDA Structured Product Labels (SPLs) and maps terms to MedDRA. PVLens integrates automation with expert oversight through a web-based review tool. In validation against 97 drug labels, PVLens achieved an F1 score of 0.882, with high recall (0.983) and moderate precision (0.799). By offering a scalable, more accurate and continuously updated alternative to SIDER, PVLens enhances real-time pharamcovigilance with improved accuracy and contemporaneous insights.
Authors: Cristian J. Vaca-Rubio, Vaishnavi Kasuluru, Engin Zeydan, Luis Blanco, Roberto Pereira, Marius Caus, Kapal Dev
Abstract: Efficient resource management is critical for Non-Terrestrial Networks (NTNs) to provide consistent, high-quality service in remote and under-served regions. While traditional single-point prediction methods, such as Long-Short Term Memory (LSTM), have been used in terrestrial networks, they often fall short in NTNs due to the complexity of satellite dynamics, signal latency and coverage variability. Probabilistic forecasting, which quantifies the uncertainties of the predictions, is a robust alternative. In this paper, we evaluate the application of probabilistic forecasting techniques, in particular SFF, to NTN resource allocation scenarios. Our results show their effectiveness in predicting bandwidth and capacity requirements in different NTN segments of probabilistic forecasting compared to single-point prediction techniques such as LSTM. The results show the potential of black probabilistic forecasting models to provide accurate and reliable predictions and to quantify their uncertainty, making them indispensable for optimizing NTN resource allocation. At the end of the paper, we also present application scenarios and a standardization roadmap for the use of probabilistic forecasting in integrated Terrestrial Network (TN)-NTN environments.
Authors: Sadaf Khademi, Mehran Shabanpour, Reza Taleei, Anastasia Oikonomou, Arash Mohammadi
Abstract: Lung cancer remains one of the leading causes of cancer-related mortality worldwide. A crucial challenge for early diagnosis is differentiating uncertain cases with similar visual characteristics and closely annotation scores. In clinical practice, radiologists rely on quantitative, hand-crafted Radiomic features extracted from Computed Tomography (CT) images, while recent research has primarily focused on deep learning solutions. More recently, Vision-Language Models (VLMs), particularly Contrastive Language-Image Pre-Training (CLIP)-based models, have gained attention for their ability to integrate textual knowledge into lung cancer diagnosis. While CLIP-Lung models have shown promising results, we identified the following potential limitations: (a) dependence on radiologists' annotated attributes, which are inherently subjective and error-prone, (b) use of textual information only during training, limiting direct applicability at inference, and (c) Convolutional-based vision encoder with randomly initialized weights, which disregards prior knowledge. To address these limitations, we introduce AutoRad-Lung, which couples an autoregressively pre-trained VLM, with prompts generated from hand-crafted Radiomics. AutoRad-Lung uses the vision encoder of the Large-Scale Autoregressive Image Model (AIMv2), pre-trained using a multi-modal autoregressive objective. Given that lung tumors are typically small, irregularly shaped, and visually similar to healthy tissue, AutoRad-Lung offers significant advantages over its CLIP-based counterparts by capturing pixel-level differences. Additionally, we introduce conditional context optimization, which dynamically generates context-specific prompts based on input Radiomics, improving cross-modal alignment.
Authors: Gongzhu Yin, Hongli Zhang, Yuchen Yang, Yi Luo
Abstract: N-ary relational facts represent semantic correlations among more than two entities. While recent studies have developed link prediction (LP) methods to infer missing relations for knowledge graphs (KGs) containing n-ary relational facts, they are generally limited to transductive settings. Fully inductive settings, where predictions are made on previously unseen entities, remain a significant challenge. As existing methods are mainly entity embedding-based, they struggle to capture entity-independent logical rules. To fill in this gap, we propose an n-ary subgraph reasoning framework for fully inductive link prediction (ILP) on n-ary relational facts. This framework reasons over local subgraphs and has a strong inductive inference ability to capture n-ary patterns. Specifically, we introduce a novel graph structure, the n-ary semantic hypergraph, to facilitate subgraph extraction. Moreover, we develop a subgraph aggregating network, NS-HART, to effectively mine complex semantic correlations within subgraphs. Theoretically, we provide a thorough analysis from the score function optimization perspective to shed light on NS-HART's effectiveness for n-ary ILP tasks. Empirically, we conduct extensive experiments on a series of inductive benchmarks, including transfer reasoning (with and without entity features) and pairwise subgraph reasoning. The results highlight the superiority of the n-ary subgraph reasoning framework and the exceptional inductive ability of NS-HART. The source code of this paper has been made publicly available at https://github.com/yin-gz/Nary-Inductive-SubGraph.
Authors: Gabriel R. Palma, Mariusz Skocze\'n, Phil Maguire
Abstract: We investigated the use of Empirical Mode Decomposition (EMD) combined with Gaussian Mixture Models (GMM), feature engineering and machine learning algorithms to optimize trading decisions. We used five, two, and one year samples of hourly candle data for GameStop, Tesla, and XRP (Ripple) markets respectively. Applying a 15 hour rolling window for each market, we collected several features based on a linear model and other classical features to predict the next hour's movement. Subsequently, a GMM filtering approach was used to identify clusters among these markets. For each cluster, we applied the EMD algorithm to extract high, medium, low and trend components from each feature collected. A simple thresholding algorithm was applied to classify market movements based on the percentage change in each market's close price. We then evaluated the performance of various machine learning models, including Random Forests (RF) and XGBoost, in classifying market movements. A naive random selection of trading decisions was used as a benchmark, which assumed equal probabilities for each outcome, and a temporal cross-validation approach was used to test models on 40%, 30%, and 20% of the dataset. Our results indicate that transforming selected features using EMD improves performance, particularly for ensemble learning algorithms like Random Forest and XGBoost, as measured by accumulated profit. Finally, GMM filtering expanded the range of learning algorithm and data source combinations that outperformed the top percentile of the random baseline.
Authors: Shuaikai Shi, Qijun Zong
Abstract: Distributed acoustic sensing (DAS) technology represents an innovative fiber-optic-based sensing methodology that enables real-time acoustic signal monitoring through the detection of minute perturbations along optical fibers. This sensing approach offers compelling advantages, including extensive measurement ranges, exceptional spatial resolution, and an expansive dynamic measurement spectrum. The integration of machine learning (ML) paradigms presents transformative potential for DAS technology, encompassing critical domains such as data augmentation, sophisticated preprocessing techniques, and advanced acoustic event classification and recognition. By leveraging ML algorithms, DAS systems can transition from traditional data processing methodologies to more automated and intelligent analytical frameworks. The computational intelligence afforded by ML-enhanced DAS technologies facilitates unprecedented monitoring capabilities across diverse critical infrastructure sectors. Particularly noteworthy are the technology's applications in transportation infrastructure, energy management systems, and Natural disaster monitoring frameworks, where the precision of data acquisition and the reliability of intelligent decision-making mechanisms are paramount. This research critically examines the comparative performance characteristics of classical machine learning methodologies and state-of-the-art deep learning models in the context of DAS data recognition and interpretation, offering comprehensive insights into the evolving landscape of intelligent sensing technologies.
Authors: Yuhao Huang, Ao Chang, Haoran Dou, Xing Tao, Xinrui Zhou, Yan Cao, Ruobing Huang, Alejandro F Frangi, Lingyun Bao, Xin Yang, Dong Ni
Abstract: Accurate segmentation of nodules in both 2D breast ultrasound (BUS) and 3D automated breast ultrasound (ABUS) is crucial for clinical diagnosis and treatment planning. Therefore, developing an automated system for nodule segmentation can enhance user independence and expedite clinical analysis. Unlike fully-supervised learning, weakly-supervised segmentation (WSS) can streamline the laborious and intricate annotation process. However, current WSS methods face challenges in achieving precise nodule segmentation, as many of them depend on inaccurate activation maps or inefficient pseudo-mask generation algorithms. In this study, we introduce a novel multi-agent reinforcement learning-based WSS framework called Flip Learning, which relies solely on 2D/3D boxes for accurate segmentation. Specifically, multiple agents are employed to erase the target from the box to facilitate classification tag flipping, with the erased region serving as the predicted segmentation mask. The key contributions of this research are as follows: (1) Adoption of a superpixel/supervoxel-based approach to encode the standardized environment, capturing boundary priors and expediting the learning process. (2) Introduction of three meticulously designed rewards, comprising a classification score reward and two intensity distribution rewards, to steer the agents' erasing process precisely, thereby avoiding both under- and over-segmentation. (3) Implementation of a progressive curriculum learning strategy to enable agents to interact with the environment in a progressively challenging manner, thereby enhancing learning efficiency. Extensively validated on the large in-house BUS and ABUS datasets, our Flip Learning method outperforms state-of-the-art WSS methods and foundation models, and achieves comparable performance as fully-supervised learning algorithms.
Authors: Giovanni Compiani, Ilya Morozov, Stephan Seiler
Abstract: We propose a demand estimation method that leverages unstructured text and image data to infer substitution patterns. Using pre-trained deep learning models, we extract embeddings from product images and textual descriptions and incorporate them into a random coefficients logit model. This approach enables researchers to estimate demand even when they lack data on product attributes or when consumers value hard-to-quantify attributes, such as visual design or functional benefits. Using data from a choice experiment, we show that our approach outperforms standard attribute-based models in counterfactual predictions of consumers' second choices. We also apply it across 40 product categories on Amazon.com and consistently find that text and image data help identify close substitutes within each category.
Authors: A. Candito, A. Dragan, R. Holbrey, A. Ribeiro, R. Donners, C. Messiou, N. Tunariu, D. -M. Koh, M. D. Blackledge, The Institute of Cancer Research, London, United Kingdom, The Royal Marsden NHS Foundation Trust, London, United Kingdom, University Hospital Basel, Basel, Switzerland
Abstract: Background: Apparent Diffusion Coefficient (ADC) values and Total Diffusion Volume (TDV) from Whole-body diffusion-weighted MRI (WB-DWI) are recognized cancer imaging biomarkers. However, manual disease delineation for ADC and TDV measurements is unfeasible in clinical practice, demanding automation. As a first step, we propose an algorithm to generate fast and reproducible probability maps of the skeleton, adjacent internal organs (liver, spleen, urinary bladder, and kidneys), and spinal canal. Methods: We developed an automated deep-learning pipeline based on a 3D patch-based Residual U-Net architecture that localizes and delineates these anatomical structures on WB-DWI. The algorithm was trained using "soft-labels" (non-binary segmentations) derived from a computationally intensive atlas-based approach. For training and validation, we employed a multi-center WB-DWI dataset comprising 532 scans from patients with Advanced Prostate Cancer (APC) or Multiple Myeloma (MM), with testing on 45 patients. Results: Our weakly-supervised deep learning model achieved an average dice score/precision/recall of 0.66/0.6/0.73 for skeletal delineations, 0.8/0.79/0.81 for internal organs, and 0.85/0.79/0.94 for spinal canal, with surface distances consistently below 3 mm. Relative median ADC and log-transformed volume differences between automated and manual expert-defined full-body delineations were below 10% and 4%, respectively. The computational time for generating probability maps was 12x faster than the atlas-based registration algorithm (25 s vs. 5 min). An experienced radiologist rated the model's accuracy "good" or "excellent" on test datasets. Conclusion: Our model offers fast and reproducible probability maps for localizing and delineating body regions on WB-DWI, enabling ADC and TDV quantification, potentially supporting clinicians in disease staging and treatment response assessment.
