Authors: Jiahua Lan, Sen Zhang, Haixia Pan, Ruijun Liu, Li Shen, Dacheng Tao
Abstract: In contrast to the human ability to continuously acquire knowledge, agents struggle with the stability-plasticity dilemma in deep reinforcement learning (DRL), which refers to the trade-off between retaining existing skills (stability) and learning new knowledge (plasticity). Current methods focus on balancing these two aspects at the network level, lacking sufficient differentiation and fine-grained control of individual neurons. To overcome this limitation, we propose Neuron-level Balance between Stability and Plasticity (NBSP) method, by taking inspiration from the observation that specific neurons are strongly relevant to task-relevant skills. Specifically, NBSP first (1) defines and identifies RL skill neurons that are crucial for knowledge retention through a goal-oriented method, and then (2) introduces a framework by employing gradient masking and experience replay techniques targeting these neurons to preserve the encoded existing skills while enabling adaptation to new tasks. Numerous experimental results on the Meta-World and Atari benchmarks demonstrate that NBSP significantly outperforms existing approaches in balancing stability and plasticity.
Authors: Krzysztof Pancerz
Abstract: We present theoretical rudiments of Petri nets over ontological graphs as well as the designed and implemented Python toolkit for dealing with such nets. In Petri nets over ontological graphs, the domain knowledge is enclosed in a form of ontologies. In this way, some valuable knowledge (especially in terms of semantic relations) can be added to model reasoning and control processes by means of Petri nets. In the implemented approach, ontological graphs are obtained from ontologies built in accordance with the OWL 2 Web Ontology Language. The implemented tool enables the users to define the structure and dynamics of Petri nets over ontological graphs.
Authors: Robert Susmaga, Izabela Szczech
Abstract: TOPSIS, a popular method for ranking alternatives is based on aggregated distances to ideal and anti-ideal points. As such, it was considered to be essentially different from widely popular and acknowledged `utility-based methods', which build rankings from weight-averaged utility values. Nonetheless, TOPSIS has recently been shown to be a natural generalization of these `utility-based methods' on the grounds that the distances it uses can be decomposed into so called weight-scaled means (WM) and weight-scaled standard deviations (WSD) of utilities. However, the influence that these two components exert on the final ranking cannot be in any way influenced in the standard TOPSIS. This is why, building on our previous results, in this paper we put forward modifications that make TOPSIS aggregations responsive to WM and WSD, achieving some amount of well interpretable control over how the rankings are influenced by WM and WSD. The modifications constitute a natural generalization of the standard TOPSIS method because, thanks to them, the generalized TOPSIS may turn into the original TOPSIS or, otherwise, following the decision maker's preferences, may trade off WM for WSD or WSD for WM. In the latter case, TOPSIS gradually reduces to a regular `utility-based method'. All in all, we believe that the proposed generalizations constitute an interesting practical tool for influencing the ranking by controlled application of a new form of decision maker's preferences.
Authors: Yutaro Yamada, Robert Tjarko Lange, Cong Lu, Shengran Hu, Chris Lu, Jakob Foerster, Jeff Clune, David Ha
Abstract: AI is increasingly playing a pivotal role in transforming how scientific discoveries are made. We introduce The AI Scientist-v2, an end-to-end agentic system capable of producing the first entirely AI generated peer-review-accepted workshop paper. This system iteratively formulates scientific hypotheses, designs and executes experiments, analyzes and visualizes data, and autonomously authors scientific manuscripts. Compared to its predecessor (v1, Lu et al., 2024 arXiv:2408.06292), The AI Scientist-v2 eliminates the reliance on human-authored code templates, generalizes effectively across diverse machine learning domains, and leverages a novel progressive agentic tree-search methodology managed by a dedicated experiment manager agent. Additionally, we enhance the AI reviewer component by integrating a Vision-Language Model (VLM) feedback loop for iterative refinement of content and aesthetics of the figures. We evaluated The AI Scientist-v2 by submitting three fully autonomous manuscripts to a peer-reviewed ICLR workshop. Notably, one manuscript achieved high enough scores to exceed the average human acceptance threshold, marking the first instance of a fully AI-generated paper successfully navigating a peer review. This accomplishment highlights the growing capability of AI in conducting all aspects of scientific research. We anticipate that further advancements in autonomous scientific discovery technologies will profoundly impact human knowledge generation, enabling unprecedented scalability in research productivity and significantly accelerating scientific breakthroughs, greatly benefiting society at large. We have open-sourced the code at https://github.com/SakanaAI/AI-Scientist-v2 to foster the future development of this transformative technology. We also discuss the role of AI in science, including AI safety.
Authors: Eser Kandogan, Nikita Bhutani, Dan Zhang, Rafael Li Chen, Sairam Gurajada, Estevam Hruschka
Abstract: Large language models (LLMs) have gained significant interest in industry due to their impressive capabilities across a wide range of tasks. However, the widespread adoption of LLMs presents several challenges, such as integration into existing applications and infrastructure, utilization of company proprietary data, models, and APIs, and meeting cost, quality, responsiveness, and other requirements. To address these challenges, there is a notable shift from monolithic models to compound AI systems, with the premise of more powerful, versatile, and reliable applications. However, progress thus far has been piecemeal, with proposals for agentic workflows, programming models, and extended LLM capabilities, without a clear vision of an overall architecture. In this paper, we propose a 'blueprint architecture' for compound AI systems for orchestrating agents and data for enterprise applications. In our proposed architecture the key orchestration concept is 'streams' to coordinate the flow of data and instructions among agents. Existing proprietary models and APIs in the enterprise are mapped to 'agents', defined in an 'agent registry' that serves agent metadata and learned representations for search and planning. Agents can utilize proprietary data through a 'data registry' that similarly registers enterprise data of various modalities. Tying it all together, data and task 'planners' break down, map, and optimize tasks and queries for given quality of service (QoS) requirements such as cost, accuracy, and latency. We illustrate an implementation of the architecture for a use-case in the HR domain and discuss opportunities and challenges for 'agentic AI' in the enterprise.
Authors: Junmo Kim, Namkyeong Lee, Jiwon Kim, Kwangsoo Kim
Abstract: Electronic health record (EHR) foundation models have been an area ripe for exploration with their improved performance in various medical tasks. Despite the rapid advances, there exists a fundamental limitation: Processing unseen medical codes out of the vocabulary. This problem limits the generality of EHR foundation models and the integration of models trained with different vocabularies. To deal with this problem, we propose MedRep for EHR foundation models based on the observational medical outcome partnership (OMOP) common data model (CDM), providing the integrated medical concept representations and the basic data augmentation strategy for patient trajectories. For concept representation learning, we enrich the information of each concept with a minimal definition through large language model (LLM) prompts and enhance the text-based representations through graph ontology of OMOP vocabulary. Trajectory augmentation randomly replaces selected concepts with other similar concepts that have closely related representations to let the model practice with the concepts out-of-vocabulary. Finally, we demonstrate that EHR foundation models trained with MedRep better maintain the prediction performance in external datasets. Our code implementation is publicly available at https://github.com/kicarussays/MedRep.
Authors: Paul J. Pritz, Kin K. Leung
Abstract: Reinforcement learning in partially observable environments is typically challenging, as it requires agents to learn an estimate of the underlying system state. These challenges are exacerbated in multi-agent settings, where agents learn simultaneously and influence the underlying state as well as each others' observations. We propose the use of learned beliefs on the underlying state of the system to overcome these challenges and enable reinforcement learning with fully decentralized training and execution. Our approach leverages state information to pre-train a probabilistic belief model in a self-supervised fashion. The resulting belief states, which capture both inferred state information as well as uncertainty over this information, are then used in a state-based reinforcement learning algorithm to create an end-to-end model for cooperative multi-agent reinforcement learning under partial observability. By separating the belief and reinforcement learning tasks, we are able to significantly simplify the policy and value function learning tasks and improve both the convergence speed and the final performance. We evaluate our proposed method on diverse partially observable multi-agent tasks designed to exhibit different variants of partial observability.
Authors: Ye Ye
Abstract: Large Language Models (LLMs) are increasingly used as autonomous agents for multi-step tasks. However, most existing frameworks fail to maintain a structured understanding of the task state, often relying on linear prompt concatenation or shallow memory buffers. This leads to brittle performance, frequent hallucinations, and poor long-range coherence. In this work, we propose the Task Memory Engine (TME), a lightweight and structured memory module that tracks task execution using a hierarchical Task Memory Tree (TMT). Each node in the tree corresponds to a task step, storing relevant input, output, status, and sub-task relationships. We introduce a prompt synthesis method that dynamically generates LLM prompts based on the active node path, significantly improving execution consistency and contextual grounding. Through case studies and comparative experiments on multi-step agent tasks, we demonstrate that TME leads to better task completion accuracy and more interpretable behavior with minimal implementation overhead. The full implementation of TME is available at https://github.com/biubiutomato/TME-Agent.
Authors: Esperan\c{c}a Amengual-Alcover, Antoni Jaume-i-Cap\'o, Miquel Mir\'o-Nicolau, Gabriel Moy\`a-Alcover, Antonia Paniza-Fullana
Abstract: The integration of Artificial Intelligence in the development of computer systems presents a new challenge: make intelligent systems explainable to humans. This is especially vital in the field of health and well-being, where transparency in decision support systems enables healthcare professionals to understand and trust automated decisions and predictions. To address this need, tools are required to guide the development of explainable AI systems. In this paper, we introduce an evaluation framework designed to support the development of explainable AI systems for health and well-being. Additionally, we present a case study that illustrates the application of the framework in practice. We believe that our framework can serve as a valuable tool not only for developing explainable AI systems in healthcare but also for any AI system that has a significant impact on individuals.
Authors: Alessio Buscemi, Daniele Proverbio, Paolo Bova, Nataliya Balabanova, Adeela Bashir, Theodor Cimpeanu, Henrique Correia da Fonseca, Manh Hong Duong, Elias Fernandez Domingos, Antonio M. Fernandes, Marcus Krellner, Ndidi Bianca Ogbo, Simon T. Powers, Fernando P. Santos, Zia Ush Shamszaman, Zhao Song, Alessandro Di Stefano, The Anh Han
Abstract: There is general agreement that fostering trust and cooperation within the AI development ecosystem is essential to promote the adoption of trustworthy AI systems. By embedding Large Language Model (LLM) agents within an evolutionary game-theoretic framework, this paper investigates the complex interplay between AI developers, regulators and users, modelling their strategic choices under different regulatory scenarios. Evolutionary game theory (EGT) is used to quantitatively model the dilemmas faced by each actor, and LLMs provide additional degrees of complexity and nuances and enable repeated games and incorporation of personality traits. Our research identifies emerging behaviours of strategic AI agents, which tend to adopt more "pessimistic" (not trusting and defective) stances than pure game-theoretic agents. We observe that, in case of full trust by users, incentives are effective to promote effective regulation; however, conditional trust may deteriorate the "social pact". Establishing a virtuous feedback between users' trust and regulators' reputation thus appears to be key to nudge developers towards creating safe AI. However, the level at which this trust emerges may depend on the specific LLM used for testing. Our results thus provide guidance for AI regulation systems, and help predict the outcome of strategic LLM agents, should they be used to aid regulation itself.
Authors: Qianou Ma, Dora Zhao, Xinran Zhao, Chenglei Si, Chenyang Yang, Ryan Louie, Ehud Reiter, Diyi Yang, Tongshuang Wu
Abstract: In the era of Large Language Models (LLMs), establishing effective evaluation methods and standards for diverse human-AI interaction systems is increasingly challenging. To encourage more transparent documentation and facilitate discussion on human-AI system evaluation design options, we present an evaluation card SPHERE, which encompasses five key dimensions: 1) What is being evaluated?; 2) How is the evaluation conducted?; 3) Who is participating in the evaluation?; 4) When is evaluation conducted?; 5) How is evaluation validated? We conduct a review of 39 human-AI systems using SPHERE, outlining current evaluation practices and areas for improvement. We provide three recommendations for improving the validity and rigor of evaluation practices.
Authors: Harishwar Reddy, Madhusudan Srinivasan, Upulee Kanewala
Abstract: Large Language Models (LLMs) have made significant strides in Natural Language Processing but remain vulnerable to fairness-related issues, often reflecting biases inherent in their training data. These biases pose risks, particularly when LLMs are deployed in sensitive areas such as healthcare, finance, and law. This paper introduces a metamorphic testing approach to systematically identify fairness bugs in LLMs. We define and apply a set of fairness-oriented metamorphic relations (MRs) to assess the LLaMA and GPT model, a state-of-the-art LLM, across diverse demographic inputs. Our methodology includes generating source and follow-up test cases for each MR and analyzing model responses for fairness violations. The results demonstrate the effectiveness of MT in exposing bias patterns, especially in relation to tone and sentiment, and highlight specific intersections of sensitive attributes that frequently reveal fairness faults. This research improves fairness testing in LLMs, providing a structured approach to detect and mitigate biases and improve model robustness in fairness-sensitive applications.
Authors: Shurui Wu, Xinyi Huang, Dingxin Lu
Abstract: As the prevalence of mental health crises increases on social media platforms, identifying and preventing potential harm has become an urgent challenge. This study introduces a large language model (LLM)-based text transfer recognition method for social network crisis intervention, enhanced with domain-specific mental health knowledge. We propose a multi-level framework that incorporates transfer learning using BERT, and integrates mental health knowledge, sentiment analysis, and behavior prediction techniques. The framework includes a crisis annotation tool trained on social media datasets from real-world events, enabling the model to detect nuanced emotional cues and identify psychological crises. Experimental results show that the proposed method outperforms traditional models in crisis detection accuracy and exhibits greater sensitivity to subtle emotional and contextual variations.
Authors: Runjin Chen, Zhenyu Zhang, Junyuan Hong, Souvik Kundu, Zhangyang Wang
Abstract: Large Language Models (LLMs), such as OpenAI's o1-series have demonstrated compelling capabilities for complex reasoning tasks via the extended chain-of-thought (CoT) reasoning mechanism. However, recent studies reveal substantial redundancy in the CoT reasoning traces, which not only increases inference latency but also negatively impacts model performance by diverting attention to unnecessary reasoning paths. To address this issue, we investigate the internal reasoning structures of LLMs and categorize them into three primary thought types: execution, reflection, and transition thoughts. Moreover, our analysis reveals that excessive reflection and transition thoughts are strongly correlated with failure cases and these thought categories exhibit clear separation in the latent space. Based on these, we introduce SEAL (Steerable reasoning calibration), a training-free approach that seamlessly calibrates the CoT process, improving accuracy while demonstrating significant efficiency gains. SEAL consists of an offline stage for extracting the reasoning steering vector in the latent space, followed by an on-the-fly calibration of the reasoning trace through representation intervention using the steering vector. Notably, the steering vector exhibits strong transferability across various tasks. Extensive experiments across multiple models (DeepSeek-R1-Distill and QwQ-32B-Preview) and benchmarks (Math500, GSM8K, LiveCodeBench) validate the effectiveness of SEAL, up to a 11% improvement in accuracy while reducing reasoning tokens by 11.8% to 50.4%. Our code is publicly available at https://github.com/VITA-Group/SEAL.
Authors: Nirvan Patil, Malhar Abhay Inamdar, Agnivo Gosai, Guruprasad Pathak, Anish Joshi, Aryan Sagavekar, Anish Joshirao, Raj Dandekar, Rajat Dandekar, Sreedath Panat
Abstract: Small Language Models (SLMs) offer efficient alternatives to LLMs for specific domains. The 2023 TinyStories study developed an English dataset that allows SLMs with 1 to 10 million parameters to produce coherent outputs. Our research expands this framework by translating the original dataset into Indian languages and creating synthetic data using LLMs. We focus on Hindi, Marathi, and Bengali, evaluating SLMs for regional language processing and understanding linguistic complexity. We show that SLMs efficiently process regional languages with significantly fewer parameters than LLMs, providing a complementary framework for ``inference based evaluation" of tokenization strategies and linguistic complexity. Our analysis shows that language-specific tokenizers outperform general-purpose ones for Indian languages. Empirical validations, supported by information-theoretic and morphological analyses, provides fundamental understanding behind the better performance of Hindi models over Marathi and Bengali. Additionally, we show that synthetic datasets outperform translated content for training SLMs. Correlation analyses reveal cross-linguistic patterns and language-specific relationships between creativity, grammatical precision, and narrative completeness. These findings advance both the practical application of SLMs to underserved languages and our theoretical understanding of neural language development.
Authors: Mohammed Mallik, Laurent Clavier, Davy P. Gaillot
Abstract: Understanding the spatial and temporal patterns of environmental exposure to radio-frequency electromagnetic fields (RF-EMF) is essential for conducting risk assessments. These assessments aim to explore potential connections between RF-EMF exposure and its effects on human health, as well as on wildlife and plant life. Existing research has used different machine learning tools for EMF exposure estimation; however, a comparative analysis of these techniques is required to better understand their performance for real-world datasets. In this work, we present both finite and infinite-width convolutional network-based methods to estimate and assess EMF exposure levels from 70 real-world sensors in Lille, France. A comparative analysis has been conducted to analyze the performance of the methods' execution time and estimation accuracy. To improve estimation accuracy for higher-resolution grids, we utilized a preconditioned gradient descent method for kernel estimation. Root Mean Square Error (RMSE) is used as the evaluation criterion for comparing the performance of these deep learning models.
Authors: Seth Drake
Abstract: Large language model (LLM)-driven AI systems may exhibit an inference failure mode we term `neural howlround,' a self-reinforcing cognitive loop where certain highly weighted inputs become dominant, leading to entrenched response patterns resistant to correction. This paper explores the mechanisms underlying this phenomenon, which is distinct from model collapse and biased salience weighting. We propose an attenuation-based correction mechanism that dynamically introduces counterbalancing adjustments and can restore adaptive reasoning, even in `locked-in' AI systems. Additionally, we discuss some other related effects arising from improperly managed reinforcement. Finally, we outline potential applications of this mitigation strategy for improving AI robustness in real-world decision-making tasks.
Authors: Samah Alkhuzaey, Floriana Grasso, Terry R. Payne, Valentina Tamma
Abstract: Ontology-based question generation is an important application of semantic-aware systems that enables the creation of large question banks for diverse learning environments. The effectiveness of these systems, both in terms of the calibre and cognitive difficulty of the resulting questions, depends heavily on the quality and modelling approach of the underlying ontologies, making it crucial to assess their fitness for this task. To date, there has been no comprehensive investigation into the specific ontology aspects or characteristics that affect the question generation process. Therefore, this paper proposes a set of requirements and task-specific metrics for evaluating the fitness of ontologies for question generation tasks in pedagogical settings. Using the ROMEO methodology, a structured framework for deriving task-specific metrics, an expert-based approach is employed to assess the performance of various ontologies in Automatic Question Generation (AQG) tasks, which is then evaluated over a set of ontologies. Our results demonstrate that ontology characteristics significantly impact the effectiveness of question generation, with different ontologies exhibiting varying performance levels. This highlights the importance of assessing ontology quality with respect to AQG tasks.
Authors: Jinming Lu, Minghao She, Wendong Mao, Zhongfeng Wang
Abstract: Fine-tuning large diffusion models for custom applications demands substantial power and time, which poses significant challenges for efficient implementation on mobile devices. In this paper, we develop a novel training accelerator specifically for Low-Rank Adaptation (LoRA) of diffusion models, aiming to streamline the process and reduce computational complexity. By leveraging a fully quantized training scheme for LoRA fine-tuning, we achieve substantial reductions in memory usage and power consumption while maintaining high model fidelity. The proposed accelerator features flexible dataflow, enabling high utilization for irregular and variable tensor shapes during the LoRA process. Experimental results show up to 1.81x training speedup and 5.50x energy efficiency improvements compared to the baseline, with minimal impact on image generation quality.
Authors: Qi Bi, Jingjun Yi, Hao Zheng, Haolan Zhan, Wei Ji, Yawen Huang, Yuexiang Li
Abstract: Domain generalization aims to learn a representation from the source domain, which can be generalized to arbitrary unseen target domains. A fundamental challenge for visual domain generalization is the domain gap caused by the dramatic style variation whereas the image content is stable. The realm of selective state space, exemplified by VMamba, demonstrates its global receptive field in representing the content. However, the way exploiting the domain-invariant property for selective state space is rarely explored. In this paper, we propose a novel Flow Factorized State Space model, dubbed as DG-Famba, for visual domain generalization. To maintain domain consistency, we innovatively map the style-augmented and the original state embeddings by flow factorization. In this latent flow space, each state embedding from a certain style is specified by a latent probability path. By aligning these probability paths in the latent space, the state embeddings are able to represent the same content distribution regardless of the style differences. Extensive experiments conducted on various visual domain generalization settings show its state-of-the-art performance.
Authors: Qi Bi, Jingjun Yi, Haolan Zhan, Wei Ji, Gui-Song Xia
Abstract: Fine-grained domain generalization (FGDG) aims to learn a fine-grained representation that can be well generalized to unseen target domains when only trained on the source domain data. Compared with generic domain generalization, FGDG is particularly challenging in that the fine-grained category can be only discerned by some subtle and tiny patterns. Such patterns are particularly fragile under the cross-domain style shifts caused by illumination, color and etc. To push this frontier, this paper presents a novel Hyperbolic State Space Hallucination (HSSH) method. It consists of two key components, namely, state space hallucination (SSH) and hyperbolic manifold consistency (HMC). SSH enriches the style diversity for the state embeddings by firstly extrapolating and then hallucinating the source images. Then, the pre- and post- style hallucinate state embeddings are projected into the hyperbolic manifold. The hyperbolic state space models the high-order statistics, and allows a better discernment of the fine-grained patterns. Finally, the hyperbolic distance is minimized, so that the impact of style variation on fine-grained patterns can be eliminated. Experiments on three FGDG benchmarks demonstrate its state-of-the-art performance.
Authors: Akram Mustafa, Usman Naseem, Mostafa Rahimi Azghadi
Abstract: Clinical document classification is essential for converting unstructured medical texts into standardised ICD-10 diagnoses, yet it faces challenges due to complex medical language, privacy constraints, and limited annotated datasets. Large Language Models (LLMs) offer promising improvements in accuracy and efficiency for this task. This study evaluates the performance and consistency of eight LLMs; four reasoning (Qwen QWQ, Deepseek Reasoner, GPT o3 Mini, Gemini 2.0 Flash Thinking) and four non-reasoning (Llama 3.3, GPT 4o Mini, Gemini 2.0 Flash, Deepseek Chat); in classifying clinical discharge summaries using the MIMIC-IV dataset. Using cTAKES to structure clinical narratives, models were assessed across three experimental runs, with majority voting determining final predictions. Results showed that reasoning models outperformed non-reasoning models in accuracy (71% vs 68%) and F1 score (67% vs 60%), with Gemini 2.0 Flash Thinking achieving the highest accuracy (75%) and F1 score (76%). However, non-reasoning models demonstrated greater stability (91% vs 84% consistency). Performance varied across ICD-10 codes, with reasoning models excelling in complex cases but struggling with abstract categories. Findings indicate a trade-off between accuracy and consistency, suggesting that a hybrid approach could optimise clinical coding. Future research should explore multi-label classification, domain-specific fine-tuning, and ensemble methods to enhance model reliability in real-world applications.
Authors: Tony Shen, Seonghwan Seo, Ross Irwin, Kieran Didi, Simon Olsson, Woo Youn Kim, Martin Ester
Abstract: Many generative applications, such as synthesis-based 3D molecular design, involve constructing compositional objects with continuous features. Here, we introduce Compositional Generative Flows (CGFlow), a novel framework that extends flow matching to generate objects in compositional steps while modeling continuous states. Our key insight is that modeling compositional state transitions can be formulated as a straightforward extension of the flow matching interpolation process. We further build upon the theoretical foundations of generative flow networks (GFlowNets), enabling reward-guided sampling of compositional structures. We apply CGFlow to synthesizable drug design by jointly designing the molecule's synthetic pathway with its 3D binding pose. Our approach achieves state-of-the-art binding affinity on all 15 targets from the LIT-PCBA benchmark, and 5.8$\times$ improvement in sampling efficiency compared to 2D synthesis-based baseline. To our best knowledge, our method is also the first to achieve state of-art-performance in both Vina Dock (-9.38) and AiZynth success rate (62.2\%) on the CrossDocked benchmark.
Authors: Meilun Zhou, Aditya Dutt, Alina Zare
Abstract: Triplet loss traditionally relies only on class labels and does not use all available information in multi-task scenarios where multiple types of annotations are available. This paper introduces a Multi-Annotation Triplet Loss (MATL) framework that extends triplet loss by incorporating additional annotations, such as bounding box information, alongside class labels in the loss formulation. By using these complementary annotations, MATL improves multi-task learning for tasks requiring both classification and localization. Experiments on an aerial wildlife imagery dataset demonstrate that MATL outperforms conventional triplet loss in both classification and localization. These findings highlight the benefit of using all available annotations for triplet loss in multi-task learning frameworks.
Authors: Constantinos Tsakonas, Konstantinos Chatzilygeroudis
Abstract: Quality-Diversity algorithms have transformed optimization by prioritizing the discovery of diverse, high-performing solutions over a single optimal result. However, traditional Quality-Diversity methods, such as MAP-Elites, rely heavily on predefined behavioral descriptors and complete prior knowledge of the task to define the behavioral space grid, limiting their flexibility and applicability. In this work, we introduce Vector Quantized-Elites (VQ-Elites), a novel Quality-Diversity algorithm that autonomously constructs a structured behavioral space grid using unsupervised learning, eliminating the need for prior task-specific knowledge. At the core of VQ-Elites is the integration of Vector Quantized Variational Autoencoders, which enables the dynamic learning of behavioral descriptors and the generation of a structured, rather than unstructured, behavioral space grid - a significant advancement over existing unsupervised Quality-Diversity approaches. This design establishes VQ-Elites as a flexible, robust, and task-agnostic optimization framework. To further enhance the performance of unsupervised Quality-Diversity algorithms, we introduce two key components: behavioral space bounding and cooperation mechanisms, which significantly improve convergence and performance. We validate VQ-Elites on robotic arm pose-reaching and mobile robot space-covering tasks. The results demonstrate its ability to efficiently generate diverse, high-quality solutions, emphasizing its adaptability, scalability, robustness to hyperparameters, and potential to extend Quality-Diversity optimization to complex, previously inaccessible domains.
Authors: Kai Hu, Zhidan Zhao, Zhifeng Hao
Abstract: Traffic data exhibits complex temporal, spatial, and spatial-temporal correlations. Most of models use either independent modules to separately extract temporal and spatial correlations or joint modules to synchronously extract them, without considering the spatial-temporal correlations. Moreover, models that consider joint spatial-temporal correlations (temporal, spatial, and spatial-temporal correlations) often encounter significant challenges in accuracy and computational efficiency which prevent such models from demonstrating the expected advantages of a joint spatial-temporal correlations architecture. To address these issues, this paper proposes an efficient pure convolutional network for traffic prediction via spatial-temporal encoding and inferring (STEI-PCN). The model introduces and designs a dynamic adjacency matrix inferring module based on absolute spatial and temporal coordinates, as well as relative spatial and temporal distance encoding, using a graph convolutional network combined with gating mechanism to capture local synchronous joint spatial-temporal correlations. Additionally, three layers of temporal dilated causal convolutional network are used to capture long-range temporal correlations. Finally, through multi-view collaborative prediction module, the model integrates the gated-activated original, local synchronous joint spatial-temporal, and long-range temporal features to achieve comprehensive prediction. This study conducts extensive experiments on flow datasets (PeMS03/04/07/08) and speed dataset (PeMS-Bay), covering multiple prediction horizons. The results show that STEI-PCN demonstrates competitive computational efficiency in both training and inference speeds, and achieves superior or slightly inferior to state-of-the-art (SOTA) models on most evaluation metrics.
Authors: Rohola Zandie, Farhan Khodaee, Yufan Xia, Elazer R. Edelman
Abstract: Cellular development follows a stochastic yet rule-governed trajectory, though the underlying principles remain elusive. Here, we propose that cellular development follows paths of least action, aligning with foundational physical laws that govern dynamic systems across nature. We introduce a computational framework that takes advantage of the deep connection between the principle of least action and maximum entropy to model developmental processes using Transformers architecture. This approach enables precise quantification of entropy production, information flow curvature, and local irreversibility for developmental asymmetry in single-cell RNA sequence data. Within this unified framework, we provide interpretable metrics: entropy to capture exploration-exploitation trade-offs, curvature to assess plasticity-elasticity dynamics, and entropy production to characterize dedifferentiation and transdifferentiation. We validate our method across both single-cell and embryonic development datasets, demonstrating its ability to reveal hidden thermodynamic and informational constraints shaping cellular fate decisions.
Authors: Ingryd V. S. T. Pereira, George D. C. Cavalcanti, Rafael M. O. Cruz
Abstract: Given the volume and speed at which fake news spreads across social media, automatic fake news detection has become a highly important task. However, this task presents several challenges, including extracting textual features that contain relevant information about fake news. Research about fake news detection shows that no single feature extraction technique consistently outperforms the others across all scenarios. Nevertheless, different feature extraction techniques can provide complementary information about the textual data and enable a more comprehensive representation of the content. This paper proposes using multi-view autoencoders to generate a joint feature representation for fake news detection by integrating several feature extraction techniques commonly used in the literature. Experiments on fake news datasets show a significant improvement in classification performance compared to individual views (feature representations). We also observed that selecting a subset of the views instead of composing a latent space with all the views can be advantageous in terms of accuracy and computational effort. For further details, including source codes, figures, and datasets, please refer to the project's repository: https://github.com/ingrydpereira/multiview-fake-news.