Authors: Hanwen Xing, Christopher Yau
Abstract: Continual learning (CL) refers to the ability to continuously learn and accumulate new knowledge while retaining useful information from past experiences. Although numerous CL methods have been proposed in recent years, it is not straightforward to deploy them directly to real-world decision-making problems due to their computational cost and lack of uncertainty quantification. To address these issues, we propose CL-BRUNO, a probabilistic, Neural Process-based CL model that performs scalable and tractable Bayesian update and prediction. Our proposed approach uses deep-generative models to create a unified probabilistic framework capable of handling different types of CL problems such as task- and class-incremental learning, allowing users to integrate information across different CL scenarios using a single model. Our approach is able to prevent catastrophic forgetting through distributional and functional regularisation without the need of retaining any previously seen samples, making it appealing to applications where data privacy or storage capacity is of concern. Experiments show that CL-BRUNO outperforms existing methods on both natural image and biomedical data sets, confirming its effectiveness in real-world applications.
Authors: Chenxi Wang, Jizhan Fang, Xiang Chen, Bozhong Tian, Ziwen Xu, Huajun Chen, Ningyu Zhang
Abstract: Recent advancements in Large Multimodal Models (LMMs) have shown promise in Autonomous Driving Systems (ADS). However, their direct application to ADS is hindered by challenges such as misunderstanding of traffic knowledge, complex road conditions, and diverse states of vehicle. To address these challenges, we propose the use of Knowledge Editing, which enables targeted modifications to a model's behavior without the need for full retraining. Meanwhile, we introduce ADS-Edit, a multimodal knowledge editing dataset specifically designed for ADS, which includes various real-world scenarios, multiple data types, and comprehensive evaluation metrics. We conduct comprehensive experiments and derive several interesting conclusions. We hope that our work will contribute to the further advancement of knowledge editing applications in the field of autonomous driving. Code and data are available in https://github.com/zjunlp/EasyEdit.
Authors: Yan-Bo Lin, Kevin Lin, Zhengyuan Yang, Linjie Li, Jianfeng Wang, Chung-Ching Lin, Xiaofei Wang, Gedas Bertasius, Lijuan Wang
Abstract: In this paper, we introduce zero-shot audio-video editing, a novel task that requires transforming original audio-visual content to align with a specified textual prompt without additional model training. To evaluate this task, we curate a benchmark dataset, AvED-Bench, designed explicitly for zero-shot audio-video editing. AvED-Bench includes 110 videos, each with a 10-second duration, spanning 11 categories from VGGSound. It offers diverse prompts and scenarios that require precise alignment between auditory and visual elements, enabling robust evaluation. We identify limitations in existing zero-shot audio and video editing methods, particularly in synchronization and coherence between modalities, which often result in inconsistent outcomes. To address these challenges, we propose AvED, a zero-shot cross-modal delta denoising framework that leverages audio-video interactions to achieve synchronized and coherent edits. AvED demonstrates superior results on both AvED-Bench and the recent OAVE dataset to validate its generalization capabilities. Results are available at https://genjib.github.io/project_page/AVED/index.html
Authors: Hao Cheng, Kaidi Xu, Chenan Wang, Bhavya Kailkhura, Xue Lin, Ryan Goldhahn
Abstract: To tackle the susceptibility of deep neural networks to adversarial examples, the adversarial training has been proposed which provides a notion of security through an inner maximization problem presenting the first-order adversaries embedded within the outer minimization of the training loss. To generalize the adversarial robustness over different perturbation types, the adversarial training method has been augmented with the improved inner maximization presenting a union of multiple perturbations e.g., various $\ell_p$ norm-bounded perturbations. However, the improved inner maximization only enjoys limited flexibility in terms of the allowable perturbation types. In this work, through a gating mechanism, we assemble a set of expert networks, each one either adversarially trained to deal with a particular perturbation type or normally trained for boosting accuracy on clean data. The gating module assigns weights dynamically to each expert to achieve superior accuracy under various data types e.g., adversarial examples, adverse weather perturbations, and clean input. In order to deal with the obfuscated gradients issue, the training of the gating module is conducted together with fine-tuning of the last fully connected layers of expert networks through adversarial training approach. Using extensive experiments, we show that our Mixture of Robust Experts (MoRE) approach enables a flexible integration of a broad range of robust experts with superior performance.
Authors: Alexander Bastounis, Anders C Hansen, Verner Vla\v{c}i\'c
Abstract: The unprecedented success of deep learning (DL) makes it unchallenged when it comes to classification problems. However, it is well established that the current DL methodology produces universally unstable neural networks (NNs). The instability problem has caused an enormous research effort -- with a vast literature on so-called adversarial attacks -- yet there has been no solution to the problem. Our paper addresses why there has been no solution to the problem, as we prove the following mathematical paradox: any training procedure based on training neural networks for classification problems with a fixed architecture will yield neural networks that are either inaccurate or unstable (if accurate) -- despite the provable existence of both accurate and stable neural networks for the same classification problems. The key is that the stable and accurate neural networks must have variable dimensions depending on the input, in particular, variable dimensions is a necessary condition for stability. Our result points towards the paradox that accurate and stable neural networks exist, however, modern algorithms do not compute them. This yields the question: if the existence of neural networks with desirable properties can be proven, can one also find algorithms that compute them? There are cases in mathematics where provable existence implies computability, but will this be the case for neural networks? The contrary is true, as we demonstrate how neural networks can provably exist as approximate minimisers to standard optimisation problems with standard cost functions, however, no randomised algorithm can compute them with probability better than 1/2.
Authors: Mingjie Li, Rui Liu, Guangsi Shi, Mingfei Han, Changling Li, Lina Yao, Xiaojun Chang, Ling Chen
Abstract: Long-term time-series forecasting (LTSF) is fundamental to various real-world applications, where Transformer-based models have become the dominant framework due to their ability to capture long-range dependencies. However, these models often experience overfitting due to data redundancy in rolling forecasting settings, limiting their generalization ability particularly evident in longer sequences with highly similar adjacent data. In this work, we introduce CLMFormer, a novel framework that mitigates redundancy through curriculum learning and a memory-driven decoder. Specifically, we progressively introduce Bernoulli noise to the training samples, which effectively breaks the high similarity between adjacent data points. This curriculum-driven noise introduction aids the memory-driven decoder by supplying more diverse and representative training data, enhancing the decoder's ability to model seasonal tendencies and dependencies in the time-series data. To further enhance forecasting accuracy, we introduce a memory-driven decoder. This component enables the model to capture seasonal tendencies and dependencies in the time-series data and leverages temporal relationships to facilitate the forecasting process. Extensive experiments on six real-world LTSF benchmarks show that CLMFormer consistently improves Transformer-based models by up to 30%, demonstrating its effectiveness in long-horizon forecasting.
Authors: Pengfei Li, Jianyi Yang, Mohammad A. Islam, Shaolei Ren
Abstract: The growing carbon footprint of artificial intelligence (AI) has been undergoing public scrutiny. Nonetheless, the equally important water (withdrawal and consumption) footprint of AI has largely remained under the radar. For example, training the GPT-3 language model in Microsoft's state-of-the-art U.S. data centers can directly evaporate 700,000 liters of clean freshwater, but such information has been kept a secret. More critically, the global AI demand is projected to account for 4.2-6.6 billion cubic meters of water withdrawal in 2027, which is more than the total annual water withdrawal of 4-6 Denmark or half of the United Kingdom. This is concerning, as freshwater scarcity has become one of the most pressing challenges. To respond to the global water challenges, AI can, and also must, take social responsibility and lead by example by addressing its own water footprint. In this paper, we provide a principled methodology to estimate the water footprint of AI, and also discuss the unique spatial-temporal diversities of AI's runtime water efficiency. Finally, we highlight the necessity of holistically addressing water footprint along with carbon footprint to enable truly sustainable AI.
Authors: Yongyi Su, Xun Xu, Kui Jia
Abstract: Test-Time Adaptation aims to adapt source domain model to testing data at inference stage with success demonstrated in adapting to unseen corruptions. However, these attempts may fail under more challenging real-world scenarios. Existing works mainly consider real-world test-time adaptation under non-i.i.d. data stream and continual domain shift. In this work, we first complement the existing real-world TTA protocol with a globally class imbalanced testing set. We demonstrate that combining all settings together poses new challenges to existing methods. We argue the failure of state-of-the-art methods is first caused by indiscriminately adapting normalization layers to imbalanced testing data. To remedy this shortcoming, we propose a balanced batchnorm layer to swap out the regular batchnorm at inference stage. The new batchnorm layer is capable of adapting without biasing towards majority classes. We are further inspired by the success of self-training (ST) in learning from unlabeled data and adapt ST for test-time adaptation. However, ST alone is prone to over adaption which is responsible for the poor performance under continual domain shift. Hence, we propose to improve self-training under continual domain shift by regularizing model updates with an anchored loss. The final TTA model, termed as TRIBE, is built upon a tri-net architecture with balanced batchnorm layers. We evaluate TRIBE on four datasets representing real-world TTA settings. TRIBE consistently achieves the state-of-the-art performance across multiple evaluation protocols. The code is available at https://github.com/Gorilla-Lab-SCUT/TRIBE.
Authors: Mahyar Fazlyab, Taha Entesari, Aniket Roy, Rama Chellappa
Abstract: To improve the robustness of deep classifiers against adversarial perturbations, many approaches have been proposed, such as designing new architectures with better robustness properties (e.g., Lipschitz-capped networks), or modifying the training process itself (e.g., min-max optimization, constrained learning, or regularization). These approaches, however, might not be effective at increasing the margin in the input (feature) space. As a result, there has been an increasing interest in developing training procedures that can directly manipulate the decision boundary in the input space. In this paper, we build upon recent developments in this category by developing a robust training algorithm whose objective is to increase the margin in the output (logit) space while regularizing the Lipschitz constant of the model along vulnerable directions. We show that these two objectives can directly promote larger margins in the input space. To this end, we develop a scalable method for calculating guaranteed differentiable upper bounds on the Lipschitz constant of neural networks accurately and efficiently. The relative accuracy of the bounds prevents excessive regularization and allows for more direct manipulation of the decision boundary. Furthermore, our Lipschitz bounding algorithm exploits the monotonicity and Lipschitz continuity of the activation layers, and the resulting bounds can be used to design new layers with controllable bounds on their Lipschitz constant. Experiments on the MNIST, CIFAR-10, and Tiny-ImageNet data sets verify that our proposed algorithm obtains competitively improved results compared to the state-of-the-art.
Authors: Wasu Top Piriyakulkij, Yingheng Wang, Volodymyr Kuleshov
Abstract: We propose denoising diffusion variational inference (DDVI), a black-box variational inference algorithm for latent variable models which relies on diffusion models as flexible approximate posteriors. Specifically, our method introduces an expressive class of diffusion-based variational posteriors that perform iterative refinement in latent space; we train these posteriors with a novel regularized evidence lower bound (ELBO) on the marginal likelihood inspired by the wake-sleep algorithm. Our method is easy to implement (it fits a regularized extension of the ELBO), is compatible with black-box variational inference, and outperforms alternative classes of approximate posteriors based on normalizing flows or adversarial networks. We find that DDVI improves inference and learning in deep latent variable models across common benchmarks as well as on a motivating task in biology -- inferring latent ancestry from human genomes -- where it outperforms strong baselines on the Thousand Genomes dataset.
Authors: Mahrokh Ghoddousi Boroujeni, Andreas Krause, Giancarlo Ferrari Trecate
Abstract: Federated Learning (FL) aims to infer a shared model from private and decentralized data stored by multiple clients. Personalized FL (PFL) enhances the model's fit for each client by adapting the global model to the clients. A significant level of personalization is required for highly heterogeneous clients but can be challenging to achieve, especially when clients' datasets are small. To address this issue, we introduce the PAC-PFL framework for PFL of probabilistic models. PAC-PFL infers a shared hyper-posterior and treats each client's posterior inference as the personalization step. Unlike previous PFL algorithms, PAC-PFL does not regularize all personalized models towards a single shared model, thereby greatly enhancing its personalization flexibility. By establishing and minimizing a PAC-Bayesian generalization bound on the average true loss of clients, PAC-PFL effectively mitigates overfitting even in data-poor scenarios. Additionally, PAC-PFL provides generalization bounds for new clients joining later. PAC-PFL achieves accurate and well-calibrated predictions, as supported by our experiments.
Authors: Francesco Della Santa, Sandra Pieraccini
Abstract: In this paper, we present a novel approach for detecting the discontinuity interfaces of a discontinuous function. This approach leverages Graph-Instructed Neural Networks (GINNs) and sparse grids to address discontinuity detection also in domains of dimension larger than 3. GINNs, trained to identify troubled points on sparse grids, exploit graph structures built on the grids to achieve efficient and accurate discontinuity detection performances. We also introduce a recursive algorithm for general sparse grid-based detectors, characterized by convergence properties and easy applicability. Numerical experiments on functions with dimensions n = 2 and n = 4 demonstrate the efficiency and robust generalization properties of GINNs in detecting discontinuity interfaces. Notably, the trained GINNs offer portability and versatility, allowing integration into various algorithms and sharing among users.
Authors: Zexi Li, Lingzhi Gao, Chao Wu
Abstract: Generative artificial intelligence (GenAI) has made significant progress in understanding world knowledge and generating content from human languages across various modalities, like text-to-text large language models, text-to-image stable diffusion, and text-to-video Sora. While in this paper, we investigate the capability of GenAI for text-to-model generation, to see whether GenAI can comprehend hyper-level knowledge embedded within AI itself parameters. Specifically, we study a practical scenario termed train-once-for-all personalization, aiming to generate personalized models for diverse end-users and tasks using text prompts. Inspired by the recent emergence of neural network diffusion, we present Tina, a text-conditioned neural network diffusion for train-once-for-all personalization. Tina leverages a diffusion transformer model conditioned on task descriptions embedded using a CLIP model. Despite the astronomical number of potential personalized tasks (e.g., $1.73\times10^{13}$), by our design, Tina demonstrates remarkable in-distribution and out-of-distribution generalization even trained on small datasets ($\sim 1000$). We further verify whether and how \Tina understands world knowledge by analyzing its capabilities under zero-shot/few-shot image prompts, different numbers of personalized classes, prompts of natural language descriptions, and predicting unseen entities.
Authors: Mohammed Baharoon, Jonathan Klein, Dominik L. Michels
Abstract: Vision-language contrastive learning frameworks like CLIP enable learning representations from natural language supervision, and provide strong zero-shot classification capabilities. However, due to the nature of the supervisory signal in these paradigms, they lack the ability to learn localized features, leading to degraded performance on dense prediction tasks like segmentation and detection. On the other hand, self-supervised learning methods have shown the ability to learn granular representations, complementing the high-level features in vision-language training. In this work, we present Harmony, a framework that combines vision-language training with discriminative and generative self-supervision to learn visual features that can be generalized across different vision downstream tasks. Our framework is specifically designed to work on web-scraped data by not relying on negative examples and addressing the one-to-one correspondence issue using soft CLIP targets generated by an EMA model. We comprehensively evaluate Harmony across various vision downstream tasks and find that it significantly outperforms the baseline CLIP and the previously leading joint self and weakly-supervised methods, MaskCLIP and SLIP. Specifically, when comparing against these methods, Harmony shows superior performance in fine-tuning and zero-shot classification on ImageNet-1k, semantic segmentation on ADE20K, and both object detection and instance segmentation on MS-COCO, when pre-training a ViT-B on CC3M. We also show that Harmony outperforms other self-supervised learning methods like iBOT and MAE across all tasks evaluated. Our code is publicly at https://github.com/MohammedSB/Harmony}{https://github.com/MohammedSB/Harmony available.
URLs: https://github.com/MohammedSB/Harmony, https://github.com/MohammedSB/Harmony
Authors: Ekaterina Kukleva, Vitaly Vanchurin
Abstract: In artificial neural networks, the activation dynamics of non-trainable variables is strongly coupled to the learning dynamics of trainable variables. During the activation pass, the boundary neurons (e.g., input neurons) are mapped to the bulk neurons (e.g., hidden neurons), and during the learning pass, both bulk and boundary neurons are mapped to changes in trainable variables (e.g., weights and biases). For example, in feed-forward neural networks, forward propagation is the activation pass and backward propagation is the learning pass. We show that a composition of the two maps establishes a duality map between a subspace of non-trainable boundary variables (e.g., dataset) and a tangent subspace of trainable variables (i.e., learning). In general, the dataset-learning duality is a complex non-linear map between high-dimensional spaces. We use duality to study the emergence of criticality, or the power-law distribution of fluctuations of the trainable variables, using a toy model at learning equilibrium. In particular, we show that criticality can emerge in the learning system even from the dataset in a non-critical state, and that the power-law distribution can be modified by changing either the activation function or the loss function.
Authors: Hunter Nisonoff, Junhao Xiong, Stephan Allenspach, Jennifer Listgarten
Abstract: Generative models on discrete state-spaces have a wide range of potential applications, particularly in the domain of natural sciences. In continuous state-spaces, controllable and flexible generation of samples with desired properties has been realized using guidance on diffusion and flow models. However, these guidance approaches are not readily amenable to discrete state-space models. Consequently, we introduce a general and principled method for applying guidance on such models. Our method depends on leveraging continuous-time Markov processes on discrete state-spaces, which unlocks computational tractability for sampling from a desired guided distribution. We demonstrate the utility of our approach, Discrete Guidance, on a range of applications including guided generation of small-molecules, DNA sequences and protein sequences.
Authors: Lev Telyatnikov, Guillermo Bernardez, Marco Montagna, Mustafa Hajij, Martin Carrasco, Pavlo Vasylenko, Mathilde Papillon, Ghada Zamzmi, Michael T. Schaub, Jonas Verhellen, Pavel Snopov, Bertran Miquel-Oliver, Manel Gil-Sorribes, Alexis Molina, Victor Guallar, Theodore Long, Julian Suk, Patryk Rygiel, Alexander Nikitin, Giordan Escalona, Michael Banf, Dominik Filipiak, Max Schattauer, Liliya Imasheva, Alvaro Martinez, Halley Fritze, Marissa Masden, Valentina S\'anchez, Manuel Lecha, Andrea Cavallo, Claudio Battiloro, Matt Piekenbrock, Mauricio Tec, George Dasoulas, Nina Miolane, Simone Scardapane, Theodore Papamarkou
Abstract: This work introduces TopoBench, an open-source library designed to standardize benchmarking and accelerate research in topological deep learning (TDL). TopoBench decomposes TDL into a sequence of independent modules for data generation, loading, transforming and processing, as well as model training, optimization and evaluation. This modular organization provides flexibility for modifications and facilitates the adaptation and optimization of various TDL pipelines. A key feature of TopoBench is its support for transformations and lifting across topological domains. Mapping the topology and features of a graph to higher-order topological domains, such as simplicial and cell complexes, enables richer data representations and more fine-grained analyses. The applicability of TopoBench is demonstrated by benchmarking several TDL architectures across diverse tasks and datasets.
Authors: Amey Agrawal, Haoran Qiu, Junda Chen, \'I\~nigo Goiri, Ramachandran Ramjee, Chaojie Zhang, Alexey Tumanov, Esha Choukse
Abstract: As large language models (LLMs) handle increasingly longer contexts, serving inference requests for context lengths in the range of millions of tokens presents unique challenges. While existing techniques are effective for training, they fail to address the unique challenges of inference, such as varying prefill and decode phases and their associated latency constraints -- like Time to First Token (TTFT) and Time per Output Token (TPOT). Furthermore, no long-context inference solutions address head-of-line blocking today. We present Medha, a system for efficient long-context LLM inference that introduces three key innovations: adaptive chunking with slack-aware scheduling to prevent head-ofline blocking, Sequence Pipeline Parallelism (SPP) to reduce TTFT, and KV Cache Parallelism (KVP) to minimize TPOT. By combining these into a novel 3D parallelism serving engine, Medha achieves unprecedented scale -- supporting contexts up to 10M tokens with production-grade latency. Our evaluation shows Medha reduces median latency by up to 30x compared to state-of-the-art systems when serving a mix of short and long requests, while improving throughput by upwards of 5x. This enables, for the first time, efficient long-context LLM inference at scale without compromising on shorter request latencies or system efficiency.
Authors: Chaoqi Wang, Zhuokai Zhao, Chen Zhu, Karthik Abinav Sankararaman, Michal Valko, Xuefei Cao, Zhaorun Chen, Madian Khabsa, Yuxin Chen, Hao Ma, Sinong Wang
Abstract: Recent advancements in generative models, particularly large language models (LLMs) and diffusion models, have been driven by extensive pretraining on large datasets followed by post-training. However, current post-training methods such as reinforcement learning from human feedback (RLHF) and direct alignment from preference methods (DAP) primarily utilize single-sample comparisons. These approaches often fail to capture critical characteristics such as generative diversity and bias, which are more accurately assessed through multiple samples. To address these limitations, we introduce a novel approach that extends post-training to include multi-sample comparisons. To achieve this, we propose Multi-sample Direct Preference Optimization (mDPO) and Multi-sample Identity Preference Optimization (mIPO). These methods improve traditional DAP methods by focusing on group-wise characteristics. Empirically, we demonstrate that multi-sample comparison is more effective in optimizing collective characteristics~(e.g., diversity and bias) for generative models than single-sample comparison. Additionally, our findings suggest that multi-sample comparisons provide a more robust optimization framework, particularly for dataset with label noise.
Authors: Guangming Huang, Yunfei Long, Cunjin Luo
Abstract: Supervised contrastive learning has achieved remarkable success by leveraging label information; however, determining positive samples in multi-label scenarios remains a critical challenge. In multi-label supervised contrastive learning (MSCL), relations among multi-label samples are not yet fully defined, leading to ambiguity in identifying positive samples and formulating contrastive loss functions to construct the representation space. To address these challenges, we: (i) first define five distinct multi-label relations in MSCL to systematically identify positive samples, (ii) introduce a novel Similarity-Dissimilarity Loss that dynamically re-weights samples through computing the similarity and dissimilarity factors between positive samples and given anchors based on multi-label relations, and (iii) further provide theoretical grounded proof for our method through rigorous mathematical analysis that supports the formulation and effectiveness of the proposed loss function. We conduct the experiments across both image and text modalities, and extend the evaluation to medical domain. The results demonstrate that our method consistently outperforms baselines in a comprehensive evaluation, confirming its effectiveness and robustness. Code is available at: https://github.com/guangminghuang/similarity-dissimilarity-loss.
URLs: https://github.com/guangminghuang/similarity-dissimilarity-loss.
Authors: Mridul Gupta, Samyak Jain, Vansh Ramani, Hariprasad Kodamana, Sayan Ranu
Abstract: Graph condensation has emerged as a promising avenue to enable scalable training of GNNs by compressing the training dataset while preserving essential graph characteristics. Our study uncovers significant shortcomings in current graph condensation techniques. First, the majority of the algorithms paradoxically require training on the full dataset to perform condensation. Second, due to their gradient-emulating approach, these methods require fresh condensation for any change in hyperparameters or GNN architecture, limiting their flexibility and reusability. Finally, they fail to achieve substantial size reduction due to synthesizing fully-connected, edge-weighted graphs. To address these challenges, we present Bonsai, a novel graph condensation method empowered by the observation that \textit{computation trees} form the fundamental processing units of message-passing GNNs. Bonsai condenses datasets by encoding a careful selection of \textit{exemplar} trees that maximize the representation of all computation trees in the training set. This unique approach imparts Bonsai as the first linear-time, model-agnostic graph condensation algorithm for node classification that outperforms existing baselines across $7$ real-world datasets on accuracy, while being $22$ times faster on average. Bonsai is grounded in rigorous mathematical guarantees on the adopted approximation strategies making it robust to GNN architectures, datasets, and parameters.