Authors: Tianyi Wu, Zhiwei Xue, Yue Liu, Jiaheng Zhang, Bryan Hooi, See-Kiong Ng
Abstract: Jailbreak attacks, which aim to cause LLMs to perform unrestricted behaviors, have become a critical and challenging direction in AI safety. Despite achieving the promising attack success rate using dictionary-based evaluation, existing jailbreak attack methods fail to output detailed contents to satisfy the harmful request, leading to poor performance on GPT-based evaluation. To this end, we propose a black-box jailbreak attack termed GeneShift, by using a genetic algorithm to optimize the scenario shifts. Firstly, we observe that the malicious queries perform optimally under different scenario shifts. Based on it, we develop a genetic algorithm to evolve and select the hybrid of scenario shifts. It guides our method to elicit detailed and actionable harmful responses while keeping the seemingly benign facade, improving stealthiness. Extensive experiments demonstrate the superiority of GeneShift. Notably, GeneShift increases the jailbreak success rate from 0% to 60% when direct prompting alone would fail.
Authors: Lucian Chauvina, Somil Guptac, Angelina Ibarrac, Joshua Peeples
Abstract: Anomaly detection is a key research challenge in computer vision and machine learning with applications in many fields from quality control to radar imaging. In radar imaging, specifically synthetic aperture radar (SAR), anomaly detection can be used for the classification, detection, and segmentation of objects of interest. However, there is no method for developing and benchmarking these methods on SAR imagery. To address this issue, we introduce SAR imagery anomaly detection (SARIAD). In conjunction with Anomalib, a deep-learning library for anomaly detection, SARIAD provides a comprehensive suite of algorithms and datasets for assessing and developing anomaly detection approaches on SAR imagery. SARIAD specifically integrates multiple SAR datasets along with tools to effectively apply various anomaly detection algorithms to SAR imagery. Several anomaly detection metrics and visualizations are available. Overall, SARIAD acts as a central package for benchmarking SAR models and datasets to allow for reproducible research in the field of anomaly detection in SAR imagery. This package is publicly available: https://github.com/Advanced-Vision-and-Learning-Lab/SARIAD.
URLs: https://github.com/Advanced-Vision-and-Learning-Lab/SARIAD.
Authors: Michael Bowling, Esraa Elelimy
Abstract: Algorithms and approaches for continual reinforcement learning have gained increasing attention. Much of this early progress rests on the foundations and standard practices of traditional reinforcement learning, without questioning if they are well-suited to the challenges of continual learning agents. We suggest that many core foundations of traditional RL are, in fact, antithetical to the goals of continual reinforcement learning. We enumerate four such foundations: the Markov decision process formalism, a focus on optimal policies, the expected sum of rewards as the primary evaluation metric, and episodic benchmark environments that embrace the other three foundations. Shedding such sacredly held and taught concepts is not easy. They are self-reinforcing in that each foundation depends upon and holds up the others, making it hard to rethink each in isolation. We propose an alternative set of all four foundations that are better suited to the continual learning setting. We hope to spur on others in rethinking the traditional foundations, proposing and critiquing alternatives, and developing new algorithms and approaches enabled by better-suited foundations.
Authors: Jinfeng Zhuang, Yinrui Li, Runze Su, Ke Xu, Zhixuan Shao, Kungang Li, Ling Leng, Han Sun, Meng Qi, Yixiong Meng, Yang Tang, Zhifang Liu, Qifei Shen, Aayush Mudgal
Abstract: The predictions of click through rate (CTR) and conversion rate (CVR) play a crucial role in the success of ad-recommendation systems. A Deep Hierarchical Ensemble Network (DHEN) has been proposed to integrate multiple feature crossing modules and has achieved great success in CTR prediction. However, its performance for CVR prediction is unclear in the conversion ads setting, where an ad bids for the probability of a user's off-site actions on a third party website or app, including purchase, add to cart, sign up, etc. A few challenges in DHEN: 1) What feature-crossing modules (MLP, DCN, Transformer, to name a few) should be included in DHEN? 2) How deep and wide should DHEN be to achieve the best trade-off between efficiency and efficacy? 3) What hyper-parameters to choose in each feature-crossing module? Orthogonal to the model architecture, the input personalization features also significantly impact model performance with a high degree of freedom. In this paper, we attack this problem and present our contributions biased to the applied data science side, including: First, we propose a multitask learning framework with DHEN as the single backbone model architecture to predict all CVR tasks, with a detailed study on how to make DHEN work effectively in practice; Second, we build both on-site real-time user behavior sequences and off-site conversion event sequences for CVR prediction purposes, and conduct ablation study on its importance; Last but not least, we propose a self-supervised auxiliary loss to predict future actions in the input sequence, to help resolve the label sparseness issue in CVR prediction. Our method achieves state-of-the-art performance compared to previous single feature crossing modules with pre-trained user personalization features.
Authors: Sourya Sengupta, Satrajit Chakrabarty, Keerthi Sravan Ravi, Gopal Avinash, Ravi Soni
Abstract: Foundation models like the Segment Anything Model (SAM) excel in zero-shot segmentation for natural images but struggle with medical image segmentation due to differences in texture, contrast, and noise. Annotating medical images is costly and requires domain expertise, limiting large-scale annotated data availability. To address this, we propose SynthFM, a synthetic data generation framework that mimics the complexities of medical images, enabling foundation models to adapt without real medical data. Using SAM's pretrained encoder and training the decoder from scratch on SynthFM's dataset, we evaluated our method on 11 anatomical structures across 9 datasets (CT, MRI, and Ultrasound). SynthFM outperformed zero-shot baselines like SAM and MedSAM, achieving superior results under different prompt settings and on out-of-distribution datasets.
Authors: Ruineng Li, Daitao Xing, Huiming Sun, Yuanzhou Ha, Jinglin Shen, Chiuman Ho
Abstract: Human-centric motion control in video generation remains a critical challenge, particularly when jointly controlling camera movements and human poses in scenarios like the iconic Grammy Glambot moment. While recent video diffusion models have made significant progress, existing approaches struggle with limited motion representations and inadequate integration of camera and human motion controls. In this work, we present TokenMotion, the first DiT-based video diffusion framework that enables fine-grained control over camera motion, human motion, and their joint interaction. We represent camera trajectories and human poses as spatio-temporal tokens to enable local control granularity. Our approach introduces a unified modeling framework utilizing a decouple-and-fuse strategy, bridged by a human-aware dynamic mask that effectively handles the spatially-and-temporally varying nature of combined motion signals. Through extensive experiments, we demonstrate TokenMotion's effectiveness across both text-to-video and image-to-video paradigms, consistently outperforming current state-of-the-art methods in human-centric motion control tasks. Our work represents a significant advancement in controllable video generation, with particular relevance for creative production applications.
Authors: Aashiq Muhamed, Jacopo Bonato, Mona Diab, Virginia Smith
Abstract: Machine unlearning is a promising approach to improve LLM safety by removing unwanted knowledge from the model. However, prevailing gradient-based unlearning methods suffer from issues such as high computational costs, hyperparameter instability, poor sequential unlearning capability, vulnerability to relearning attacks, low data efficiency, and lack of interpretability. While Sparse Autoencoders are well-suited to improve these aspects by enabling targeted activation-based unlearning, prior approaches underperform gradient-based methods. This work demonstrates that, contrary to these earlier findings, SAEs can significantly improve unlearning when employed dynamically. We introduce $\textbf{Dynamic DAE Guardrails}$ (DSG), a novel method for precision unlearning that leverages principled feature selection and a dynamic classifier. Our experiments show DSG substantially outperforms leading unlearning methods, achieving superior forget-utility trade-offs. DSG addresses key drawbacks of gradient-based approaches for unlearning -- offering enhanced computational efficiency and stability, robust performance in sequential unlearning, stronger resistance to relearning attacks, better data efficiency including zero-shot settings, and more interpretable unlearning.
Authors: Michael Elrod, Niloufar Mehrabi, Rahul Amin, Manveen Kaur, Long Cheng, Jim Martin, Abolfazl Razi
Abstract: Mission planning for a fleet of cooperative autonomous drones in applications that involve serving distributed target points, such as disaster response, environmental monitoring, and surveillance, is challenging, especially under partial observability, limited communication range, and uncertain environments. Traditional path-planning algorithms struggle in these scenarios, particularly when prior information is not available. To address these challenges, we propose a novel framework that integrates Graph Neural Networks (GNNs), Deep Reinforcement Learning (DRL), and transformer-based mechanisms for enhanced multi-agent coordination and collective task execution. Our approach leverages GNNs to model agent-agent and agent-goal interactions through adaptive graph construction, enabling efficient information aggregation and decision-making under constrained communication. A transformer-based message-passing mechanism, augmented with edge-feature-enhanced attention, captures complex interaction patterns, while a Double Deep Q-Network (Double DQN) with prioritized experience replay optimizes agent policies in partially observable environments. This integration is carefully designed to address specific requirements of multi-agent navigation, such as scalability, adaptability, and efficient task execution. Experimental results demonstrate superior performance, with 90% service provisioning and 100% grid coverage (node discovery), while reducing the average steps per episode to 200, compared to 600 for benchmark methods such as particle swarm optimization (PSO), greedy algorithms and DQN.
Authors: Ryoma Sato, Shinji Ito
Abstract: While classical formulations of multi-armed bandit problems assume that each arm's reward is independent and stationary, real-world applications often involve non-stationary environments and interdependencies between arms. In particular, selecting one arm may influence the future rewards of other arms, a scenario not adequately captured by existing models such as rotting bandits or restless bandits. To address this limitation, we propose the influential bandit problem, which models inter-arm interactions through an unknown, symmetric, positive semi-definite interaction matrix that governs the dynamics of arm losses. We formally define this problem and establish two regret lower bounds, including a superlinear $\Omega(T^2 / \log^2 T)$ bound for the standard UCB algorithm and an algorithm-independent $\Omega(T)$ bound, which highlight the inherent difficulty of the setting. We then introduce a new algorithm based on a lower confidence bound (LCB) estimator tailored to the structure of the loss dynamics. Under mild assumptions, our algorithm achieves a regret of $O(KT \log T)$, which is nearly optimal in terms of its dependence on the time horizon. The algorithm is simple to implement and computationally efficient. Empirical evaluations on both synthetic and real-world datasets demonstrate the presence of inter-arm influence and confirm the superior performance of our method compared to conventional bandit algorithms.
Authors: Yizi Zhang, Yanchen Wang, Mehdi Azabou, Alexandre Andre, Zixuan Wang, Hanrui Lyu, The International Brain Laboratory, Eva Dyer, Liam Paninski, Cole Hurwitz
Abstract: Recent work has demonstrated that large-scale, multi-animal models are powerful tools for characterizing the relationship between neural activity and behavior. Current large-scale approaches, however, focus exclusively on either predicting neural activity from behavior (encoding) or predicting behavior from neural activity (decoding), limiting their ability to capture the bidirectional relationship between neural activity and behavior. To bridge this gap, we introduce a multimodal, multi-task model that enables simultaneous Neural Encoding and Decoding at Scale (NEDS). Central to our approach is a novel multi-task-masking strategy, which alternates between neural, behavioral, within-modality, and cross-modality masking. We pretrain our method on the International Brain Laboratory (IBL) repeated site dataset, which includes recordings from 83 animals performing the same visual decision-making task. In comparison to other large-scale models, we demonstrate that NEDS achieves state-of-the-art performance for both encoding and decoding when pretrained on multi-animal data and then fine-tuned on new animals. Surprisingly, NEDS's learned embeddings exhibit emergent properties: even without explicit training, they are highly predictive of the brain regions in each recording. Altogether, our approach is a step towards a foundation model of the brain that enables seamless translation between neural activity and behavior.
Authors: Rudra Dhar, Adyansh Kakran, Amey Karan, Karthik Vaidhyanathan, Vasudeva Varma
Abstract: Architectural Knowledge Management (AKM) is crucial for software development but remains challenging due to the lack of standardization and high manual effort. Architecture Decision Records (ADRs) provide a structured approach to capture Architecture Design Decisions (ADDs), but their adoption is limited due to the manual effort involved and insufficient tool support. Our previous work has shown that Large Language Models (LLMs) can assist in generating ADDs. However, simply prompting the LLM does not produce quality ADDs. Moreover, using third-party LLMs raises privacy concerns, while self-hosting them poses resource challenges. To this end, we experimented with different approaches like few-shot, retrieval-augmented generation (RAG) and fine-tuning to enhance LLM's ability to generate ADDs. Our results show that both techniques improve effectiveness. Building on this, we propose Domain Specific Retreival Augumented Few Shot Fine Tuninng, DRAFT, which combines the strengths of all these three approaches for more effective ADD generation. DRAFT operates in two phases: an offline phase that fine-tunes an LLM on generating ADDs augmented with retrieved examples and an online phase that generates ADDs by leveraging retrieved ADRs and the fine-tuned model. We evaluated DRAFT against existing approaches on a dataset of 4,911 ADRs and various LLMs and analyzed them using automated metrics and human evaluations. Results show DRAFT outperforms all other approaches in effectiveness while maintaining efficiency. Our findings indicate that DRAFT can aid architects in drafting ADDs while addressing privacy and resource constraints.
Authors: Boxuan Ma, Md Akib Zabed Khan, Tianyuan Yang, Agoritsa Polyzou, Shin'ichi Konomi
Abstract: Large Language Models (LLMs) have made significant strides in natural language processing and are increasingly being integrated into recommendation systems. However, their potential in educational recommendation systems has yet to be fully explored. This paper investigates the use of LLMs as a general-purpose recommendation model, leveraging their vast knowledge derived from large-scale corpora for course recommendation tasks. We explore a variety of approaches, ranging from prompt-based methods to more advanced fine-tuning techniques, and compare their performance against traditional recommendation models. Extensive experiments were conducted on a real-world MOOC dataset, evaluating using LLMs as course recommendation systems across key dimensions such as accuracy, diversity, and novelty. Our results demonstrate that LLMs can achieve good performance comparable to traditional models, highlighting their potential to enhance educational recommendation systems. These findings pave the way for further exploration and development of LLM-based approaches in the context of educational recommendations.
Authors: Erica van der Sar, Alessandro Zocca, Sandjai Bhulai
Abstract: Power grid operation is becoming increasingly complex due to the rising integration of renewable energy sources and the need for more adaptive control strategies. Reinforcement Learning (RL) has emerged as a promising approach to power network control (PNC), offering the potential to enhance decision-making in dynamic and uncertain environments. The Learning To Run a Power Network (L2RPN) competitions have played a key role in accelerating research by providing standardized benchmarks and problem formulations, leading to rapid advancements in RL-based methods. This survey provides a comprehensive and structured overview of RL applications for power grid topology optimization, categorizing existing techniques, highlighting key design choices, and identifying gaps in current research. Additionally, we present a comparative numerical study evaluating the impact of commonly applied RL-based methods, offering insights into their practical effectiveness. By consolidating existing research and outlining open challenges, this survey aims to provide a foundation for future advancements in RL-driven power grid optimization.
Authors: Leo Kampen, Carlos Rabat Villarreal, Louis Yu, Santu Karmaker, Dongji Feng
Abstract: In this paper, we conducted a Multi-Perspective Comparative Narrative Analysis (CNA) on three prominent LLMs: GPT-3.5, PaLM2, and Llama2. We applied identical prompts and evaluated their outputs on specific tasks, ensuring an equitable and unbiased comparison between various LLMs. Our study revealed that the three LLMs generated divergent responses to the same prompt, indicating notable discrepancies in their ability to comprehend and analyze the given task. Human evaluation was used as the gold standard, evaluating four perspectives to analyze differences in LLM performance.