Authors: Amin Abyaneh, Mahrokh G. Boroujeni, Hsiu-Chin Lin, Giancarlo Ferrari-Trecate
Abstract: Imitation learning is a data-driven approach to learning policies from expert behavior, but it is prone to unreliable outcomes in out-of-sample (OOS) regions. While previous research relying on stable dynamical systems guarantees convergence to a desired state, it often overlooks transient behavior. We propose a framework for learning policies modeled by contractive dynamical systems, ensuring that all policy rollouts converge regardless of perturbations, and in turn, enable efficient OOS recovery. By leveraging recurrent equilibrium networks and coupling layers, the policy structure guarantees contractivity for any parameter choice, which facilitates unconstrained optimization. We also provide theoretical upper bounds for worst-case and expected loss to rigorously establish the reliability of our method in deployment. Empirically, we demonstrate substantial OOS performance improvements for simulated robotic manipulation and navigation tasks.
Authors: Anh Tu Ngo, Chuan Song Heng, Nandish Chattopadhyay, Anupam Chattopadhyay
Abstract: Deep Neural Networks (DNNs) have gained considerable traction in recent years due to the unparalleled results they gathered. However, the cost behind training such sophisticated models is resource intensive, resulting in many to consider DNNs to be intellectual property (IP) to model owners. In this era of cloud computing, high-performance DNNs are often deployed all over the internet so that people can access them publicly. As such, DNN watermarking schemes, especially backdoor-based watermarks, have been actively developed in recent years to preserve proprietary rights. Nonetheless, there lies much uncertainty on the robustness of existing backdoor watermark schemes, towards both adversarial attacks and unintended means such as fine-tuning neural network models. One reason for this is that no complete guarantee of robustness can be assured in the context of backdoor-based watermark. In this paper, we extensively evaluate the persistence of recent backdoor-based watermarks within neural networks in the scenario of fine-tuning, we propose/develop a novel data-driven idea to restore watermark after fine-tuning without exposing the trigger set. Our empirical results show that by solely introducing training data after fine-tuning, the watermark can be restored if model parameters do not shift dramatically during fine-tuning. Depending on the types of trigger samples used, trigger accuracy can be reinstated to up to 100%. Our study further explores how the restoration process works using loss landscape visualization, as well as the idea of introducing training data in fine-tuning stage to alleviate watermark vanishing.
Authors: Yunbo Hou, Haoran Ye, Shuwen Yang, Yingxue Zhang, Siyuan Xu, Guojie Song
Abstract: Global placement, a critical step in designing the physical layout of computer chips, is essential to optimize chip performance. Prior global placement methods optimize each circuit design individually from scratch. Their neglect of transferable knowledge limits solution efficiency and chip performance as circuit complexity drastically increases. This study presents TransPlace, a global placement framework that learns to place millions of mixed-size cells in continuous space. TransPlace introduces i) Netlist Graph to efficiently model netlist topology, ii) Cell-flow and relative position encoding to learn SE(2)-invariant representation, iii) a tailored graph neural network architecture for informed parameterization of placement knowledge, and iv) a two-stage strategy for coarse-to-fine placement. Compared to state-of-the-art placement methods, TransPlace-trained on a few high-quality placements-can place unseen circuits with 1.2x speedup while reducing congestion by 30%, timing by 9%, and wirelength by 5%.
Authors: Hao Yuan Bai, Xue Liu
Abstract: Spatiotemporal data is ubiquitous, and forecasting it has important applications in many domains. However, its complex cross-component dependencies and non-linear temporal dynamics can be challenging for traditional techniques. Existing methods address this by learning the two dimensions separately. Here, we introduce Temporal Graphormer (T-Graphormer), a Transformer-based approach capable of modelling spatiotemporal correlations simultaneously. By adding temporal encodings in the Graphormer architecture, each node attends to all other tokens within the graph sequence, enabling the model to learn rich spacetime patterns with minimal predefined inductive biases. We show the effectiveness of T-Graphormer on real-world traffic prediction benchmark datasets. Compared to state-of-the-art methods, T-Graphormer reduces root mean squared error (RMSE) and mean absolute percentage error (MAPE) by up to 20% and 10%.
Authors: Jacob Fein-Ashley
Abstract: We present Linear Diffusion Networks (LDNs), a novel architecture that reinterprets sequential data processing as a unified diffusion process. Our model integrates adaptive diffusion modules with localized nonlinear updates and a diffusion-inspired attention mechanism. This design enables efficient global information propagation while preserving fine-grained temporal details. LDN overcomes the limitations of conventional recurrent and transformer models by allowing full parallelization across time steps and supporting robust multi-scale temporal representations. Experiments on benchmark sequence modeling tasks demonstrate that LDN delivers competitive performance across ImageNet and LRA tasks.
Authors: Takashi Morita
Abstract: Human and animal memory for sequentially presented items is well-documented to be more accurate for those at the beginning and end of the sequence, phenomena known as the primacy and recency effects, respectively. By contrast, artificial neural network (ANN) models are typically designed with a memory that decays monotonically over time. Accordingly, ANNs are expected to show the recency effect but not the primacy effect. Contrary to this theoretical expectation, however, the present study reveals a counterintuitive finding: a recently developed ANN architecture, called structured state-space models, exhibits the primacy effect when trained and evaluated on a synthetic task that mirrors psychological memory experiments. Given that this model was originally designed for recovering neuronal activity patterns observed in biological brains, this result provides a novel perspective on the psychological primacy effect while also posing a non-trivial puzzle for the current theories in machine learning.
Authors: Mohit Pandey, Gopeshh Subbaraj, Artem Cherkasov, Martin Ester, Emmanuel Bengio
Abstract: Generative Flow Networks (GFlowNets) have recently emerged as a suitable framework for generating diverse and high-quality molecular structures by learning from rewards treated as unnormalized distributions. Previous works in this framework often restrict exploration by using predefined molecular fragments as building blocks, limiting the chemical space that can be accessed. In this work, we introduce Atomic GFlowNets (A-GFNs), a foundational generative model leveraging individual atoms as building blocks to explore drug-like chemical space more comprehensively. We propose an unsupervised pre-training approach using drug-like molecule datasets, which teaches A-GFNs about inexpensive yet informative molecular descriptors such as drug-likeliness, topological polar surface area, and synthetic accessibility scores. These properties serve as proxy rewards, guiding A-GFNs towards regions of chemical space that exhibit desirable pharmacological properties. We further implement a goal-conditioned finetuning process, which adapts A-GFNs to optimize for specific target properties. In this work, we pretrain A-GFN on a subset of ZINC dataset, and by employing robust evaluation metrics we show the effectiveness of our approach when compared to other relevant baseline methods for a wide range of drug design tasks.
Authors: Yuncheng Guo, Xiaodong Gu
Abstract: Large-scale pre-trained Vision-Language Models (VLMs) have become essential for transfer learning across diverse tasks. However, adapting these models with limited few-shot data often leads to overfitting, diminishing their performance on new tasks. To tackle this issue, we propose a novel Multi-Modal Representation Learning (MMRL) framework that introduces a shared, learnable, and modality-agnostic representation space. MMRL projects the space tokens to text and image representation tokens, facilitating more effective multi-modal interactions. Unlike previous approaches that solely optimize class token features, MMRL integrates representation tokens at higher layers of the encoders--where dataset-specific features are more prominent--while preserving generalized knowledge in the lower layers. During training, both representation and class features are optimized, with trainable projection layer applied to the representation tokens, whereas the class token projection layer remains frozen to retain pre-trained knowledge. Furthermore, a regularization term is introduced to align the class features and text features with the zero-shot features from the frozen VLM, thereby safeguarding the model's generalization capacity. For inference, a decoupling strategy is employed, wherein both representation and class features are utilized for base classes, while only the class features, which retain more generalized knowledge, are used for new tasks. Extensive experiments across 15 datasets demonstrate that MMRL outperforms state-of-the-art methods, achieving a balanced trade-off between task-specific adaptation and generalization. Code is available at https://github.com/yunncheng/MMRL.
Authors: Moritz A. Zanger, Pascal R. Van der Vaart, Wendelin B\"ohmer, Matthijs T. J. Spaan
Abstract: Uncertainty quantification is a critical aspect of reinforcement learning and deep learning, with numerous applications ranging from efficient exploration and stable offline reinforcement learning to outlier detection in medical diagnostics. The scale of modern neural networks, however, complicates the use of many theoretically well-motivated approaches such as full Bayesian inference. Approximate methods like deep ensembles can provide reliable uncertainty estimates but still remain computationally expensive. In this work, we propose contextual similarity distillation, a novel approach that explicitly estimates the variance of an ensemble of deep neural networks with a single model, without ever learning or evaluating such an ensemble in the first place. Our method builds on the predictable learning dynamics of wide neural networks, governed by the neural tangent kernel, to derive an efficient approximation of the predictive variance of an infinite ensemble. Specifically, we reinterpret the computation of ensemble variance as a supervised regression problem with kernel similarities as regression targets. The resulting model can estimate predictive variance at inference time with a single forward pass, and can make use of unlabeled target-domain data or data augmentations to refine its uncertainty estimates. We empirically validate our method across a variety of out-of-distribution detection benchmarks and sparse-reward reinforcement learning environments. We find that our single-model method performs competitively and sometimes superior to ensemble-based baselines and serves as a reliable signal for efficient exploration. These results, we believe, position contextual similarity distillation as a principled and scalable alternative for uncertainty quantification in reinforcement learning and general deep learning.
Authors: Juhyeong Kim
Abstract: I propose Semi-Decision-Focused Learning, a practical adaptation of Decision-Focused Learning for portfolio optimization. Rather than directly optimizing complex financial metrics, I employ simple target portfolios (Max-Sortino or One-Hot) and train models with a convex, cross-entropy loss. I further incorporate Deep Ensemble methods to reduce variance and stabilize performance. Experiments on two universes (one upward-trending and another range-bound) show consistent outperformance over baseline portfolios, demonstrating the effectiveness and robustness of my approach. Code is available at https://github.com/sDFLwDE/sDFLwDE
Authors: Matt Franchi, Nikhil Garg, Wendy Ju, Emma Pierson
Abstract: Street scene datasets, collected from Street View or dashboard cameras, offer a promising means of detecting urban objects and incidents like street flooding. However, a major challenge in using these datasets is their lack of reliable labels: there are myriad types of incidents, many types occur rarely, and ground-truth measures of where incidents occur are lacking. Here, we propose BayFlood, a two-stage approach which circumvents this difficulty. First, we perform zero-shot classification of where incidents occur using a pretrained vision-language model (VLM). Second, we fit a spatial Bayesian model on the VLM classifications. The zero-shot approach avoids the need to annotate large training sets, and the Bayesian model provides frequent desiderata in urban settings - principled measures of uncertainty, smoothing across locations, and incorporation of external data like stormwater accumulation zones. We comprehensively validate this two-stage approach, showing that VLMs provide strong zero-shot signal for floods across multiple cities and time periods, the Bayesian model improves out-of-sample prediction relative to baseline methods, and our inferred flood risk correlates with known external predictors of risk. Having validated our approach, we show it can be used to improve urban flood detection: our analysis reveals 113,738 people who are at high risk of flooding overlooked by current methods, identifies demographic biases in existing methods, and suggests locations for new flood sensors. More broadly, our results showcase how Bayesian modeling of zero-shot LM annotations represents a promising paradigm because it avoids the need to collect large labeled datasets and leverages the power of foundation models while providing the expressiveness and uncertainty quantification of Bayesian models.