Authors: Zhaoyu Liu, Kan Jiang, Murong Ma, Zhe Hou, Yun Lin, Jin Song Dong
Abstract: Analyzing Fast, Frequent, and Fine-grained (F$^3$) events presents a significant challenge in video analytics and multi-modal LLMs. Current methods struggle to identify events that satisfy all the F$^3$ criteria with high accuracy due to challenges such as motion blur and subtle visual discrepancies. To advance research in video understanding, we introduce F$^3$Set, a benchmark that consists of video datasets for precise F$^3$ event detection. Datasets in F$^3$Set are characterized by their extensive scale and comprehensive detail, usually encompassing over 1,000 event types with precise timestamps and supporting multi-level granularity. Currently, F$^3$Set contains several sports datasets, and this framework may be extended to other applications as well. We evaluated popular temporal action understanding methods on F$^3$Set, revealing substantial challenges for existing techniques. Additionally, we propose a new method, F$^3$ED, for F$^3$ event detections, achieving superior performance. The dataset, model, and benchmark code are available at https://github.com/F3Set/F3Set.
Authors: Shengyuan Ye, Bei Ouyang, Liekang Zeng, Tianyi Qian, Xiaowen Chu, Jian Tang, Xu Chen
Abstract: Generative large language models (LLMs) have garnered significant attention due to their exceptional capabilities in various AI tasks. Traditionally deployed in cloud datacenters, LLMs are now increasingly moving towards more accessible edge platforms to protect sensitive user data and ensure privacy preservation. The limited computational resources of individual edge devices, however, can result in excessively prolonged inference latency and overwhelmed memory usage. While existing research has explored collaborative edge computing to break the resource wall of individual devices, these solutions yet suffer from massive communication overhead and under-utilization of edge resources. Furthermore, they focus exclusively on optimizing the prefill phase, neglecting the crucial autoregressive decoding phase for generative LLMs. To address that, we propose Jupiter, a fast, scalable, and resource-efficient collaborative edge AI system for generative LLM inference. Jupiter introduces a flexible pipelined architecture as a principle and differentiates its system design according to the differentiated characteristics of the prefill and decoding phases. For prefill phase, Jupiter submits a novel intra-sequence pipeline parallelism and develops a meticulous parallelism planning strategy to maximize resource efficiency; For decoding, Jupiter devises an effective outline-based pipeline parallel decoding mechanism combined with speculative decoding, which further magnifies inference acceleration. Extensive evaluation based on realistic implementation demonstrates that Jupiter remarkably outperforms state-of-the-art approaches under various edge environment setups, achieving up to 26.1x end-to-end latency reduction while rendering on-par generation quality.
Authors: Shiyi Ding, Ying Chen
Abstract: Recent advances in large language models (LLMs) provide new opportunities for context understanding in virtual reality (VR). However, VR contexts are often highly localized and personalized, limiting the effectiveness of general-purpose LLMs. To address this challenge, we present RAG-VR, the first 3D question-answering system for VR that incorporates retrieval-augmented generation (RAG), which augments an LLM with external knowledge retrieved from a localized knowledge database to improve the answer quality. RAG-VR includes a pipeline for extracting comprehensive knowledge about virtual environments and user conditions for accurate answer generation. To ensure efficient retrieval, RAG-VR offloads the retrieval process to a nearby edge server and uses only essential information during retrieval. Moreover, we train the retriever to effectively distinguish among relevant, irrelevant, and hard-to-differentiate information in relation to questions. RAG-VR improves answer accuracy by 17.9%-41.8% and reduces end-to-end latency by 34.5%-47.3% compared with two baseline systems.
Authors: Yingqian Xu, Xiaohan Li, Caihua Wan, Ran Zhang, Bin He, Shiqiang Liu, Jihao Xia, Dehao Kong, Shilong Xiong, Guoqiang Yu, Xiufeng Han
Abstract: Bayesian networks play an increasingly important role in data mining, inference, and reasoning with the rapid development of artificial intelligence. In this paper, we present proof-of-concept experiments demonstrating the use of spin-orbit torque magnetic tunnel junctions (SOT-MTJs) in Bayesian network reasoning. Not only can the target probability distribution function (PDF) of a Bayesian network be precisely formulated by a conditional probability table as usual but also quantitatively parameterized by a probabilistic forward propagating neuron network. Moreover, the parameters of the network can also approach the optimum through a simple point-by point training algorithm, by leveraging which we do not need to memorize all historical data nor statistically summarize conditional probabilities behind them, significantly improving storage efficiency and economizing data pretreatment. Furthermore, we developed a simple medical diagnostic system using the SOT-MTJ as a random number generator and sampler, showcasing the application of SOT-MTJ-based Bayesian reasoning. This SOT-MTJ-based Bayesian reasoning shows great promise in the field of artificial probabilistic neural network, broadening the scope of spintronic device applications and providing an efficient and low-storage solution for complex reasoning tasks.
Authors: Yuxuan Zeng, Wenhao Xie, Wei Cao, Tan Peng, Yue Hou, Ziyu Wang, Jing Shi
Abstract: The thermoelectric performance of materials exhibits complex nonlinear dependencies on both elemental types and their proportions, rendering traditional trial-and-error approaches inefficient and time-consuming for material discovery. In this work, we present a deep learning model capable of accurately predicting thermoelectric properties of doped materials directly from their chemical formulas, achieving state-of-the-art performance. To enhance interpretability, we further incorporate sensitivity analysis techniques to elucidate how physical descriptors affect the thermoelectric figure of merit (zT). Moreover, we establish a coupled framework that integrates a surrogate model with a multi-objective genetic algorithm to efficiently explore the vast compositional space for high-performance candidates. Experimental validation confirms the discovery of a novel thermoelectric material with superior $zT$ values in the medium-temperature regime.
Authors: Ruohao Zhan, Yijin Li, Yisheng He, Shuo Chen, Yichen Shen, Xinyu Chen, Zilong Dong, Zhaoyang Huang, Guofeng Zhang
Abstract: Sketches serve as fundamental blueprints in artistic creation because sketch editing is easier and more intuitive than pixel-level RGB image editing for painting artists, yet sketch generation remains unexplored despite advancements in generative models. We propose a novel framework CoProSketch, providing prominent controllability and details for sketch generation with diffusion models. A straightforward method is fine-tuning a pretrained image generation diffusion model with binarized sketch images. However, we find that the diffusion models fail to generate clear binary images, which makes the produced sketches chaotic. We thus propose to represent the sketches by unsigned distance field (UDF), which is continuous and can be easily decoded to sketches through a lightweight network. With CoProSketch, users generate a rough sketch from a bounding box and a text prompt. The rough sketch can be manually edited and fed back into the model for iterative refinement and will be decoded to a detailed sketch as the final result. Additionally, we curate the first large-scale text-sketch paired dataset as the training data. Experiments demonstrate superior semantic consistency and controllability over baselines, offering a practical solution for integrating user feedback into generative workflows.
Authors: Vishal Gandhi, Sagar Gandhi
Abstract: Advancements in emotion aware language processing increasingly shape vital NLP applications ranging from conversational AI and affective computing to computational psychology and creative content generation. Existing emotion datasets either lack emotional granularity or fail to capture necessary stylistic diversity, limiting the advancement of effective emotion conditioned text generation systems. Seeking to bridge this crucial gap between granularity and style diversity, this paper introduces a novel systematically constructed dataset named ELSA Emotion and Language Style Alignment Dataset leveraging fine grained emotion taxonomies adapted from existing sources such as dair ai emotion dataset and GoEmotions taxonomy. This dataset comprises multiple emotionally nuanced variations of original sentences regenerated across distinct contextual styles such as conversational, formal, poetic, and narrative, using advanced Large Language Models LLMs. Rigorous computational evaluation using metrics such as perplexity, embedding variance, readability, lexical diversity, and semantic coherence measures validates the datasets emotional authenticity, linguistic fluency, and textual diversity. Comprehensive metric analyses affirm its potential to support deeper explorations into emotion conditioned style adaptive text generation. By enabling precision tuned emotionally nuanced language modeling, our dataset creates fertile ground for research on fine grained emotional control, prompt driven explanation, interpretability, and style adaptive expressive language generation with LLMs.
Authors: Yuyang Xu, Renjun Hu, Haochao Ying, Jian Wu, Xing Shi, Wei Lin
Abstract: Multiple-choice question (MCQ) benchmarks are widely used for evaluating Large Language Models (LLMs), yet their reliability is undermined by benchmark contamination. In this study, we reframe contamination as an inherent aspect of learning and seek to disentangle genuine capability acquisition from superficial memorization in LLM evaluation. First, by analyzing model performance under different memorization conditions, we uncover a counterintuitive trend: LLMs perform worse on memorized MCQs than on non-memorized ones, indicating the coexistence of two distinct learning phenomena, i.e., rote memorization and genuine capability learning. To disentangle them, we propose TrinEval, a novel evaluation framework that reformulates MCQs into an alternative trinity format, reducing memorization while preserving knowledge assessment. Experiments validate TrinEval's effectiveness in reformulation, and its evaluation reveals that common LLMs may memorize by rote 20.5% of knowledge points (in MMLU on average).
Authors: Steffen Herbold
Abstract: Sorting is a tedious but simple task for human intelligence and can be solved fairly easily algorithmically. However, for Large Language Models (LLMs) this task is surprisingly hard, as some properties of sorting are among known weaknesses of LLMs: being faithful to the input data, logical comparisons between values, and strictly differentiating between syntax (used for sorting) and semantics (typically learned by embeddings). Within this paper, we describe the new SortBench benchmark for LLMs that comes with different difficulties and that can be easily scaled in terms of difficulty. We apply this benchmark to seven state-of-the-art LLMs, including current test-time reasoning models. Our results show that while the o3-mini model is very capable at sorting in general, even this can be fooled if strings are defined to mix syntactical and semantical aspects, e.g., by asking to sort numbers written-out as word. Furthermore, all models have problems with the faithfulness to the input of long lists, i.e., they drop items and add new ones. Our results also show that test-time reasoning has a tendency to overthink problems which leads to performance degradation. Finally, models without test-time reasoning like GPT-4o are not much worse than reasoning models.
Authors: Lucia Celli, Giovanni Peccati
Abstract: Using entropic inequalities from information theory, we provide new bounds on the total variation and 2-Wasserstein distances between a conditionally Gaussian law and a Gaussian law with invertible covariance matrix. We apply our results to quantify the speed of convergence to Gaussian of a randomly initialized fully connected neural network and its derivatives - evaluated in a finite number of inputs - when the initialization is Gaussian and the sizes of the inner layers diverge to infinity. Our results require mild assumptions on the activation function, and allow one to recover optimal rates of convergence in a variety of distances, thus improving and extending the findings of Basteri and Trevisan (2023), Favaro et al. (2023), Trevisan (2024) and Apollonio et al. (2024). One of our main tools are the quantitative cumulant estimates established in Hanin (2024). As an illustration, we apply our results to bound the total variation distance between the Bayesian posterior law of the neural network and its derivatives, and the posterior law of the corresponding Gaussian limit: this yields quantitative versions of a posterior CLT by Hron et al. (2022), and extends several estimates by Trevisan (2024) to the total variation metric.
Authors: Hoang-Loc La, Phuong Hoai Ha
Abstract: Many studies estimate energy consumption using proxy metrics like memory usage, FLOPs, and inference latency, with the assumption that reducing these metrics will also lower energy consumption in neural networks. This paper, however, takes a different approach by introducing an energy-efficient Neural Architecture Search (NAS) method that directly focuses on identifying architectures that minimize energy consumption while maintaining acceptable accuracy. Unlike previous methods that primarily target vision and language tasks, the approach proposed here specifically addresses tabular datasets. Remarkably, the optimal architecture suggested by this method can reduce energy consumption by up to 92% compared to architectures recommended by conventional NAS.
Authors: Yucheng Liu, Longyu Jiang
Abstract: Signal separation in the passive underwater acoustic domain has heavily relied on deep learning techniques to isolate ship radiated noise. However, the separation networks commonly used in this domain stem from speech separation applications and may not fully consider the unique aspects of underwater acoustics beforehand, such as the influence of different propagation media, signal frequencies and modulation characteristics. This oversight highlights the need for tailored approaches that account for the specific characteristics of underwater sound propagation. This study introduces a novel temporal network designed to separate ship radiated noise by employing a dual-path model and a feature decoupling approach. The mixed signals' features are transformed into a space where they exhibit greater independence, with each dimension's significance decoupled. Subsequently, a fusion of local and global attention mechanisms is employed in the separation layer. Extensive comparisons showcase the effectiveness of this method when compared to other prevalent network models, as evidenced by its performance in the ShipsEar and DeepShip datasets.
Authors: Markus Flicke, Glenn Angrabeit, Madhav Iyengar, Vitalii Protsenko, Illia Shakun, Jovan Cicvaric, Bora Kargi, Haoyu He, Lukas Schuler, Lewin Scholz, Kavyanjali Agnihotri, Yong Cao, Andreas Geiger
Abstract: Scholar Inbox is a new open-access platform designed to address the challenges researchers face in staying current with the rapidly expanding volume of scientific literature. We provide personalized recommendations, continuous updates from open-access archives (arXiv, bioRxiv, etc.), visual paper summaries, semantic search, and a range of tools to streamline research workflows and promote open research access. The platform's personalized recommendation system is trained on user ratings, ensuring that recommendations are tailored to individual researchers' interests. To further enhance the user experience, Scholar Inbox also offers a map of science that provides an overview of research across domains, enabling users to easily explore specific topics. We use this map to address the cold start problem common in recommender systems, as well as an active learning strategy that iteratively prompts users to rate a selection of papers, allowing the system to learn user preferences quickly. We evaluate the quality of our recommendation system on a novel dataset of 800k user ratings, which we make publicly available, as well as via an extensive user study. https://www.scholar-inbox.com/
Authors: Arman Khaledian, Amirreza Ghadiridehkordi, Nariman Khaledian
Abstract: Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm for grounding large language models in external knowledge sources, improving the precision of agents responses. However, high-dimensional language model embeddings, often in the range of hundreds to thousands of dimensions, can present scalability challenges in terms of storage and latency, especially when processing massive financial text corpora. This paper investigates the use of Principal Component Analysis (PCA) to reduce embedding dimensionality, thereby mitigating computational bottlenecks without incurring large accuracy losses. We experiment with a real-world dataset and compare different similarity and distance metrics under both full-dimensional and PCA-compressed embeddings. Our results show that reducing vectors from 3,072 to 110 dimensions provides a sizeable (up to $60\times$) speedup in retrieval operations and a $\sim 28.6\times$ reduction in index size, with only moderate declines in correlation metrics relative to human-annotated similarity scores. These findings demonstrate that PCA-based compression offers a viable balance between retrieval fidelity and resource efficiency, essential for real-time systems such as Zanista AI's \textit{Newswitch} platform. Ultimately, our study underscores the practicality of leveraging classical dimensionality reduction techniques to scale RAG architectures for knowledge-intensive applications in finance and trading, where speed, memory efficiency, and accuracy must jointly be optimized.