Authors: Jiazhu Dai, Yubing Lu
Abstract: Graph neural networks (GNNs) are widely used for graph-structured data but are vulnerable to membership inference attacks (MIAs) in graph classification tasks, which determine if a graph was part of the training dataset, potentially causing data leakage. Existing MIAs rely on prediction probability vectors, but they become ineffective when only prediction labels are available. We propose a Graph-level Label-Only Membership Inference Attack (GLO-MIA), which is based on the intuition that the target model's predictions on training data are more stable than those on testing data. GLO-MIA generates a set of perturbed graphs for target graph by adding perturbations to its effective features and queries the target model with the perturbed graphs to get their prediction labels, which are then used to calculate robustness score of the target graph. Finally, by comparing the robustness score with a predefined threshold, the membership of the target graph can be inferred correctly with high probability. Our evaluation on three datasets and four GNN models shows that GLO-MIA achieves an attack accuracy of up to 0.825, outperforming baseline work by 8.5% and closely matching the performance of probability-based MIAs, even with only prediction labels.
Authors: Stefano De Carli, Nicola Licini, Davide Previtali, Fabio Previdi, Antonio Ferramosca
Abstract: Type 1 Diabetes (T1D) management is a complex task due to many variability factors. Artificial Pancreas (AP) systems have alleviated patient burden by automating insulin delivery through advanced control algorithms. However, the effectiveness of these systems depends on accurate modeling of glucose-insulin dynamics, which traditional mathematical models often fail to capture due to their inability to adapt to patient-specific variations. This study introduces a Biological-Informed Recurrent Neural Network (BIRNN) framework to address these limitations. The BIRNN leverages a Gated Recurrent Units (GRU) architecture augmented with physics-informed loss functions that embed physiological constraints, ensuring a balance between predictive accuracy and consistency with biological principles. The framework is validated using the commercial UVA/Padova simulator, outperforming traditional linear models in glucose prediction accuracy and reconstruction of unmeasured states, even under circadian variations in insulin sensitivity. The results demonstrate the potential of BIRNN for personalized glucose regulation and future adaptive control strategies in AP systems.
Authors: Xin Cai
Abstract: In this article, we primarily examine a variety of RL-based and RL-free methods designed to address Reinforcement Learning from Human Feedback (RLHF) and Large Reasoning Models (LRMs). We begin with a concise overview of the typical steps involved in RLHF and LRMs. Next, we reinterpret several RL-based and RL-free algorithms through the perspective of neural structured bandit prediction, providing a clear conceptual framework that uncovers a deeper connection between these seemingly distinct approaches. Following this, we briefly review some core principles of reinforcement learning, drawing attention to an often-overlooked aspect in existing RLHF studies. This leads to a detailed derivation of the standard RLHF objective within a full RL context, demonstrating its equivalence to neural structured bandit prediction. Finally, by reinvestigating the principles behind Proximal Policy Optimization (PPO), we pinpoint areas needing adjustment, which culminates in the introduction of the Generalized Reinforce Optimization (GRO) framework, seamlessly integrating RL-based and RL-free methods in RLHF. We look forward to the community's efforts to empirically validate GRO and invite constructive feedback.
Authors: Eshed Gal, Moshe Eliasof, Carola-Bibiane Sch\"onlieb, Eldad Haber, Eran Treister
Abstract: Graph Neural Networks (GNNs) have emerged as a powerful tool for learning and inferring from graph-structured data, and are widely used in a variety of applications, often considering large amounts of data and large graphs. However, training on such data requires large memory and extensive computations. In this paper, we introduce a novel framework for efficient multiscale training of GNNs, designed to integrate information across multiscale representations of a graph. Our approach leverages a hierarchical graph representation, taking advantage of coarse graph scales in the training process, where each coarse scale graph has fewer nodes and edges. Based on this approach, we propose a suite of GNN training methods: such as coarse-to-fine, sub-to-full, and multiscale gradient computation. We demonstrate the effectiveness of our methods on various datasets and learning tasks.
Authors: Hao Cheng, Kaidi Xu, Sijia Liu, Pin-Yu Chen, Pu Zhao, Xue Lin
Abstract: Although deep neural networks (DNNs) have achieved a great success in various computer vision tasks, it is recently found that they are vulnerable to adversarial attacks. In this paper, we focus on the so-called \textit{backdoor attack}, which injects a backdoor trigger to a small portion of training data (also known as data poisoning) such that the trained DNN induces misclassification while facing examples with this trigger. To be specific, we carefully study the effect of both real and synthetic backdoor attacks on the internal response of vanilla and backdoored DNNs through the lens of Gard-CAM. Moreover, we show that the backdoor attack induces a significant bias in neuron activation in terms of the $\ell_\infty$ norm of an activation map compared to its $\ell_1$ and $\ell_2$ norm. Spurred by our results, we propose the \textit{$\ell_\infty$-based neuron pruning} to remove the backdoor from the backdoored DNN. Experiments show that our method could effectively decrease the attack success rate, and also hold a high classification accuracy for clean images.
Authors: Jie Wang, Rui Gao, Yao Xie
Abstract: We study distributionally robust optimization with Sinkhorn distance -- a variant of Wasserstein distance based on entropic regularization. We derive a convex programming dual reformulation for general nominal distributions, transport costs, and loss functions. To solve the dual reformulation, we develop a stochastic mirror descent algorithm with biased subgradient estimators and derive its computational complexity guarantees. Finally, we provide numerical examples using synthetic and real data to demonstrate its superior performance.
Authors: Bin Wang, Wenbin Pei, Bing Xue, Mengjie Zhang
Abstract: Deep convolutional neural networks have proven their effectiveness, and have been acknowledged as the most dominant method for image classification. However, a severe drawback of deep convolutional neural networks is poor explainability. Unfortunately, in many real-world applications, users need to understand the rationale behind the predictions of deep convolutional neural networks when determining whether they should trust the predictions or not. To resolve this issue, a novel genetic algorithm-based method is proposed for the first time to automatically evolve local explanations that can assist users to assess the rationality of the predictions. Furthermore, the proposed method is model-agnostic, i.e., it can be utilised to explain any deep convolutional neural network models. In the experiments, ResNet is used as an example model to be explained, and the ImageNet dataset is selected as the benchmark dataset. DenseNet and MobileNet are further explained to demonstrate the model-agnostic characteristic of the proposed method. The evolved local explanations on four images, randomly selected from ImageNet, are presented, which show that the evolved local explanations are straightforward to be recognised by humans. Moreover, the evolved explanations can explain the predictions of deep convolutional neural networks on all four images very well by successfully capturing meaningful interpretable features of the sample images. Further analysis based on the 30 runs of the experiments exhibits that the evolved local explanations can also improve the probabilities/confidences of the deep convolutional neural network models in making the predictions. The proposed method can obtain local explanations within one minute, which is more than ten times faster than LIME (the state-of-the-art method).
Authors: Renhao Huang, Hao Xue, Maurice Pagnucco, Flora Salim, Yang Song
Abstract: Vision-based trajectory prediction is an important task that supports safe and intelligent behaviours in autonomous systems. Many advanced approaches have been proposed over the years with improved spatial and temporal feature extraction. However, human behaviour is naturally diverse and uncertain. Given the past trajectory and surrounding environment information, an agent can have multiple plausible trajectories in the future. To tackle this problem, an essential task named multi-future trajectory prediction (MTP) has recently been studied. This task aims to generate a diverse, acceptable and explainable distribution of future predictions for each agent. In this paper, we present the first survey for MTP with our unique taxonomies and a comprehensive analysis of frameworks, datasets and evaluation metrics. We also compare models on existing MTP datasets and conduct experiments on the ForkingPath dataset. Finally, we discuss multiple future directions that can help researchers develop novel multi-future trajectory prediction systems and other diverse learning tasks similar to MTP.
Authors: Payam Abdisarabshali, Nicholas Accurso, Filippo Malandra, Weifeng Su, Seyyedali Hosseinalipour
Abstract: Federated learning (FL) is a popular distributed machine learning (ML) technique in Internet of Things (IoT) networks, where resource-constrained devices collaboratively train ML models while preserving data privacy. However, implementation of FL over 5G-and-beyond wireless networks faces key challenges caused by (i) dynamics of the wireless network conditions and (ii) the coexistence of multiple FL-services in the system. In this paper, we unveil two key phenomena that arise from these challenges: over/under-provisioning of resources and perspective-driven load balancing, both of which significantly impact FL performance in IoT environments. We take the first steps towards addressing these phenomena by proposing a novel distributed ML architecture called elastic FL (EFL). EFL unleashes the full potential of Open RAN (O-RAN) systems and introduces an elastic resource provisioning methodology to execute FL-services. It further constitutes a multi-time-scale FL management system that introduces three dedicated network control functionalities tailored for FL-services, including (i) non-real-time (non-RT) system descriptor, which trains ML-based applications to predict both system and FL-related dynamics and parameters; (ii) near-RT FL controller, which handles O-RAN slicing and mobility management for the seamless execution of FL-services; (iii) FL MAC scheduler, which conducts real-time resource allocation to the end clients of various FL-services. We finally prototype EFL to demonstrate its potential in improving the performance of FL-services.
Authors: Abdullah Makkeh, Marcel Graetz, Andreas C. Schneider, David A. Ehrlich, Viola Priesemann, Michael Wibral
Abstract: Despite the impressive performance of biological and artificial networks, an intuitive understanding of how their local learning dynamics contribute to network-level task solutions remains a challenge to this date. Efforts to bring learning to a more local scale indeed lead to valuable insights, however, a general constructive approach to describe local learning goals that is both interpretable and adaptable across diverse tasks is still missing. We have previously formulated a local information processing goal that is highly adaptable and interpretable for a model neuron with compartmental structure. Building on recent advances in Partial Information Decomposition (PID), we here derive a corresponding parametric local learning rule, which allows us to introduce 'infomorphic' neural networks. We demonstrate the versatility of these networks to perform tasks from supervised, unsupervised and memory learning. By leveraging the interpretable nature of the PID framework, infomorphic networks represent a valuable tool to advance our understanding of the intricate structure of local learning.
Authors: Joschka Birk, Erik Buhmann, Cedric Ewen, Gregor Kasieczka, David Shih
Abstract: We introduce the first generative model trained on the JetClass dataset. Our model generates jets at the constituent level, and it is a permutation-equivariant continuous normalizing flow (CNF) trained with the flow matching technique. It is conditioned on the jet type, so that a single model can be used to generate the ten different jet types of JetClass. For the first time, we also introduce a generative model that goes beyond the kinematic features of jet constituents. The JetClass dataset includes more features, such as particle-ID and track impact parameter, and we demonstrate that our CNF can accurately model all of these additional features as well. Our generative model for JetClass expands on the versatility of existing jet generation techniques, enhancing their potential utility in high-energy physics research, and offering a more comprehensive understanding of the generated jets.
Authors: Yi Xin, Jianjiang Yang, Siqi Luo, Haodi Zhou, Junlong Du, Xiaohong Liu, Yue Fan, Qing Li, Yuntao Du
Abstract: Large-scale pre-trained vision models (PVMs) have shown great potential for adaptability across various downstream vision tasks. However, with state-of-the-art PVMs growing to billions or even trillions of parameters, the standard full fine-tuning paradigm is becoming unsustainable due to high computational and storage demands. In response, researchers are exploring parameter-efficient fine-tuning (PEFT), which seeks to exceed the performance of full fine-tuning with minimal parameter modifications. This survey provides a comprehensive overview and future directions for visual PEFT, offering a systematic review of the latest advancements. First, we provide a formal definition of PEFT and discuss model pre-training methods. We then categorize existing methods into three categories: addition-based, partial-based, and unified-based. Finally, we introduce the commonly used datasets and applications and suggest potential future research challenges. A comprehensive collection of resources is available at https://github.com/synbol/Awesome-Parameter-Efficient-Transfer-Learning.