Authors: Junliang Guo, Yang Ye, Tianyu He, Haoyu Wu, Yushu Jiang, Tim Pearce, Jiang Bian
Abstract: World modeling is a crucial task for enabling intelligent agents to effectively interact with humans and operate in dynamic environments. In this work, we propose MineWorld, a real-time interactive world model on Minecraft, an open-ended sandbox game which has been utilized as a common testbed for world modeling. MineWorld is driven by a visual-action autoregressive Transformer, which takes paired game scenes and corresponding actions as input, and generates consequent new scenes following the actions. Specifically, by transforming visual game scenes and actions into discrete token ids with an image tokenizer and an action tokenizer correspondingly, we consist the model input with the concatenation of the two kinds of ids interleaved. The model is then trained with next token prediction to learn rich representations of game states as well as the conditions between states and actions simultaneously. In inference, we develop a novel parallel decoding algorithm that predicts the spatial redundant tokens in each frame at the same time, letting models in different scales generate $4$ to $7$ frames per second and enabling real-time interactions with game players. In evaluation, we propose new metrics to assess not only visual quality but also the action following capacity when generating new scenes, which is crucial for a world model. Our comprehensive evaluation shows the efficacy of MineWorld, outperforming SoTA open-sourced diffusion based world models significantly. The code and model have been released.
Authors: Pepita Barnard, Maria J Galvez Trigo, Dominic Price, Sue Cobb, Gisela Reyes-Cruz, Gustavo Berumen, David Branson III, Mojtaba A. Khanesar, Mercedes Torres Torres, Michel Valstar
Abstract: Mental models and expectations underlying human-human interaction (HHI) inform human-robot interaction (HRI) with domestic robots. To ease collaborative home tasks by improving domestic robot speech and behaviours for human-robot communication, we designed a study to understand how people communicated when failure occurs. To identify patterns of natural communication, particularly in response to robotic failures, participants instructed Laundrobot to move laundry into baskets using natural language and gestures. Laundrobot either worked error-free, or in one of two error modes. Participants were not advised Laundrobot would be a human actor, nor given information about error modes. Video analysis from 42 participants found speech patterns, included laughter, verbal expressions, and filler words, such as ``oh'' and ``ok'', also, sequences of body movements, including touching one's own face, increased pointing with a static finger, and expressions of surprise. Common strategies deployed when errors occurred, included correcting and teaching, taking responsibility, and displays of frustration. The strength of reaction to errors diminished with exposure, possibly indicating acceptance or resignation. Some used strategies similar to those used to communicate with other technologies, such as smart assistants. An anthropomorphic robot may not be ideally suited to this kind of task. Laundrobot's appearance, morphology, voice, capabilities, and recovery strategies may have impacted how it was perceived. Some participants indicated Laundrobot's actual skills were not aligned with expectations; this made it difficult to know what to expect and how much Laundrobot understood. Expertise, personality, and cultural differences may affect responses, however these were not assessed.
Authors: Yin Jou Huang, Rafik Hadfi
Abstract: There is a growing interest in assessing the personality traits of Large language models (LLMs). However, traditional personality assessments based on self-report questionnaires may fail to capture their true behavioral nuances due to inherent biases and meta-knowledge contamination. This paper introduces a novel multi-observer framework for LLM personality assessment that draws inspiration from informant-report methods in psychology. Instead of relying solely on self-assessments, our approach employs multiple observer agents configured with a specific relationship context (e.g., family, friend, or workplace) to simulate interactive scenarios with a subject LLM. These observers engage in dialogues and subsequently provide ratings across the Big Five personality dimensions. Our experiments reveal that LLMs possess systematic biases in self-report personality ratings. Moreover, aggregating observer ratings effectively reduces non-systematic biases and achieves optimal reliability with 5-7 observers. The findings highlight the significant impact of relationship context on personality perception and demonstrate that a multi-observer paradigm yields a more robust and context-sensitive evaluation of LLM personality traits.
Authors: Dawei Zhou, Suzhi Gang, Decheng Liu, Tongliang Liu, Nannan Wang, Xinbo Gao
Abstract: Malicious applications of visual manipulation have raised serious threats to the security and reputation of users in many fields. To alleviate these issues, adversarial noise-based defenses have been enthusiastically studied in recent years. However, ``data-only" methods tend to distort fake samples in the low-level feature space rather than the high-level semantic space, leading to limitations in resisting malicious manipulation. Frontier research has shown that integrating knowledge in deep learning can produce reliable and generalizable solutions. Inspired by these, we propose a knowledge-guided adversarial defense (KGAD) to actively force malicious manipulation models to output semantically confusing samples. Specifically, in the process of generating adversarial noise, we focus on constructing significant semantic confusions at the domain-specific knowledge level, and exploit a metric closely related to visual perception to replace the general pixel-wise metrics. The generated adversarial noise can actively interfere with the malicious manipulation model by triggering knowledge-guided and perception-related disruptions in the fake samples. To validate the effectiveness of the proposed method, we conduct qualitative and quantitative experiments on human perception and visual quality assessment. The results on two different tasks both show that our defense provides better protection compared to state-of-the-art methods and achieves great generalizability.
Authors: Gaetano Signorelli, Michele Lombardi
Abstract: The problem of ensuring constraints satisfaction on the output of machine learning models is critical for many applications, especially in safety-critical domains. Modern approaches rely on penalty-based methods at training time, which do not guarantee to avoid constraints violations; or constraint-specific model architectures (e.g., for monotonocity); or on output projection, which requires to solve an optimization problem that might be computationally demanding. We present the Hypersherical Constrained Representation, a novel method to enforce constraints in the output space for convex and bounded feasibility regions (generalizable to star domains). Our method operates on a different representation system, where Euclidean coordinates are converted into hyperspherical coordinates relative to the constrained region, which can only inherently represent feasible points. Experiments on a synthetic and a real-world dataset show that our method has predictive performance comparable to the other approaches, can guarantee 100% constraint satisfaction, and has a minimal computational cost at inference time.
Authors: Yilin Ning, Yian Ma, Mingxuan Liu, Xin Li, Nan Liu
Abstract: Fairness in artificial intelligence (AI) prediction models is increasingly emphasized to support responsible adoption in high-stakes domains such as health care and criminal justice. Guidelines and implementation frameworks highlight the importance of both predictive accuracy and equitable outcomes. However, current fairness toolkits often evaluate classification performance disparities in isolation, with limited attention to other critical aspects such as calibration. To address these gaps, we present seeBias, an R package for comprehensive evaluation of model fairness and predictive performance. seeBias offers an integrated evaluation across classification, calibration, and other performance domains, providing a more complete view of model behavior. It includes customizable visualizations to support transparent reporting and responsible AI implementation. Using public datasets from criminal justice and healthcare, we demonstrate how seeBias supports fairness evaluations, and uncovers disparities that conventional fairness metrics may overlook. The R package is available on GitHub, and a Python version is under development.
Authors: Tongyan Wu, Amine Bentellis, Alona Sakhnenko, Jeanette Miriam Lorenz
Abstract: Hybrid classical-quantum models aim to harness the strengths of both quantum computing and classical machine learning, but their practical potential remains poorly understood. In this work, we develop a unified mathematical framework for analyzing generalization in hybrid models, offering insight into how these systems learn from data. We establish a novel generalization bound of the form $O\big( \sqrt{\frac{T\log{T}}{N}} + \frac{\alpha}{\sqrt{N}}\big)$ for $N$ training data points, $T$ trainable quantum gates, and bounded fully-connected layers $||F|| \leq \alpha$. This bound decomposes cleanly into quantum and classical contributions, extending prior work on both components and clarifying their interaction. We apply our results to the quantum-classical convolutional neural network (QCCNN), an architecture that integrates quantum convolutional layers with classical processing. Alongside the bound, we highlight conceptual limitations of applying classical statistical learning theory in the hybrid setting and suggest promising directions for future theoretical work.
Authors: Pietro Foti, Andreas Brendel
Abstract: Recently, neural speech codecs (NSCs) trained as generative models have shown superior performance compared to conventional codecs at low bitrates. Although most state-of-the-art NSCs are trained as Generative Adversarial Networks (GANs), Diffusion Models (DMs), a recent class of generative models, represent a promising alternative due to their superior performance in image generation relative to GANs. Consequently, DMs have been successfully applied for audio and speech coding among various other audio generation applications. However, the design of diffusion-based NSCs has not yet been explored in a systematic way. We address this by providing a comprehensive analysis of diffusion-based NSCs divided into three contributions. First, we propose a categorization based on the conditioning and output domains of the DM. This simple conceptual framework allows us to define a design space for diffusion-based NSCs and to assign a category to existing approaches in the literature. Second, we systematically investigate unexplored designs by creating and evaluating new diffusion-based NSCs within the conceptual framework. Finally, we compare the proposed models to existing GAN and DM baselines through objective metrics and subjective listening tests.
Authors: Kerol Djoumessi, Samuel Ofosu Mensah, Philipp Berens
Abstract: In many medical imaging tasks, convolutional neural networks (CNNs) efficiently extract local features hierarchically. More recently, vision transformers (ViTs) have gained popularity, using self-attention mechanisms to capture global dependencies, but lacking the inherent spatial localization of convolutions. Therefore, hybrid models combining CNNs and ViTs have been developed to combine the strengths of both architectures. However, such hybrid CNN-ViT models are difficult to interpret, which hinders their application in medical imaging. In this work, we introduce an interpretable-by-design hybrid fully convolutional CNN-Transformer architecture for medical image classification. Unlike widely used post-hoc saliency methods for ViTs, our approach generates faithful and localized evidence maps that directly reflect the model's decision process. We evaluated our method on two medical image classification tasks using color fundus images. Our model not only achieves state-of-the-art predictive performance compared to both black-box and interpretable models but also provides class-specific sparse evidence maps in a single forward pass. The code is available at: https://anonymous.4open.science/r/Expl-CNN-Transformer/.
URLs: https://anonymous.4open.science/r/Expl-CNN-Transformer/.
Authors: Robin D. Pesl
Abstract: Modern enterprise computing systems integrate numerous subsystems to resolve a common task by yielding emergent behavior. A widespread approach is using services implemented with Web technologies like REST or OpenAPI, which offer an interaction mechanism and service documentation standard, respectively. Each service represents a specific business functionality, allowing encapsulation and easier maintenance. Despite the reduced maintenance costs on an individual service level, increased integration complexity arises. Consequently, automated service composition approaches have arisen to mitigate this issue. Nevertheless, these approaches have not achieved high acceptance in practice due to their reliance on complex formal modeling. Within this Ph.D. thesis, we analyze the application of Large Language Models (LLMs) to automatically integrate the services based on a natural language input. The result is a reusable service composition, e.g., as program code. While not always generating entirely correct results, the result can still be helpful by providing integration engineers with a close approximation of a suitable solution, which requires little effort to become operational. Our research involves (i) introducing a software architecture for automated service composition using LLMs, (ii) analyzing Retrieval Augmented Generation (RAG) for service discovery, (iii) proposing a novel natural language query-based benchmark for service discovery, and (iv) extending the benchmark to complete service composition scenarios. We have presented our software architecture as Compositio Prompto, the analysis of RAG for service discovery, and submitted a proposal for the service discovery benchmark. Open topics are primarily the extension of the service discovery benchmark to service composition scenarios and the improvements of the service composition generation, e.g., using fine-tuning or LLM agents.
Authors: Na Li, Chuke Wang, Yu Gu, Zhifeng Li
Abstract: Voice conversion (VC) transforms source speech into a target voice by preserving the content. However, timbre information from the source speaker is inherently embedded in the content representations, causing significant timbre leakage and reducing similarity to the target speaker. To address this, we introduce a residual block to a content extractor. The residual block consists of two weighted branches: 1) universal semantic dictionary based Content Feature Re-expression (CFR) module, supplying timbre-free content representation. 2) skip connection to the original content layer, providing complementary fine-grained information. In the CFR module, each dictionary entry in the universal semantic dictionary represents a phoneme class, computed statistically using speech from multiple speakers, creating a stable, speaker-independent semantic set. We introduce a CFR method to obtain timbre-free content representations by expressing each content frame as a weighted linear combination of dictionary entries using corresponding phoneme posteriors as weights. Extensive experiments across various VC frameworks demonstrate that our approach effectively mitigates timbre leakage and significantly improves similarity to the target speaker.
Authors: Charles Rathkopf
Abstract: Generative AI is increasingly used in scientific domains, from protein folding to climate modeling. But these models produce distinctive errors known as hallucinations - outputs that are incorrect yet superficially plausible. Worse, some arguments suggest that hallucinations are an inevitable consequence of the mechanisms underlying generative inference. Fortunately, such arguments rely on a conception of hallucination defined solely with respect to internal properties of the model, rather than in reference to the empirical target system. This conception fails to distinguish epistemically benign errors from those that threaten scientific inference. I introduce the concept of corrosive hallucination to capture the epistemically troubling subclass: misrepresentations that are substantively misleading and resistant to systematic anticipation. I argue that although corrosive hallucinations do pose a threat to scientific reliability, they are not inevitable. Scientific workflows such as those surrounding AlphaFold and GenCast, both of which serve as case studies, can neutralize their effects by imposing theoretical constraints during training, and by strategically screening for errors at inference time. When embedded in such workflows, generative AI can reliably contribute to scientific knowledge.
Authors: Farshad Noravesh, Reza Haffari, Layki Soon, Arghya Pal
Abstract: Hierarchical graph pooling(HGP) are designed to consider the fact that conventional graph neural networks(GNN) are inherently flat and are also not multiscale. However, most HGP methods suffer not only from lack of considering global topology of the graph and focusing on the feature learning aspect, but also they do not align local and global features since graphs should inherently be analyzed in a multiscale way. LGRPool is proposed in the present paper as a HGP in the framework of expectation maximization in machine learning that aligns local and global aspects of message passing with each other using a regularizer to force the global topological information to be inline with the local message passing at different scales through the representations at different layers of HGP. Experimental results on some graph classification benchmarks show that it slightly outperforms some baselines.
Authors: Thomas Tsouparopoulos, Iordanis Koutsopoulos
Abstract: As edge devices become more capable and pervasive in wireless networks, there is growing interest in leveraging their collective compute power for distributed learning. However, optimizing learning at the network edge entails unique challenges, particularly when moving beyond conventional settings and objectives. While Federated Learning (FL) has emerged as a key paradigm for distributed model training, critical challenges persist. First, existing approaches often overlook the trade-off between predictive accuracy and interpretability. Second, they struggle to integrate inherently explainable models such as decision trees because their non-differentiable structure makes them not amenable to backpropagation-based training algorithms. Lastly, they lack meaningful mechanisms for continual Machine Learning (ML) model adaptation through Continual Learning (CL) in resource-limited environments. In this paper, we pave the way for a set of novel optimization problems that emerge in distributed learning at the network edge with wirelessly interconnected edge devices, and we identify key challenges and future directions. Specifically, we discuss how Multi-objective optimization (MOO) can be used to address the trade-off between predictive accuracy and explainability when using complex predictive models. Next, we discuss the implications of integrating inherently explainable tree-based models into distributed learning settings. Finally, we investigate how CL strategies can be effectively combined with FL to support adaptive, lifelong learning when limited-size buffers are used to store past data for retraining. Our approach offers a cohesive set of tools for designing privacy-preserving, adaptive, and trustworthy ML solutions tailored to the demands of edge computing and intelligent services.