URLs: https://github.com/synbol/Awesome-Parameter-Efficient-Transfer-Learning.
Authors: Yizheng Wang, Xiang Li, Ziming Yan, Shuaifeng Ma, Jinshuai Bai, Bokai Liu, Timon Rabczuk, Yinghua Liu
Abstract: Homogenization is a fundamental tool for studying multiscale physical phenomena. Traditional numerical homogenization methods, heavily reliant on finite element analysis, demand significant computational resources, especially for complex geometries, materials, and high-resolution problems. To address these challenges, we propose PreFine-Homo, a novel numerical homogenization framework comprising two phases: pretraining and fine-tuning. In the pretraining phase, a Fourier Neural Operator (FNO) is trained on large datasets to learn the mapping from input geometries and material properties to displacement fields. In the fine-tuning phase, the pretrained predictions serve as initial solutions for iterative algorithms, drastically reducing the number of iterations needed for convergence. The pretraining phase of PreFine-Homo delivers homogenization results up to 1000 times faster than conventional methods, while the fine-tuning phase further enhances accuracy. Moreover, the fine-tuning phase grants PreFine-Homo unlimited generalization capabilities, enabling continuous learning and improvement as data availability increases. We validate PreFine-Homo by predicting the effective elastic tensor for 3D periodic materials, specifically Triply Periodic Minimal Surfaces (TPMS). The results demonstrate that PreFine-Homo achieves high precision, exceptional efficiency, robust learning capabilities, and strong extrapolation ability, establishing it as a powerful tool for multiscale homogenization tasks.
Authors: Maad Ebrahim, Abdelhakim Hafid
Abstract: Real-time Internet of Things (IoT) applications require real-time support to handle the ever-growing demand for computing resources to process IoT workloads. Fog Computing provides high availability of such resources in a distributed manner. However, these resources must be efficiently managed to distribute unpredictable traffic demands among heterogeneous Fog resources. This paper proposes a fully distributed load-balancing solution with Multi-Agent Reinforcement Learning (MARL) that intelligently distributes IoT workloads to optimize the waiting time while providing fair resource utilization in the Fog network. These agents use transfer learning for life-long self-adaptation to dynamic changes in the environment. By leveraging distributed decision-making, MARL agents effectively minimize the waiting time compared to a single centralized agent solution and other baselines, enhancing end-to-end execution delay. Besides performance gain, a fully distributed solution allows for a global-scale implementation where agents can work independently in small collaboration regions, leveraging nearby local resources. Furthermore, we analyze the impact of a realistic frequency to observe the state of the environment, unlike the unrealistic common assumption in the literature of having observations readily available in real-time for every required action. The findings highlight the trade-off between realism and performance using an interval-based Gossip-based multi-casting protocol against assuming real-time observation availability for every generated workload.
Authors: Ed Davis, Ian Gallagher, Daniel John Lawson, Patrick Rubin-Delanchy
Abstract: Dynamic graphs provide a flexible data abstraction for modelling many sorts of real-world systems, such as transport, trade, and social networks. Graph neural networks (GNNs) are powerful tools allowing for different kinds of prediction and inference on these systems, but getting a handle on uncertainty, especially in dynamic settings, is a challenging problem. In this work we propose to use a dynamic graph representation known in the tensor literature as the unfolding, to achieve valid prediction sets via conformal prediction. This representation, a simple graph, can be input to any standard GNN and does not require any modification to existing GNN architectures or conformal prediction routines. One of our key contributions is a careful mathematical consideration of the different inference scenarios which can arise in a dynamic graph modelling context. For a range of practically relevant cases, we obtain valid prediction sets with almost no assumptions, even dispensing with exchangeability. In a more challenging scenario, which we call the semi-inductive regime, we achieve valid prediction under stronger assumptions, akin to stationarity. We provide real data examples demonstrating validity, showing improved accuracy over baselines, and sign-posting different failure modes which can occur when those assumptions are violated.
Authors: David Benrimoh, Caitrin Armstrong, Joseph Mehltretter, Robert Fratila, Kelly Perlman, Sonia Israel, Adam Kapelner, Sagar V. Parikh, Jordan F. Karp, Katherine Heller, Gustavo Turecki
Abstract: INTRODUCTION: The pharmacological treatment of Major Depressive Disorder (MDD) relies on a trial-and-error approach. We introduce an artificial intelligence (AI) model aiming to personalize treatment and improve outcomes, which was deployed in the Artificial Intelligence in Depression Medication Enhancement (AIDME) Study. OBJECTIVES: 1) Develop a model capable of predicting probabilities of remission across multiple pharmacological treatments for adults with at least moderate major depression. 2) Validate model predictions and examine them for amplification of harmful biases. METHODS: Data from previous clinical trials of antidepressant medications were standardized into a common framework and included 9,042 adults with moderate to severe major depression. Feature selection retained 25 clinical and demographic variables. Using Bayesian optimization, a deep learning model was trained on the training set, refined using the validation set, and tested once on the held-out test set. RESULTS: In the evaluation on the held-out test set, the model demonstrated achieved an AUC of 0.65. The model outperformed a null model on the test set (p = 0.01). The model demonstrated clinical utility, achieving an absolute improvement in population remission rate in hypothetical and actual improvement testing. While the model did identify one drug (escitalopram) as generally outperforming the other drugs (consistent with the input data), there was otherwise significant variation in drug rankings. On bias testing, the model did not amplify potentially harmful biases. CONCLUSIONS: We demonstrate the first model capable of predicting outcomes for 10 different treatment options for patients with MDD, intended to be used at or near the start of treatment to personalize treatment. The model was put into clinical practice during the AIDME randomized controlled trial whose results are reported separately.
Authors: Dominick Reilly, Rajatsubhra Chakraborty, Arkaprava Sinha, Manish Kumar Govind, Pu Wang, Francois Bremond, Le Xue, Srijan Das
Abstract: Current Large Language Vision Models (LLVMs) trained on web videos perform well in general video understanding but struggle with fine-grained details, complex human-object interactions (HOI), and view-invariant representation learning essential for Activities of Daily Living (ADL). This limitation stems from a lack of specialized ADL video instruction-tuning datasets and insufficient modality integration to capture discriminative action representations. To address this, we propose a semi-automated framework for curating ADL datasets, creating ADL-X, a multiview, multimodal RGBS instruction-tuning dataset. Additionally, we introduce LLAVIDAL, an LLVM integrating videos, 3D skeletons, and HOIs to model ADL's complex spatiotemporal relationships. For training LLAVIDAL a simple joint alignment of all modalities yields suboptimal results; thus, we propose a Multimodal Progressive (MMPro) training strategy, incorporating modalities in stages following a curriculum. We also establish ADL MCQ and video description benchmarks to assess LLVM performance in ADL tasks. Trained on ADL-X, LLAVIDAL achieves state-of-the-art performance across ADL benchmarks. Code and data will be made publicly available at: https://adl-x.github.io/.
Authors: Sumio Watanabe
Abstract: Mathematical equivalence between statistical mechanics and machine learning theory has been known since the 20th century, and research based on this equivalence has provided novel methodologies in both theoretical physics and statistical learning theory. It is well known that algebraic approaches in statistical mechanics such as operator algebra enable us to analyze phase transition phenomena mathematically. In this paper, we review and prospect algebraic research in machine learning theory for theoretical physicists who are interested in artificial intelligence. If a learning machine has a hierarchical structure or latent variables, then the random Hamiltonian cannot be expressed by any quadratic perturbation because it has singularities. To study an equilibrium state defined by such a singular random Hamiltonian, algebraic approaches are necessary to derive the asymptotic form of the free energy and the generalization error. We also introduce the most recent advance: the theoretical foundation for the alignment of artificial intelligence is now being constructed based on algebraic learning theory. This paper is devoted to the memory of Professor Huzihiro Araki who is a pioneering founder of algebraic research in both statistical mechanics and quantum field theory.
Authors: Luxi He, Yangsibo Huang, Weijia Shi, Tinghao Xie, Haotian Liu, Yue Wang, Luke Zettlemoyer, Chiyuan Zhang, Danqi Chen, Peter Henderson
Abstract: Recent studies show that image and video generation models can be prompted to reproduce copyrighted content from their training data, raising serious legal concerns about copyright infringement. Copyrighted characters (e.g., Mario, Batman) present a significant challenge: at least one lawsuit has already awarded damages based on the generation of such characters. Consequently, commercial services like DALL-E have started deploying interventions. However, little research has systematically examined these problems: (1) Can users easily prompt models to generate copyrighted characters, even if it is unintentional?; (2) How effective are the existing mitigation strategies? To address these questions, we introduce a novel evaluation framework with metrics that assess both the generated image's similarity to copyrighted characters and its consistency with user intent, grounded in a set of popular copyrighted characters from diverse studios and regions. We show that state-of-the-art image and video generation models can still generate characters even if characters' names are not explicitly mentioned, sometimes with only two generic keywords (e.g., prompting with "videogame, plumber" consistently generates Nintendo's Mario character). We also introduce semi-automatic techniques to identify such keywords or descriptions that trigger character generation. Using this framework, we evaluate mitigation strategies, including prompt rewriting and new approaches we propose. Our findings reveal that common methods, such as DALL-E's prompt rewriting, are insufficient alone and require supplementary strategies like negative prompting. Our work provides empirical grounding for discussions on copyright mitigation strategies and offers actionable insights for model deployers implementing these safeguards.
Authors: Ningya Feng, Junwei Pan, Jialong Wu, Baixu Chen, Ximei Wang, Qian Li, Xian Hu, Jie Jiang, Mingsheng Long
Abstract: Lifelong user behavior sequences are crucial for capturing user interests and predicting user responses in modern recommendation systems. A two-stage paradigm is typically adopted to handle these long sequences: a subset of relevant behaviors is first searched from the original long sequences via an attention mechanism in the first stage and then aggregated with the target item to construct a discriminative representation for prediction in the second stage. In this work, we identify and characterize, for the first time, a neglected deficiency in existing long-sequence recommendation models: a single set of embeddings struggles with learning both attention and representation, leading to interference between these two processes. Initial attempts to address this issue with some common methods (e.g., linear projections -- a technique borrowed from language processing) proved ineffective, shedding light on the unique challenges of recommendation models. To overcome this, we propose the Decoupled Attention and Representation Embeddings (DARE) model, where two distinct embedding tables are initialized and learned separately to fully decouple attention and representation. Extensive experiments and analysis demonstrate that DARE provides more accurate searches of correlated behaviors and outperforms baselines with AUC gains up to 0.9% on public datasets and notable improvements on Tencent's advertising platform. Furthermore, decoupling embedding spaces allows us to reduce the attention embedding dimension and accelerate the search procedure by 50% without significant performance impact, enabling more efficient, high-performance online serving. Code in PyTorch for experiments, including model analysis, is available at https://github.com/thuml/DARE.
Authors: Xinyu Zhang, Yuhan Liu, Haonan Chang, Liam Schramm, Abdeslam Boularias
Abstract: Designing a universal policy architecture that performs well across diverse robots and task configurations remains a key challenge. In this work, we address this by representing robot actions as sequential data and generating actions through autoregressive sequence modeling. Existing autoregressive architectures generate end-effector waypoints sequentially as word tokens in language modeling, which are limited to low-frequency control tasks. Unlike language, robot actions are heterogeneous and often include continuous values -- such as joint positions, 2D pixel coordinates, and end-effector poses -- which are not easily suited for language-based modeling. Based on this insight, we introduce a straightforward enhancement: we extend causal transformers' single-token prediction to support predicting a variable number of tokens in a single step through our Chunking Causal Transformer (CCT). This enhancement enables robust performance across diverse tasks of various control frequencies, greater efficiency by having fewer autoregression steps, and lead to a hybrid action sequence design by mixing different types of actions and using a different chunk size for each action type. Based on CCT, we propose the Autoregressive Policy (ARP) architecture, which solves manipulation tasks by generating hybrid action sequences. We evaluate ARP across diverse robotic manipulation environments, including Push-T, ALOHA, and RLBench, and show that ARP, as a universal architecture, matches or outperforms the environment-specific state-of-the-art in all tested benchmarks, while being more efficient in computation and parameter sizes. Videos of our real robot demonstrations, all source code and the pretrained models of ARP can be found at http://github.com/mlzxy/arp.
Authors: Jiacheng Wang, Xiang Chen, Renjiu Hu, Rongguang Wang, Min Liu, Yaonan Wang, Jiazheng Wang, Hao Li, Hang Zhang
Abstract: Co-examination of second-harmonic generation (SHG) and bright-field (BF) microscopy enables the differentiation of tissue components and collagen fibers, aiding the analysis of human breast and pancreatic cancer tissues. However, large discrepancies between SHG and BF images pose challenges for current learning-based registration models in aligning SHG to BF. In this paper, we propose a novel multi-modal registration framework that employs fidelity-imposed displacement editing to address these challenges. The framework integrates batch-wise contrastive learning, feature-based pre-alignment, and instance-level optimization. Experimental results from the Learn2Reg COMULISglobe SHG-BF Challenge validate the effectiveness of our method, securing the 1st place on the online leaderboard.
Authors: Amirabbas Afzali, Borna Khodabandeh, Ali Rasekh, Mahyar JafariNodeh, Sepehr kazemi, Simon Gottschalk
Abstract: Contrastive learning models have demonstrated impressive abilities to capture semantic similarities by aligning representations in the embedding space. However, their performance can be limited by the quality of the training data and its inherent biases. While Preference Optimization (PO) methods such as Reinforcement Learning from Human Feedback (RLHF) and Direct Preference Optimization (DPO) have been applied to align generative models with human preferences, their use in contrastive learning has yet to be explored. This paper introduces a novel method for training contrastive learning models using different PO methods to break down complex concepts. Our method systematically aligns model behavior with desired preferences, enhancing performance on the targeted task. In particular, we focus on enhancing model robustness against typographic attacks and inductive biases, commonly seen in contrastive vision-language models like CLIP. Our experiments demonstrate that models trained using PO outperform standard contrastive learning techniques while retaining their ability to handle adversarial challenges and maintain accuracy on other downstream tasks. This makes our method well-suited for tasks requiring fairness, robustness, and alignment with specific preferences. We evaluate our method for tackling typographic attacks on images and explore its ability to disentangle gender concepts and mitigate gender bias, showcasing the versatility of our approach.
Authors: Yanwei Wang, Lirui Wang, Yilun Du, Balakumar Sundaralingam, Xuning Yang, Yu-Wei Chao, Claudia Perez-D'Arpino, Dieter Fox, Julie Shah
Abstract: Generative policies trained with human demonstrations can autonomously accomplish multimodal, long-horizon tasks. However, during inference, humans are often removed from the policy execution loop, limiting the ability to guide a pre-trained policy towards a specific sub-goal or trajectory shape among multiple predictions. Naive human intervention may inadvertently exacerbate distribution shift, leading to constraint violations or execution failures. To better align policy output with human intent without inducing out-of-distribution errors, we propose an Inference-Time Policy Steering (ITPS) framework that leverages human interactions to bias the generative sampling process, rather than fine-tuning the policy on interaction data. We evaluate ITPS across three simulated and real-world benchmarks, testing three forms of human interaction and associated alignment distance metrics. Among six sampling strategies, our proposed stochastic sampling with diffusion policy achieves the best trade-off between alignment and distribution shift. Videos are available at https://yanweiw.github.io/itps/.
Authors: Ryan G. Dempsey, Jonathan Ethier, Halim Yanikomeroglu
Abstract: Radio deployments and spectrum planning benefit from path loss predictions. Obstructions along a communications link are often considered implicitly or through derived metrics such as representative clutter height or total obstruction depth. In this paper, we propose a path-specific path loss prediction method that uses convolutional neural networks to automatically perform feature extraction from 2-D obstruction height maps. Our methods result in low prediction error in a variety of environments without requiring derived metrics.
Authors: Sankalp Sinha, Mohammad Sadil Khan, Muhammad Usama, Shino Sam, Didier Stricker, Sk Aziz Ali, Muhammad Zeshan Afzal
Abstract: Generating high-fidelity 3D content from text prompts remains a significant challenge in computer vision due to the limited size, diversity, and annotation depth of the existing datasets. To address this, we introduce MARVEL-40M+, an extensive dataset with 40 million text annotations for over 8.9 million 3D assets aggregated from seven major 3D datasets. Our contribution is a novel multi-stage annotation pipeline that integrates open-source pretrained multi-view VLMs and LLMs to automatically produce multi-level descriptions, ranging from detailed (150-200 words) to concise semantic tags (10-20 words). This structure supports both fine-grained 3D reconstruction and rapid prototyping. Furthermore, we incorporate human metadata from source datasets into our annotation pipeline to add domain-specific information in our annotation and reduce VLM hallucinations. Additionally, we develop MARVEL-FX3D, a two-stage text-to-3D pipeline. We fine-tune Stable Diffusion with our annotations and use a pretrained image-to-3D network to generate 3D textured meshes within 15s. Extensive evaluations show that MARVEL-40M+ significantly outperforms existing datasets in annotation quality and linguistic diversity, achieving win rates of 72.41% by GPT-4 and 73.40% by human evaluators. Project page is available at https://sankalpsinha-cmos.github.io/MARVEL/.
Authors: Zhixuan Liang, Yao Mu, Yixiao Wang, Tianxing Chen, Wenqi Shao, Wei Zhan, Masayoshi Tomizuka, Ping Luo, Mingyu Ding
Abstract: Dexterous manipulation with contact-rich interactions is crucial for advanced robotics. While recent diffusion-based planning approaches show promise for simple manipulation tasks, they often produce unrealistic ghost states (e.g., the object automatically moves without hand contact) or lack adaptability when handling complex sequential interactions. In this work, we introduce DexHandDiff, an interaction-aware diffusion planning framework for adaptive dexterous manipulation. DexHandDiff models joint state-action dynamics through a dual-phase diffusion process which consists of pre-interaction contact alignment and post-contact goal-directed control, enabling goal-adaptive generalizable dexterous manipulation. Additionally, we incorporate dynamics model-based dual guidance and leverage large language models for automated guidance function generation, enhancing generalizability for physical interactions and facilitating diverse goal adaptation through language cues. Experiments on physical interaction tasks such as door opening, pen and block re-orientation, object relocation, and hammer striking demonstrate DexHandDiff's effectiveness on goals outside training distributions, achieving over twice the average success rate (59.2% vs. 29.5%) compared to existing methods. Our framework achieves an average of 70.7% success rate on goal adaptive dexterous tasks, highlighting its robustness and flexibility in contact-rich manipulation.
Authors: Nardiena A. Pratama, Shaoyang Fan, Gianluca Demartini
Abstract: Human-annotated content is often used to train machine learning (ML) models. However, recently, language and multi-modal foundational models have been used to replace and scale-up human annotator's efforts. This study compares human-generated and ML-generated annotations of images representing diverse socio-economic contexts. We aim to understand differences in perception and identify potential biases in content interpretation. Our dataset comprises images of people from various geographical regions and income levels, covering various daily activities and home environments. We compare human and ML-generated annotations semantically and evaluate their impact on predictive models. Our results show highest similarity between ML captions and human labels from a low-level perspective, i.e., types of words that appear and sentence structures, but all three annotations are alike in how similar or dissimilar they perceive images across different regions. Additionally, ML Captions resulted in best overall region classification performance, while ML Objects and ML Captions performed best overall for income regression. The varying performance of annotation sets highlights the notion that all annotations are important, and that human-generated annotations are yet to be replaceable.
Authors: Yihao Wang, Marcus Klasson, Matias Turkulainen, Shuzhe Wang, Juho Kannala, Arno Solin
Abstract: Gaussian splatting enables fast novel view synthesis in static 3D environments. However, reconstructing real-world environments remains challenging as distractors or occluders break the multi-view consistency assumption required for accurate 3D reconstruction. Most existing methods rely on external semantic information from pre-trained models, introducing additional computational overhead as pre-processing steps or during optimization. In this work, we propose a novel method, DeSplat, that directly separates distractors and static scene elements purely based on volume rendering of Gaussian primitives. We initialize Gaussians within each camera view for reconstructing the view-specific distractors to separately model the static 3D scene and distractors in the alpha compositing stages. DeSplat yields an explicit scene separation of static elements and distractors, achieving comparable results to prior distractor-free approaches without sacrificing rendering speed. We demonstrate DeSplat's effectiveness on three benchmark data sets for distractor-free novel view synthesis. See the project website at https://aaltoml.github.io/desplat/.
Authors: Bikang Pan, Qun Li, Xiaoying Tang, Wei Huang, Zhen Fang, Feng Liu, Jingya Wang, Jingyi Yu, Ye Shi
Abstract: The emergence of vision-language foundation models, such as CLIP, has revolutionized image-text representation, enabling a broad range of applications via prompt learning. Despite its promise, real-world datasets often contain noisy labels that can degrade prompt learning performance. In this paper, we demonstrate that using mean absolute error (MAE) loss in prompt learning, named PromptMAE, significantly enhances robustness against noisy labels while maintaining high accuracy. Though MAE is straightforward and recognized for its robustness, it is rarely used in noisy-label learning due to its slow convergence and poor performance outside prompt learning scenarios. To elucidate the robustness of PromptMAE, we leverage feature learning theory to show that MAE can suppress the influence of noisy samples, thereby improving the signal-to-noise ratio and enhancing overall robustness. Additionally, we introduce PromptOT, a prompt-based optimal transport data purification method to enhance the robustness further. PromptOT employs text features in vision-language models as prototypes to construct an optimal transportation matrix. This matrix effectively partitions datasets into clean and noisy subsets, allowing for the application of cross-entropy loss to the clean subset and MAE loss to the noisy subset. Our Noise-Label Prompt Learning method, named NLPrompt, offers a simple and efficient approach that leverages the expressive representations and precise alignment capabilities of vision-language models for robust prompt learning. We validate NLPrompt through extensive experiments across various noise settings, demonstrating significant performance improvements.
Authors: Sanghwan Kim, Rui Xiao, Mariana-Iuliana Georgescu, Stephan Alaniz, Zeynep Akata
Abstract: Vision-Language Models (VLMs) trained with contrastive loss have achieved significant advancements in various vision and language tasks. However, the global nature of the contrastive loss makes VLMs focus predominantly on foreground objects, neglecting other crucial information in the image, which limits their effectiveness in downstream tasks. To address these challenges, we propose COSMOS: CrOSs-MOdality Self-distillation for vision-language pre-training that integrates a novel text-cropping strategy and cross-attention module into a self-supervised learning framework. We create global and local views of images and texts (i.e., multi-modal augmentations), which are essential for self-distillation in VLMs. We further introduce a cross-attention module, enabling COSMOS to learn comprehensive cross-modal representations optimized via a cross-modality self-distillation loss. COSMOS consistently outperforms previous strong baselines on various zero-shot downstream tasks, including retrieval, classification, and semantic segmentation. Additionally, it surpasses CLIP-based models trained on larger datasets in visual perception and contextual understanding tasks. Code is available at https://github.com/ExplainableML/cosmos.