Authors: Zhao Dong, Ka Chen, Zhaoyang Lv, Hong-Xing Yu, Yunzhi Zhang, Cheng Zhang, Yufeng Zhu, Stephen Tian, Zhengqin Li, Geordie Moffatt, Sean Christofferson, James Fort, Xiaqing Pan, Mingfei Yan, Jiajun Wu, Carl Yuheng Ren, Richard Newcombe
Abstract: We introduce Digital Twin Catalog (DTC), a new large-scale photorealistic 3D object digital twin dataset. A digital twin of a 3D object is a highly detailed, virtually indistinguishable representation of a physical object, accurately capturing its shape, appearance, physical properties, and other attributes. Recent advances in neural-based 3D reconstruction and inverse rendering have significantly improved the quality of 3D object reconstruction. Despite these advancements, there remains a lack of a large-scale, digital twin quality real-world dataset and benchmark that can quantitatively assess and compare the performance of different reconstruction methods, as well as improve reconstruction quality through training or fine-tuning. Moreover, to democratize 3D digital twin creation, it is essential to integrate creation techniques with next-generation egocentric computing platforms, such as AR glasses. Currently, there is no dataset available to evaluate 3D object reconstruction using egocentric captured images. To address these gaps, the DTC dataset features 2,000 scanned digital twin-quality 3D objects, along with image sequences captured under different lighting conditions using DSLR cameras and egocentric AR glasses. This dataset establishes the first comprehensive real-world evaluation benchmark for 3D digital twin creation tasks, offering a robust foundation for comparing and improving existing reconstruction methods. The DTC dataset is already released at https://www.projectaria.com/datasets/dtc/ and we will also make the baseline evaluations open-source.
Authors: Alireza Fathalizadeh, Roozbeh Razavi-Far
Abstract: Continual generalized category discovery has been introduced and studied in the literature as a method that aims to continuously discover and learn novel categories in incoming data batches while avoiding catastrophic forgetting of previously learned categories. A key component in addressing this challenge is the model's ability to separate novel samples, where Extreme Value Theory (EVT) has been effectively employed. In this work, we propose a novel method that integrates EVT with proxy anchors to define boundaries around proxies using a probability of inclusion function, enabling the rejection of unknown samples. Additionally, we introduce a novel EVT-based loss function to enhance the learned representation, achieving superior performance compared to other deep-metric learning methods in similar settings. Using the derived probability functions, novel samples are effectively separated from previously known categories. However, category discovery within these novel samples can sometimes overestimate the number of new categories. To mitigate this issue, we propose a novel EVT-based approach to reduce the model size and discard redundant proxies. We also incorporate experience replay and knowledge distillation mechanisms during the continual learning stage to prevent catastrophic forgetting. Experimental results demonstrate that our proposed approach outperforms state-of-the-art methods in continual generalized category discovery scenarios.
Authors: Johannes Mae{\ss}, Gr\'egoire Montavon, Shinichi Nakajima, Klaus-Robert M\"uller, Thomas Schnake
Abstract: As machine learning models are increasingly considered for high-stakes domains, effective explanation methods are crucial to ensure that their prediction strategies are transparent to the user. Over the years, numerous metrics have been proposed to assess quality of explanations. However, their practical applicability remains unclear, in particular due to a limited understanding of which specific aspects each metric rewards. In this paper we propose a new framework based on spectral analysis of explanation outcomes to systematically capture the multifaceted properties of different explanation techniques. Our analysis uncovers two distinct factors of explanation quality-stability and target sensitivity-that can be directly observed through spectral decomposition. Experiments on both MNIST and ImageNet show that popular evaluation techniques (e.g., pixel-flipping, entropy) partially capture the trade-offs between these factors. Overall, our framework provides a foundational basis for understanding explanation quality, guiding the development of more reliable techniques for evaluating explanations.
Authors: Mahshad Lotfinia, Arash Tayebiarasteh, Samaneh Samiei, Mehdi Joodaki, Soroosh Tayebi Arasteh
Abstract: Reliable artificial intelligence (AI) models for medical image analysis often depend on large and diverse labeled datasets. Federated learning (FL) offers a decentralized and privacy-preserving approach to training but struggles in highly non-independent and identically distributed (non-IID) settings, where institutions with more representative data may experience degraded performance. Moreover, existing large-scale FL studies have been limited to adult datasets, neglecting the unique challenges posed by pediatric data, which introduces additional non-IID variability. To address these limitations, we analyzed n=398,523 adult chest radiographs from diverse institutions across multiple countries and n=9,125 pediatric images, leveraging transfer learning from general-purpose self-supervised image representations to classify pneumonia and cases with no abnormality. Using state-of-the-art vision transformers, we found that FL improved performance only for smaller adult datasets (P<0.001) but degraded performance for larger datasets (P<0.064) and pediatric cases (P=0.242). However, equipping FL with self-supervised weights significantly enhanced outcomes across pediatric cases (P=0.031) and most adult datasets (P<0.008), except the largest dataset (P=0.052). These findings underscore the potential of easily deployable general-purpose self-supervised image representations to address non-IID challenges in clinical FL applications and highlight their promise for enhancing patient outcomes and advancing pediatric healthcare, where data scarcity and variability remain persistent obstacles.
Authors: Mohamed S. Talamali, Genki Miyauchi, Thomas Watteyne, Micael S. Couceiro, Roderich Gross
Abstract: Unmanned Aerial Vehicles (UAVs) are expected to transform logistics, reducing delivery time, costs, and emissions. This study addresses an on-demand delivery , in which fleets of UAVs are deployed to fulfil orders that arrive stochastically. Unlike previous work, it considers UAVs with heterogeneous, unknown energy storage capacities and assumes no knowledge of the energy consumption models. We propose a decentralised deployment strategy that combines auction-based task allocation with online learning. Each UAV independently decides whether to bid for orders based on its energy storage charge level, the parcel mass, and delivery distance. Over time, it refines its policy to bid only for orders within its capability. Simulations using realistic UAV energy models reveal that, counter-intuitively, assigning orders to the least confident bidders reduces delivery times and increases the number of successfully fulfilled orders. This strategy is shown to outperform threshold-based methods which require UAVs to exceed specific charge levels at deployment. We propose a variant of the strategy which uses learned policies for forecasting. This enables UAVs with insufficient charge levels to commit to fulfilling orders at specific future times, helping to prioritise early orders. Our work provides new insights into long-term deployment of UAV swarms, highlighting the advantages of decentralised energy-aware decision-making coupled with online learning in real-world dynamic environments.
Authors: Low Jian He, Harry Walsh, Ozge Mercanoglu Sincan, Richard Bowden
Abstract: This work tackles the challenge of continuous sign language segmentation, a key task with huge implications for sign language translation and data annotation. We propose a transformer-based architecture that models the temporal dynamics of signing and frames segmentation as a sequence labeling problem using the Begin-In-Out (BIO) tagging scheme. Our method leverages the HaMeR hand features, and is complemented with 3D Angles. Extensive experiments show that our model achieves state-of-the-art results on the DGS Corpus, while our features surpass prior benchmarks on BSLCorpus.
Authors: Gaya Mehenni, Amal Zouaq
Abstract: We present MedHal, a novel large-scale dataset specifically designed to evaluate if models can detect hallucinations in medical texts. Current hallucination detection methods face significant limitations when applied to specialized domains like medicine, where they can have disastrous consequences. Existing medical datasets are either too small, containing only a few hundred samples, or focus on a single task like Question Answering or Natural Language Inference. MedHal addresses these gaps by: (1) incorporating diverse medical text sources and tasks; (2) providing a substantial volume of annotated samples suitable for training medical hallucination detection models; and (3) including explanations for factual inconsistencies to guide model learning. We demonstrate MedHal's utility by training and evaluating a baseline medical hallucination detection model, showing improvements over general-purpose hallucination detection approaches. This resource enables more efficient evaluation of medical text generation systems while reducing reliance on costly expert review, potentially accelerating the development of medical AI research.
Authors: Gesina Schwalbe, Georgii Mikriukov, Edgar Heinert, Stavros Gerolymatos, Mert Keser, Alois Knoll, Matthias Rottmann, Annika M\"utze
Abstract: The thriving research field of concept-based explainable artificial intelligence (C-XAI) investigates how human-interpretable semantic concepts embed in the latent spaces of deep neural networks (DNNs). Post-hoc approaches therein use a set of examples to specify a concept, and determine its embeddings in DNN latent space using data driven techniques. This proved useful to uncover biases between different target (foreground or concept) classes. However, given that the background is mostly uncontrolled during training, an important question has been left unattended so far: Are/to what extent are state-of-the-art, data-driven post-hoc C-XAI approaches themselves prone to biases with respect to their backgrounds? E.g., wild animals mostly occur against vegetation backgrounds, and they seldom appear on roads. Even simple and robust C-XAI methods might abuse this shortcut for enhanced performance. A dangerous performance degradation of the concept-corner cases of animals on the road could thus remain undiscovered. This work validates and thoroughly confirms that established Net2Vec-based concept segmentation techniques frequently capture background biases, including alarming ones, such as underperformance on road scenes. For the analysis, we compare 3 established techniques from the domain of background randomization on >50 concepts from 2 datasets, and 7 diverse DNN architectures. Our results indicate that even low-cost setups can provide both valuable insight and improved background robustness.
Authors: Sebasti\'an Barbas Laina, Simon Boche, Sotiris Papatheodorou, Simon Schaefer, Jaehyung Jung, Stefan Leutenegger
Abstract: Geometrically accurate and semantically expressive map representations have proven invaluable to facilitate robust and safe mobile robot navigation and task planning. Nevertheless, real-time, open-vocabulary semantic understanding of large-scale unknown environments is still an open problem. In this paper we present FindAnything, an open-world mapping and exploration framework that incorporates vision-language information into dense volumetric submaps. Thanks to the use of vision-language features, FindAnything bridges the gap between pure geometric and open-vocabulary semantic information for a higher level of understanding while allowing to explore any environment without the help of any external source of ground-truth pose information. We represent the environment as a series of volumetric occupancy submaps, resulting in a robust and accurate map representation that deforms upon pose updates when the underlying SLAM system corrects its drift, allowing for a locally consistent representation between submaps. Pixel-wise vision-language features are aggregated from efficient SAM (eSAM)-generated segments, which are in turn integrated into object-centric volumetric submaps, providing a mapping from open-vocabulary queries to 3D geometry that is scalable also in terms of memory usage. The open-vocabulary map representation of FindAnything achieves state-of-the-art semantic accuracy in closed-set evaluations on the Replica dataset. This level of scene understanding allows a robot to explore environments based on objects or areas of interest selected via natural language queries. Our system is the first of its kind to be deployed on resource-constrained devices, such as MAVs, leveraging vision-language information for real-world robotic tasks.
Authors: Youwei Yu, Lantao Liu
Abstract: Deep reinforcement learning can seamlessly transfer agile locomotion and navigation skills from the simulator to real world. However, bridging the sim-to-real gap with domain randomization or adversarial methods often demands expert physics knowledge to ensure policy robustness. Even so, cutting-edge simulators may fall short of capturing every real-world detail, and the reconstructed environment may introduce errors due to various perception uncertainties. To address these challenges, we propose Neural Fidelity Calibration (NFC), a novel framework that employs conditional score-based diffusion models to calibrate simulator physical coefficients and residual fidelity domains online during robot execution. Specifically, the residual fidelity reflects the simulation model shift relative to the real-world dynamics and captures the uncertainty of the perceived environment, enabling us to sample realistic environments under the inferred distribution for policy fine-tuning. Our framework is informative and adaptive in three key ways: (a) we fine-tune the pretrained policy only under anomalous scenarios, (b) we build sequential NFC online with the pretrained NFC's proposal prior, reducing the diffusion model's training burden, and (c) when NFC uncertainty is high and may degrade policy improvement, we leverage optimistic exploration to enable hallucinated policy optimization. Our framework achieves superior simulator calibration precision compared to state-of-the-art methods across diverse robots with high-dimensional parametric spaces. We study the critical contribution of residual fidelity to policy improvement in simulation and real-world experiments. Notably, our approach demonstrates robust robot navigation under challenging real-world conditions, such as a broken wheel axle on snowy surfaces.
Authors: Julian B\"aumler, Louis Bl\"ocher, Lars-Joel Frey, Xian Chen, Markus Bayer, Christian Reuter
Abstract: The dissemination of online hate speech can have serious negative consequences for individuals, online communities, and entire societies. This and the large volume of hateful online content prompted both practitioners', i.e., in content moderation or law enforcement, and researchers' interest in machine learning models to automatically classify instances of hate speech. Whereas most scientific works address hate speech classification as a binary task, practice often requires a differentiation into sub-types, e.g., according to target, severity, or legality, which may overlap for individual content. Hence, researchers created datasets and machine learning models that approach hate speech classification in textual data as a multi-label problem. This work presents the first systematic and comprehensive survey of scientific literature on this emerging research landscape in English (N=46). We contribute with a concise overview of 28 datasets suited for training multi-label classification models that reveals significant heterogeneity regarding label-set, size, meta-concept, annotation process, and inter-annotator agreement. Our analysis of 24 publications proposing suitable classification models further establishes inconsistency in evaluation and a preference for architectures based on Bidirectional Encoder Representation from Transformers (BERT) and Recurrent Neural Networks (RNNs). We identify imbalanced training data, reliance on crowdsourcing platforms, small and sparse datasets, and missing methodological alignment as critical open issues and formulate ten recommendations for research.
Authors: Vineeth Sai Narajala, Idan Habler
Abstract: The Model Context Protocol (MCP), introduced by Anthropic, provides a standardized framework for artificial intelligence (AI) systems to interact with external data sources and tools in real-time. While MCP offers significant advantages for AI integration and capability extension, it introduces novel security challenges that demand rigorous analysis and mitigation. This paper builds upon foundational research into MCP architecture and preliminary security assessments to deliver enterprise-grade mitigation frameworks and detailed technical implementation strategies. Through systematic threat modeling and analysis of MCP implementations and analysis of potential attack vectors, including sophisticated threats like tool poisoning, we present actionable security patterns tailored for MCP implementers and adopters. The primary contribution of this research lies in translating theoretical security concerns into a practical, implementable framework with actionable controls, thereby providing essential guidance for the secure enterprise adoption and governance of integrated AI systems.
Authors: Renu Sharma, Debasmita Pal, Arun Ross
Abstract: One of the major challenges in machine learning is maintaining the accuracy of the deployed model (e.g., a classifier) in a non-stationary environment. The non-stationary environment results in distribution shifts and, consequently, a degradation in accuracy. Continuous learning of the deployed model with new data could be one remedy. However, the question arises as to how we should update the model with new training data so that it retains its accuracy on the old data while adapting to the new data. In this work, we propose a task-conditioned ensemble of models to maintain the performance of the existing model. The method involves an ensemble of expert models based on task membership information. The in-domain models-based on the local outlier concept (different from the expert models) provide task membership information dynamically at run-time to each probe sample. To evaluate the proposed method, we experiment with three setups: the first represents distribution shift between tasks (LivDet-Iris-2017), the second represents distribution shift both between and within tasks (LivDet-Iris-2020), and the third represents disjoint distribution between tasks (Split MNIST). The experiments highlight the benefits of the proposed method. The source code is available at https://github.com/iPRoBe-lab/Continuous_Learning_FE_DM.
URLs: https://github.com/iPRoBe-lab/Continuous_Learning_FE_DM.
Authors: Athanasios Athanasopoulos, Mat\'u\v{s} Mihal\'ak, Marcin Pietrasik
Abstract: One of the key safety considerations of battery manufacturing is thermal runaway, the uncontrolled increase in temperature which can lead to fires, explosions, and emissions of toxic gasses. As such, development of automated systems capable of detecting such events is of considerable importance in both academic and industrial contexts. In this work, we investigate the use of deep learning for detecting thermal runaway in the battery production line of VDL Nedcar, a Dutch automobile manufacturer. Specifically, we collect data from the production line to represent both baseline (non thermal runaway) and thermal runaway conditions. Thermal runaway was simulated through the use of external heat and smoke sources. The data consisted of both optical and thermal images which were then preprocessed and fused before serving as input to our models. In this regard, we evaluated three deep-learning models widely used in computer vision including shallow convolutional neural networks, residual neural networks, and vision transformers on two performance metrics. Furthermore, we evaluated these models using explainability methods to gain insight into their ability to capture the relevant feature information from their inputs. The obtained results indicate that the use of deep learning is a viable approach to thermal runaway detection in battery production lines.
Authors: Jialu Li, Shoubin Yu, Han Lin, Jaemin Cho, Jaehong Yoon, Mohit Bansal
Abstract: Recent advancements in text-to-video (T2V) diffusion models have significantly enhanced the visual quality of the generated videos. However, even recent T2V models find it challenging to follow text descriptions accurately, especially when the prompt requires accurate control of spatial layouts or object trajectories. A recent line of research uses layout guidance for T2V models that require fine-tuning or iterative manipulation of the attention map during inference time. This significantly increases the memory requirement, making it difficult to adopt a large T2V model as a backbone. To address this, we introduce Video-MSG, a training-free Guidance method for T2V generation based on Multimodal planning and Structured noise initialization. Video-MSG consists of three steps, where in the first two steps, Video-MSG creates Video Sketch, a fine-grained spatio-temporal plan for the final video, specifying background, foreground, and object trajectories, in the form of draft video frames. In the last step, Video-MSG guides a downstream T2V diffusion model with Video Sketch through noise inversion and denoising. Notably, Video-MSG does not need fine-tuning or attention manipulation with additional memory during inference time, making it easier to adopt large T2V models. Video-MSG demonstrates its effectiveness in enhancing text alignment with multiple T2V backbones (VideoCrafter2 and CogVideoX-5B) on popular T2V generation benchmarks (T2VCompBench and VBench). We provide comprehensive ablation studies about noise inversion ratio, different background generators, background object detection, and foreground object segmentation.
Authors: Alessio Lombardi (Buro Happold, London), Li Duan (Birmingham City University), Ahmed Elnagar (Buro Happold, London), Ahmed Zaalouk (Birmingham City University), Khalid Ismail (Birmingham City University), Edlira Vakaj (Birmingham City University)
Abstract: The architecture, engineering, and construction (AEC) industry still heavily relies on information stored in drawings for building construction, maintenance, compliance and error checks. However, information extraction (IE) from building drawings is often time-consuming and costly, especially when dealing with historical buildings. Drawing search can be simplified by leveraging the information stored in the title block portion of the drawing, which can be seen as drawing metadata. However, title block IE can be complex especially when dealing with historical drawings which do not follow existing standards for uniformity. This work performs a comparison of existing methods for this kind of IE task, and then proposes a novel title block detection and IE pipeline which outperforms existing methods, in particular when dealing with complex, noisy historical drawings. The pipeline is obtained by combining a lightweight Convolutional Neural Network and GPT-4o, the proposed inference pipeline detects building engineering title blocks with high accuracy, and then extract structured drawing metadata from the title blocks, which can be used for drawing search, filtering and grouping. The work demonstrates high accuracy and efficiency in IE for both vector (CAD) and hand-drawn (historical) drawings. A user interface (UI) that leverages the extracted metadata for drawing search is established and deployed on real projects, which demonstrates significant time savings. Additionally, an extensible domain-expert-annotated dataset for title block detection is developed, via an efficient AEC-friendly annotation workflow that lays the foundation for future work.
Authors: Alexandre Bazin (Huaxi), Alain Gutierrez (Huaxi), Marianne Huchard (Huaxi), Pierre Martin (Huaxi), Yulin (Huaxi), Zhang
Abstract: A widely used Agile practice for requirements is to produce a set of user stories (also called ``agile product backlog''), which roughly includes a list of pairs (role, feature), where the role handles the feature for a certain purpose. In the context of Software Product Lines, the requirements for a family of similar systems is thus a family of user-story sets, one per system, leading to a 3-dimensional dataset composed of sets of triples (system, role, feature). In this paper, we combine Triadic Concept Analysis (TCA) and Large Language Model (LLM) prompting to suggest the user-story set required to develop a new system relying on the variability logic of an existing system family. This process consists in 1) computing 3-dimensional variability expressed as a set of TCA implications, 2) providing the designer with intelligible design options, 3) capturing the designer's selection of options, 4) proposing a first user-story set corresponding to this selection, 5) consolidating its validity according to the implications identified in step 1, while completing it if necessary, and 6) leveraging LLM to have a more comprehensive website. This process is evaluated with a dataset comprising the user-story sets of 67 similar-purpose websites.
Authors: Nomisha Kurian
Abstract: To build AI interfaces that children can intuitively understand and use, designers need a design grammar that truly serves children's developmental needs. This paper bridges Artificial Intelligence design for children -- an emerging field still defining its best practices -- and children's animation, a well-established field with decades of experience in engaging young viewers through emotionally resonant, cognitively accessible storytelling. Pairing Piagetian developmental theory with design pattern extraction from 52 works of Disney animation, the paper presents six design insights transferable to child-centred AI interface design: (1) emotional expressiveness and visual clarity, (2) musical and auditory scaffolding, (3) audiovisual synchrony for emotional comfort, (4) sidekick-style personas, (5) support for symbolic play and imaginative exploration, and (6) predictable and scaffolded interaction structures. These strategies -- long refined in Disney animation -- function as multimodal scaffolds for attention, understanding, and emotional attunement, thereby forming a structured design grammar familiar to children and transferable to AI interface design. By reframing cinematic storytelling as design logic for AI, the paper offers heuristics for crafting intuitive AI interfaces that align with children's cognitive stages and emotional needs. The work contributes to design theory by showing how sensory, affective and narrative techniques can inform developmentally attuned AI design for children. Future directions include empirical testing, cultural adaptation, and participatory co-design.
Authors: Fangzhi Xu, Hang Yan, Chang Ma, Haiteng Zhao, Qiushi Sun, Kanzhi Cheng, Junxian He, Jun Liu, Zhiyong Wu
Abstract: Advancing LLM reasoning skills has captivated wide interest. However, current post-training techniques rely heavily on supervisory signals, such as outcome supervision or auxiliary reward models, which face the problem of scalability and high annotation costs. This motivates us to enhance LLM reasoning without the need for external supervision. We introduce a generalizable and purely unsupervised self-training framework, named Genius. Without external auxiliary, Genius requires to seek the optimal response sequence in a stepwise manner and optimize the LLM. To explore the potential steps and exploit the optimal ones, Genius introduces a stepwise foresight re-sampling strategy to sample and estimate the step value by simulating future outcomes. Further, we recognize that the unsupervised setting inevitably induces the intrinsic noise and uncertainty. To provide a robust optimization, we propose an advantage-calibrated optimization (ACO) loss function to mitigate estimation inconsistencies. Combining these techniques together, Genius provides an advanced initial step towards self-improve LLM reasoning with general queries and without supervision, revolutionizing reasoning scaling laws given the vast availability of general queries. The code will be released at https://github.com/xufangzhi/Genius.
Authors: Team Seawead, Ceyuan Yang, Zhijie Lin, Yang Zhao, Shanchuan Lin, Zhibei Ma, Haoyuan Guo, Hao Chen, Lu Qi, Sen Wang, Feng Cheng, Feilong Zuo Xuejiao Zeng, Ziyan Yang, Fangyuan Kong, Zhiwu Qing, Fei Xiao, Meng Wei, Tuyen Hoang, Siyu Zhang, Peihao Zhu, Qi Zhao, Jiangqiao Yan, Liangke Gui, Sheng Bi, Jiashi Li, Yuxi Ren, Rui Wang, Huixia Li, Xuefeng Xiao, Shu Liu, Feng Ling, Heng Zhang, Houmin Wei, Huafeng Kuang, Jerry Duncan, Junda Zhang, Junru Zheng, Li Sun, Manlin Zhang, Renfei Sun, Xiaobin Zhuang, Xiaojie Li, Xin Xia, Xuyan Chi, Yanghua Peng, Yuping Wang, Yuxuan Wang, Zhongkai Zhao, Zhuo Chen, Zuquan Song, Zhenheng Yang, Jiashi Feng, Jianchao Yang, Lu Jiang
Abstract: This technical report presents a cost-efficient strategy for training a video generation foundation model. We present a mid-sized research model with approximately 7 billion parameters (7B) called Seaweed-7B trained from scratch using 665,000 H100 GPU hours. Despite being trained with moderate computational resources, Seaweed-7B demonstrates highly competitive performance compared to contemporary video generation models of much larger size. Design choices are especially crucial in a resource-constrained setting. This technical report highlights the key design decisions that enhance the performance of the medium-sized diffusion model. Empirically, we make two observations: (1) Seaweed-7B achieves performance comparable to, or even surpasses, larger models trained on substantially greater GPU resources, and (2) our model, which exhibits strong generalization ability, can be effectively adapted across a wide range of downstream applications either by lightweight fine-tuning or continue training. See the project page at https://seaweed.video/
URLs: https://seaweed.video/
Authors: Alessia Loi, Loona Macabre, J\'er\'emy Fersula, Keivan Amini, Leo Cazenille, Fabien Caura, Alexandre Guerre, St\'ephane Gourichon, Olivier Dauchot, Nicolas Bredeche
Abstract: This paper describes the Pogobot, an open-source and open-hardware platform specifically designed for research involving swarm robotics. Pogobot features vibration-based locomotion, infrared communication, and an array of sensors in a cost-effective package (approx. 250~euros/unit). The platform's modular design, comprehensive API, and extensible architecture facilitate the implementation of swarm intelligence algorithms and distributed online reinforcement learning algorithms. Pogobots offer an accessible alternative to existing platforms while providing advanced capabilities including directional communication between units. More than 200 Pogobots are already being used on a daily basis at Sorbonne Universit\'e and PSL to study self-organizing systems, programmable active matter, discrete reaction-diffusion-advection systems as well as models of social learning and evolution.
Authors: Jiho Kim, Philippe Laban, Xiang 'Anthony' Chen, Kenneth C. Arnold
Abstract: Writing well requires not only expressing ideas but also refining them through revision, a process facilitated by reflection. Prior research suggests that feedback delivered through dialogues, such as those in writing center tutoring sessions, can help writers reflect more thoughtfully on their work compared to static feedback. Recent advancements in multi-modal large language models (LLMs) now offer new possibilities for supporting interactive and expressive voice-based reflection in writing. In particular, we propose that LLM-generated static feedback can be repurposed as conversation starters, allowing writers to seek clarification, request examples, and ask follow-up questions, thereby fostering deeper reflection on their writing. We argue that voice-based interaction can naturally facilitate this conversational exchange, encouraging writers' engagement with higher-order concerns, facilitating iterative refinement of their reflections, and reduce cognitive load compared to text-based interactions. To investigate these effects, we propose a formative study exploring how text vs. voice input influence writers' reflection and subsequent revisions. Findings from this study will inform the design of intelligent and interactive writing tools, offering insights into how voice-based interactions with LLM-powered conversational agents can support reflection and revision.
Authors: Yiliu Sun, Yanfang Zhang, Zicheng Zhao, Sheng Wan, Dacheng Tao, Chen Gong
Abstract: Nowadays, Large Language Models (LLMs) have been gradually employed to solve complex tasks. To face the challenge, task decomposition has become an effective way, which proposes to divide a complex task into multiple simpler subtasks and then solve them separately so that the difficulty of the original task can be reduced. However, the performance of existing task decomposition methods can be suboptimal when the task contains overly complex logic and constraints. In this situation, the solution generated by LLMs may deviate from the original purpose of the task, or contain redundant or even erroneous content. Therefore, inspired by the fact that humans possess two thinking systems including fast thinking and slow thinking, this paper introduces a new task decomposition method termed ``Fast-Slow-Thinking'' (FST), which stimulates LLMs to solve tasks through the cooperation of Fast Thinking (FT) and Slow Thinking (ST) steps. Here FT focuses more on the general and concise aspect of the task, and ST focuses more on the details of the task. In FT, LLMs are prompted to remove the constraints of the original task, therefore simplifying it to a general and concise one. In ST, we recall the constraints removed in FT, so that LLMs can improve the answer generated in FT to meet the requirements of the original task. Therefore, our FST method enables LLMs to consider a complex problem via a human-like cognition process from coarse to fine, the effectiveness of which has been well demonstrated by the experiments on three types of tasks.
Authors: Sahil Sethi, David Chen, Thomas Statchen, Michael C. Burkhart, Nipun Bhandari, Bashar Ramadan, Brett Beaulieu-Jones
Abstract: Deep learning-based electrocardiogram (ECG) classification has shown impressive performance but clinical adoption has been slowed by the lack of transparent and faithful explanations. Post hoc methods such as saliency maps may fail to reflect a model's true decision process. Prototype-based reasoning offers a more transparent alternative by grounding decisions in similarity to learned representations of real ECG segments, enabling faithful, case-based explanations. We introduce ProtoECGNet, a prototype-based deep learning model for interpretable, multi-label ECG classification. ProtoECGNet employs a structured, multi-branch architecture that reflects clinical interpretation workflows: it integrates a 1D CNN with global prototypes for rhythm classification, a 2D CNN with time-localized prototypes for morphology-based reasoning, and a 2D CNN with global prototypes for diffuse abnormalities. Each branch is trained with a prototype loss designed for multi-label learning, combining clustering, separation, diversity, and a novel contrastive loss that encourages appropriate separation between prototypes of unrelated classes while allowing clustering for frequently co-occurring diagnoses. We evaluate ProtoECGNet on all 71 diagnostic labels from the PTB-XL dataset, demonstrating competitive performance relative to state-of-the-art black-box models while providing structured, case-based explanations. To assess prototype quality, we conduct a structured clinician review of the final model's projected prototypes, finding that they are rated as representative and clear. ProtoECGNet shows that prototype learning can be effectively scaled to complex, multi-label time-series classification, offering a practical path toward transparent and trustworthy deep learning models for clinical decision support.