Authors: Zhifei Yang, Keyang Lu, Chao Zhang, Jiaxing Qi, Hanqi Jiang, Ruifei Ma, Shenglin Yin, Yifan Xu, Mingzhe Xing, Zhen Xiao, Jieyi Long, Guangyao Zhai
Abstract: Controllable 3D scene generation has extensive applications in virtual reality and interior design, where the generated scenes should exhibit high levels of realism and controllability in terms of geometry. Scene graphs provide a suitable data representation that facilitates these applications. However, current graph-based methods for scene generation are constrained to text-based inputs and exhibit insufficient adaptability to flexible user inputs, hindering the ability to precisely control object geometry. To address this issue, we propose MMGDreamer, a dual-branch diffusion model for scene generation that incorporates a novel Mixed-Modality Graph, visual enhancement module, and relation predictor. The mixed-modality graph allows object nodes to integrate textual and visual modalities, with optional relationships between nodes. It enhances adaptability to flexible user inputs and enables meticulous control over the geometry of objects in the generated scenes. The visual enhancement module enriches the visual fidelity of text-only nodes by constructing visual representations using text embeddings. Furthermore, our relation predictor leverages node representations to infer absent relationships between nodes, resulting in more coherent scene layouts. Extensive experimental results demonstrate that MMGDreamer exhibits superior control of object geometry, achieving state-of-the-art scene generation performance. Project page: https://yangzhifeio.github.io/project/MMGDreamer.
Authors: Fenglu Hong, Ravi Raju, Jonathan Lingjie Li, Bo Li, Urmish Thakker, Avinash Ravichandran, Swayambhoo Jain, Changran Hu
Abstract: Speculative decoding is an effective method for accelerating inference of large language models (LLMs) by employing a small draft model to predict the output of a target model. However, when adapting speculative decoding to domain-specific target models, the acceptance rate of the generic draft model drops significantly due to domain shift. In this work, we systematically investigate knowledge distillation techniques for training domain draft models to improve their speculation accuracy. We compare white-box and black-box distillation approaches and explore their effectiveness in various data accessibility scenarios, including historical user queries, curated domain data, and synthetically generated alignment data. Our experiments across Function Calling, Biology, and Chinese domains show that offline distillation consistently outperforms online distillation by 11% to 25%, white-box distillation surpasses black-box distillation by 2% to 10%, and data scaling trends hold across domains. Additionally, we find that synthetic data can effectively align draft models and achieve 80% to 93% of the performance of training on historical user queries. These findings provide practical guidelines for training domain-specific draft models to improve speculative decoding efficiency.
Authors: Junzhe Li, Xuerui Qiu, Linrui Xu, Liya Guo, Delin Qu, Tingting Long, Chun Fan, Ming Li
Abstract: Unified multimodal models (UMMs) have emerged as a powerful paradigm in foundational computer vision research, demonstrating significant potential in both image understanding and generation. However, existing research in the face domain primarily focuses on $\textbf{coarse}$ facial attribute understanding, with limited capacity to handle $\textbf{fine-grained}$ facial attributes and without addressing generation capabilities. To overcome these limitations, we propose Uni$\textbf{F}^2$ace, the first UMM tailored specifically for fine-grained face understanding and generation. In general, we train Uni$\textbf{F}^2$ace on a self-constructed, specialized dataset utilizing two mutually beneficial diffusion techniques and a two-level mixture-of-experts architecture. Specifically, we first build a large-scale facial dataset, Uni$\textbf{F}^2$ace-130K, which contains 130K image-text pairs with one million question-answering pairs that span a wide range of facial attributes. Second, we establish a theoretical connection between discrete diffusion score matching and masked generative models, optimizing both evidence lower bounds simultaneously, which significantly improves the model's ability to synthesize facial details. Finally, we introduce both token-level and sequence-level mixture-of-experts, enabling efficient fine-grained representation learning for both understanding and generation tasks. Extensive experiments on Uni$\textbf{F}^2$ace-130K demonstrate that Uni$\textbf{F}^2$ace outperforms existing UMMs and generative models, achieving superior performance across both understanding and generation tasks.
Authors: Hongyu Ke, Jack Morris, Kentaro Oguchi, Xiaofei Cao, Yongkang Liu, Haoxin Wang, Yi Ding
Abstract: 3D visual perception tasks, such as 3D detection from multi-camera images, are essential components of autonomous driving and assistance systems. However, designing computationally efficient methods remains a significant challenge. In this paper, we propose a Mamba-based framework called MamBEV, which learns unified Bird's Eye View (BEV) representations using linear spatio-temporal SSM-based attention. This approach supports multiple 3D perception tasks with significantly improved computational and memory efficiency. Furthermore, we introduce SSM based cross-attention, analogous to standard cross attention, where BEV query representations can interact with relevant image features. Extensive experiments demonstrate MamBEV's promising performance across diverse visual perception metrics, highlighting its advantages in input scaling efficiency compared to existing benchmark models.
Authors: Caifeng Zou, Zachary E. Ross, Robert W. Clayton, Fan-Chi Lin, Kamyar Azizzadenesheli
Abstract: Numerical simulations of seismic wave propagation are crucial for investigating velocity structures and improving seismic hazard assessment. However, standard methods such as finite difference or finite element are computationally expensive. Recent studies have shown that a new class of machine learning models, called neural operators, can solve the elastodynamic wave equation orders of magnitude faster than conventional methods. Full waveform inversion is a prime beneficiary of the accelerated simulations. Neural operators, as end-to-end differentiable operators, combined with automatic differentiation, provide an alternative approach to the adjoint-state method. Since neural operators do not involve the Born approximation, when used for full waveform inversion they have the potential to include additional phases and alleviate cycle-skipping problems present in traditional adjoint-state formulations. In this study, we demonstrate the first application of neural operators for full waveform inversion on a real seismic dataset, which consists of several nodal transects collected across the San Gabriel, Chino, and San Bernardino basins in the Los Angeles metropolitan area.
Authors: Minh-Nhat Phung, Minh-Binh Tran
Abstract: From the perspective of control theory, neural differential equations (neural ODEs) have become an important tool for supervised learning. In the fundamental work of Ruiz-Balet and Zuazua (SIAM REVIEW 2023), the authors pose an open problem regarding the connection between control theory, optimal transport theory, and neural differential equations. More precisely, they inquire how one can quantify the closeness of the optimal flows in neural transport equations to the true dynamic optimal transport. In this work, we propose a construction of neural differential equations that converge to the true dynamic optimal transport in the limit, providing a significant step in solving the formerly mentioned open problem.
Authors: Julian Junyan Wang, Victor Xiaoqi Wang
Abstract: This study provides the first comprehensive assessment of consistency and reproducibility in Large Language Model (LLM) outputs in finance and accounting research. We evaluate how consistently LLMs produce outputs given identical inputs through extensive experimentation with 50 independent runs across five common tasks: classification, sentiment analysis, summarization, text generation, and prediction. Using three OpenAI models (GPT-3.5-turbo, GPT-4o-mini, and GPT-4o), we generate over 3.4 million outputs from diverse financial source texts and data, covering MD&As, FOMC statements, finance news articles, earnings call transcripts, and financial statements. Our findings reveal substantial but task-dependent consistency, with binary classification and sentiment analysis achieving near-perfect reproducibility, while complex tasks show greater variability. More advanced models do not consistently demonstrate better consistency and reproducibility, with task-specific patterns emerging. LLMs significantly outperform expert human annotators in consistency and maintain high agreement even where human experts significantly disagree. We further find that simple aggregation strategies across 3-5 runs dramatically improve consistency. We also find that aggregation may come with an additional benefit of improved accuracy for sentiment analysis when using newer models. Simulation analysis reveals that despite measurable inconsistency in LLM outputs, downstream statistical inferences remain remarkably robust. These findings address concerns about what we term "G-hacking," the selective reporting of favorable outcomes from multiple Generative AI runs, by demonstrating that such risks are relatively low for finance and accounting tasks.
Authors: Alexandra Arzberger, Ramin Tavakoli Kolagari
Abstract: Light Detection and Ranging (LiDAR) is an essential sensor technology for autonomous driving as it can capture high-resolution 3D data. As 3D object detection systems (OD) can interpret such point cloud data, they play a key role in the driving decisions of autonomous vehicles. Consequently, such 3D OD must be robust against all types of perturbations and must therefore be extensively tested. One approach is the use of adversarial examples, which are small, sometimes sophisticated perturbations in the input data that change, i.e., falsify, the prediction of the OD. These perturbations are carefully designed based on the weaknesses of the OD. The robustness of the OD cannot be quantified with adversarial examples in general, because if the OD is vulnerable to a given attack, it is unclear whether this is due to the robustness of the OD or whether the attack algorithm produces particularly strong adversarial examples. The contribution of this work is Hi-ALPS -- Hierarchical Adversarial-example-based LiDAR Perturbation Level System, where higher robustness of the OD is required to withstand the perturbations as the perturbation levels increase. In doing so, the Hi-ALPS levels successively implement a heuristic followed by established adversarial example approaches. In a series of comprehensive experiments using Hi-ALPS, we quantify the robustness of six state-of-the-art 3D OD under different types of perturbations. The results of the experiments show that none of the OD is robust against all Hi-ALPS levels; an important factor for the ranking is that human observers can still correctly recognize the perturbed objects, as the respective perturbations are small. To increase the robustness of the OD, we discuss the applicability of state-of-the-art countermeasures. In addition, we derive further suggestions for countermeasures based on our experimental results.
Authors: Jaihoon Kim, Taehoon Yoon, Jisung Hwang, Minhyuk Sung
Abstract: We propose an inference-time scaling approach for pretrained flow models. Recently, inference-time scaling has gained significant attention in LLMs and diffusion models, improving sample quality or better aligning outputs with user preferences by leveraging additional computation. For diffusion models, particle sampling has allowed more efficient scaling due to the stochasticity at intermediate denoising steps. On the contrary, while flow models have gained popularity as an alternative to diffusion models--offering faster generation and high-quality outputs in state-of-the-art image and video generative models--efficient inference-time scaling methods used for diffusion models cannot be directly applied due to their deterministic generative process. To enable efficient inference-time scaling for flow models, we propose three key ideas: 1) SDE-based generation, enabling particle sampling in flow models, 2) Interpolant conversion, broadening the search space and enhancing sample diversity, and 3) Rollover Budget Forcing (RBF), an adaptive allocation of computational resources across timesteps to maximize budget utilization. Our experiments show that SDE-based generation, particularly variance-preserving (VP) interpolant-based generation, improves the performance of particle sampling methods for inference-time scaling in flow models. Additionally, we demonstrate that RBF with VP-SDE achieves the best performance, outperforming all previous inference-time scaling approaches.
Authors: Aaron Serianni, Tyler Zhu, Olga Russakovsky, Vikram V. Ramaswamy
Abstract: Computer vision models have been shown to exhibit and amplify biases across a wide array of datasets and tasks. Existing methods for quantifying bias in classification models primarily focus on dataset distribution and model performance on subgroups, overlooking the internal workings of a model. We introduce the Attention-IoU (Attention Intersection over Union) metric and related scores, which use attention maps to reveal biases within a model's internal representations and identify image features potentially causing the biases. First, we validate Attention-IoU on the synthetic Waterbirds dataset, showing that the metric accurately measures model bias. We then analyze the CelebA dataset, finding that Attention-IoU uncovers correlations beyond accuracy disparities. Through an investigation of individual attributes through the protected attribute of Male, we examine the distinct ways biases are represented in CelebA. Lastly, by subsampling the training set to change attribute correlations, we demonstrate that Attention-IoU reveals potential confounding variables not present in dataset labels.