Authors: Dayu Yang, Antoine Simoulin, Xin Qian, Xiaoyi Liu, Yuwei Cao, Zhaopu Teng, Grey Yang
Abstract: High-quality code documentation is crucial for software development especially in the era of AI. However, generating it automatically using Large Language Models (LLMs) remains challenging, as existing approaches often produce incomplete, unhelpful, or factually incorrect outputs. We introduce DocAgent, a novel multi-agent collaborative system using topological code processing for incremental context building. Specialized agents (Reader, Searcher, Writer, Verifier, Orchestrator) then collaboratively generate documentation. We also propose a multi-faceted evaluation framework assessing Completeness, Helpfulness, and Truthfulness. Comprehensive experiments show DocAgent significantly outperforms baselines consistently. Our ablation study confirms the vital role of the topological processing order. DocAgent offers a robust approach for reliable code documentation generation in complex and proprietary repositories.
Authors: Boyang Deng, Songyou Peng, Kyle Genova, Gordon Wetzstein, Noah Snavely, Leonidas Guibas, Thomas Funkhouser
Abstract: We present a system using Multimodal LLMs (MLLMs) to analyze a large database with tens of millions of images captured at different times, with the aim of discovering patterns in temporal changes. Specifically, we aim to capture frequent co-occurring changes ("trends") across a city over a certain period. Unlike previous visual analyses, our analysis answers open-ended queries (e.g., "what are the frequent types of changes in the city?") without any predetermined target subjects or training labels. These properties cast prior learning-based or unsupervised visual analysis tools unsuitable. We identify MLLMs as a novel tool for their open-ended semantic understanding capabilities. Yet, our datasets are four orders of magnitude too large for an MLLM to ingest as context. So we introduce a bottom-up procedure that decomposes the massive visual analysis problem into more tractable sub-problems. We carefully design MLLM-based solutions to each sub-problem. During experiments and ablation studies with our system, we find it significantly outperforms baselines and is able to discover interesting trends from images captured in large cities (e.g., "addition of outdoor dining,", "overpass was painted blue," etc.). See more results and interactive demos at https://boyangdeng.com/visual-chronicles.
Authors: Sonia Joseph, Praneet Suresh, Ethan Goldfarb, Lorenz Hufe, Yossi Gandelsman, Robert Graham, Danilo Bzdok, Wojciech Samek, Blake Aaron Richards
Abstract: While vision models are highly capable, their internal mechanisms remain poorly understood -- a challenge which sparse autoencoders (SAEs) have helped address in language, but which remains underexplored in vision. We address this gap by training SAEs on CLIP's vision transformer and uncover key differences between vision and language processing, including distinct sparsity patterns for SAEs trained across layers and token types. We then provide the first systematic analysis on the steerability of CLIP's vision transformer by introducing metrics to quantify how precisely SAE features can be steered to affect the model's output. We find that 10-15\% of neurons and features are steerable, with SAEs providing thousands more steerable features than the base model. Through targeted suppression of SAE features, we then demonstrate improved performance on three vision disentanglement tasks (CelebA, Waterbirds, and typographic attacks), finding optimal disentanglement in middle model layers, and achieving state-of-the-art performance on defense against typographic attacks.
Authors: Yanlin Wang, Kefeng Duan, Dewu Zheng, Ensheng Shi, Fengji Zhang, Yanli Wang, Jiachi Chen, Xilin Liu, Yuchi Ma, Hongyu Zhang, Qianxiang Wang, Zibin Zheng
Abstract: Code intelligence is an emerging domain in software engineering, aiming to improve the effectiveness and efficiency of various code-related tasks. Recent research suggests that incorporating contextual information beyond the basic original task inputs (i.e., source code) can substantially enhance model performance. Such contextual signals may be obtained directly or indirectly from sources such as API documentation or intermediate representations like abstract syntax trees can significantly improve the effectiveness of code intelligence. Despite growing academic interest, there is a lack of systematic analysis of context in code intelligence. To address this gap, we conduct an extensive literature review of 146 relevant studies published between September 2007 and August 2024. Our investigation yields four main contributions. (1) A quantitative analysis of the research landscape, including publication trends, venues, and the explored domains; (2) A novel taxonomy of context types used in code intelligence; (3) A task-oriented analysis investigating context integration strategies across diverse code intelligence tasks; (4) A critical evaluation of evaluation methodologies for context-aware methods. Based on these findings, we identify fundamental challenges in context utilization in current code intelligence systems and propose a research roadmap that outlines key opportunities for future research.
Authors: Diego Perez-Liebana, Katja Hofmann, Sharada Prasanna Mohanty, Noboru Kuno, Andre Kramer, Sam Devlin, Raluca D. Gaina, Daniel Ionita
Abstract: Learning in multi-agent scenarios is a fruitful research direction, but current approaches still show scalability problems in multiple games with general reward settings and different opponent types. The Multi-Agent Reinforcement Learning in Malm\"O (MARL\"O) competition is a new challenge that proposes research in this domain using multiple 3D games. The goal of this contest is to foster research in general agents that can learn across different games and opponent types, proposing a challenge as a milestone in the direction of Artificial General Intelligence.
Authors: Peter Slattery, Alexander K. Saeri, Emily A. C. Grundy, Jess Graham, Michael Noetel, Risto Uuk, James Dao, Soroush Pour, Stephen Casper, Neil Thompson
Abstract: The risks posed by Artificial Intelligence (AI) are of considerable concern to academics, auditors, policymakers, AI companies, and the public. However, a lack of shared understanding of AI risks can impede our ability to comprehensively discuss, research, and react to them. This paper addresses this gap by creating an AI Risk Repository to serve as a common frame of reference. This comprises a living database of 777 risks extracted from 43 taxonomies, which can be filtered based on two overarching taxonomies and easily accessed, modified, and updated via our website and online spreadsheets. We construct our Repository with a systematic review of taxonomies and other structured classifications of AI risk followed by an expert consultation. We develop our taxonomies of AI risk using a best-fit framework synthesis. Our high-level Causal Taxonomy of AI Risks classifies each risk by its causal factors (1) Entity: Human, AI; (2) Intentionality: Intentional, Unintentional; and (3) Timing: Pre-deployment; Post-deployment. Our mid-level Domain Taxonomy of AI Risks classifies risks into seven AI risk domains: (1) Discrimination & toxicity, (2) Privacy & security, (3) Misinformation, (4) Malicious actors & misuse, (5) Human-computer interaction, (6) Socioeconomic & environmental, and (7) AI system safety, failures, & limitations. These are further divided into 23 subdomains. The AI Risk Repository is, to our knowledge, the first attempt to rigorously curate, analyze, and extract AI risk frameworks into a publicly accessible, comprehensive, extensible, and categorized risk database. This creates a foundation for a more coordinated, coherent, and complete approach to defining, auditing, and managing the risks posed by AI systems.
Authors: Liqiang Jing, Zhehui Huang, Xiaoyang Wang, Wenlin Yao, Wenhao Yu, Kaixin Ma, Hongming Zhang, Xinya Du, Dong Yu
Abstract: Large Language Models (LLMs) and Large Vision-Language Models (LVLMs) have demonstrated impressive language/vision reasoning abilities, igniting the recent trend of building agents for targeted applications such as shopping assistants or AI software engineers. Recently, many data science benchmarks have been proposed to investigate their performance in the data science domain. However, existing data science benchmarks still fall short when compared to real-world data science applications due to their simplified settings. To bridge this gap, we introduce DSBench, a comprehensive benchmark designed to evaluate data science agents with realistic tasks. This benchmark includes 466 data analysis tasks and 74 data modeling tasks, sourced from Eloquence and Kaggle competitions. DSBench offers a realistic setting by encompassing long contexts, multimodal task backgrounds, reasoning with large data files and multi-table structures, and performing end-to-end data modeling tasks. Our evaluation of state-of-the-art LLMs, LVLMs, and agents shows that they struggle with most tasks, with the best agent solving only 34.12% of data analysis tasks and achieving a 34.74% Relative Performance Gap (RPG). These findings underscore the need for further advancements in developing more practical, intelligent, and autonomous data science agents.
Authors: DiJia Su, Sainbayar Sukhbaatar, Michael Rabbat, Yuandong Tian, Qinqing Zheng
Abstract: In human cognition theory, human thinking is governed by two systems: the fast and intuitive System 1 and the slower but more deliberative System 2. Recent studies have shown that incorporating System 2 process into Transformers including large language models (LLMs), significantly enhances their reasoning capabilities. Nevertheless, models that purely resemble System 2 thinking require substantially higher computational costs and are much slower to respond. To address this challenge, we present Dualformer, a single Transformer model that seamlessly integrates both the fast and slow reasoning modes. Dualformer is obtained by training on data with randomized reasoning traces, where different parts of the traces are dropped during training. The dropping strategies are specifically tailored according to the trace structure, analogous to analyzing our thinking process and creating shortcuts with patterns. At inference time, our model can be configured to output only the solutions (fast mode) or both the reasoning chain and the final solution (slow mode), or automatically decide which mode to engage (auto mode). In all cases, Dualformer outperforms the corresponding baseline models in both performance and computational efficiency: (1) in slow mode, Dualformer optimally solves unseen 30 x 30 maze navigation tasks 97.6% of the time, surpassing the Searchformer (trained on data with complete reasoning traces) baseline performance of 93.3%, while only using 45.5% fewer reasoning steps; (2) in fast mode, Dualformer completes those tasks with an 80% optimal rate, significantly outperforming the Solution-Only model (trained on solution-only data), which has an optimal rate of only 30%. For math problems, our techniques have also achieved improved performance with LLM fine-tuning, showing its generalization beyond task-specific models.
Authors: Han Wang, Yilin Zhao, Dian Li, Xiaohan Wang, Gang Liu, Xuguang Lan, Hui Wang
Abstract: Humor is previously regarded as a gift exclusive to humans for the following reasons. Humor is a culturally nuanced aspect of human language, presenting challenges for its understanding and generation. Humor generation necessitates a multi-hop reasoning process, with each hop founded on proper rationales. Although many studies, such as those related to GPT-o1, focus on logical reasoning with reflection and correction, they still fall short in humor generation. Due to the sparsity of the knowledge graph in creative thinking, it is arduous to achieve multi-hop reasoning. Consequently, in this paper, we propose a more robust framework for addressing the humor reasoning task, named LoL. LoL aims to inject external information to mitigate the sparsity of the knowledge graph, thereby enabling multi-hop reasoning. In the first stage of LoL, we put forward an automatic instruction-evolution method to incorporate the deeper and broader thinking processes underlying humor. Judgment-oriented instructions are devised to enhance the model's judgment capability, dynamically supplementing and updating the sparse knowledge graph. Subsequently, through reinforcement learning, the reasoning logic for each online-generated response is extracted using GPT-4o. In this process, external knowledge is re-introduced to aid the model in logical reasoning and the learning of human preferences. Finally, experimental results indicate that the combination of these two processes can enhance both the model's judgment ability and its generative capacity. These findings deepen our comprehension of the creative capabilities of large language models (LLMs) and offer approaches to boost LLMs' creative abilities for cross-domain innovative applications.
Authors: Lu Qiu, Yi Chen, Yuying Ge, Yixiao Ge, Ying Shan, Xihui Liu
Abstract: The advent of Multimodal Large Language Models, leveraging the power of Large Language Models, has recently demonstrated superior multimodal understanding and reasoning abilities, heralding a new era for artificial general intelligence. However, achieving AGI necessitates more than just comprehension and reasoning. A crucial capability required is effective planning in diverse scenarios, which involves making reasonable decisions based on complex environments to solve real-world problems. Despite its importance, the planning abilities of current MLLMs in varied scenarios remain underexplored. In this paper, we introduce EgoPlan-Bench2, a rigorous and comprehensive benchmark designed to assess the planning capabilities of MLLMs across a wide range of real-world scenarios. EgoPlan-Bench2 encompasses everyday tasks spanning 4 major domains and 24 detailed scenarios, closely aligned with human daily life. EgoPlan-Bench2 is constructed through a semi-automatic process utilizing egocentric videos, complemented by manual verification. Grounded in a first-person perspective, it mirrors the way humans approach problem-solving in everyday life. We evaluate 21 competitive MLLMs and provide an in-depth analysis of their limitations, revealing that they face significant challenges in real-world planning. To further improve the planning proficiency of current MLLMs, we propose a training-free approach using multimodal Chain-of-Thought (CoT) prompting through investigating the effectiveness of various multimodal prompts in complex planning. Our approach enhances the performance of GPT-4V by 10.24 on EgoPlan-Bench2 without additional training. Our work not only sheds light on the current limitations of MLLMs in planning, but also provides insights for future enhancements in this critical area. We have made data and code available at https://qiulu66.github.io/egoplanbench2/.
Authors: Safayat Bin Hakim, Muhammad Adil, Alvaro Velasquez, Houbing Herbert Song
Abstract: In this paper, we propose an Adaptive Neuro-Symbolic Learning and Reasoning Framework for digital twin technology called ``ANSR-DT." Digital twins in industrial environments often struggle with interpretability, real-time adaptation, and human input integration. Our approach addresses these challenges by combining CNN-LSTM dynamic event detection with reinforcement learning and symbolic reasoning to enable adaptive intelligence with interpretable decision processes. This integration enhances environmental understanding while promoting continuous learning, leading to more effective real-time decision-making in human-machine collaborative applications. We evaluated ANSR-DT on synthetic industrial data, observing significant improvements over traditional approaches, with up to 99.5% accuracy for dynamic pattern recognition. The framework demonstrated superior adaptability with extended reinforcement learning training, improving explained variance from 0.447 to 0.547. Future work aims at scaling to larger datasets to test rule management beyond the current 14 rules. Our open-source implementation promotes reproducibility and establishes a foundation for future research in adaptive, interpretable digital twins for industrial applications.
Authors: Jaesik Yoon, Hyeonseo Cho, Doojin Baek, Yoshua Bengio, Sungjin Ahn
Abstract: Diffusion models have recently emerged as a powerful tool for planning. However, unlike Monte Carlo Tree Search (MCTS)-whose performance naturally improves with additional test-time computation (TTC), standard diffusion-based planners offer only limited avenues for TTC scalability. In this paper, we introduce Monte Carlo Tree Diffusion (MCTD), a novel framework that integrates the generative strength of diffusion models with the adaptive search capabilities of MCTS. Our method reconceptualizes denoising as a tree-structured process, allowing partially denoised plans to be iteratively evaluated, pruned, and refined. By selectively expanding promising trajectories while retaining the flexibility to revisit and improve suboptimal branches, MCTD achieves the benefits of MCTS such as controlling exploration-exploitation trade-offs within the diffusion framework. Empirical results on challenging long-horizon tasks show that MCTD outperforms diffusion baselines, yielding higher-quality solutions as TTC increases.
Authors: Vignesh Prabhakar, Md Amirul Islam, Adam Atanas, Yao-Ting Wang, Joah Han, Aastha Jhunjhunwala, Rucha Apte, Robert Clark, Kang Xu, Zihan Wang, Kai Liu
Abstract: Large Language Models (LLMs) have demonstrated remarkable potential in advancing scientific knowledge and addressing complex challenges. In this work, we introduce OmniScience, a specialized large reasoning model for general science, developed through three key components: (1) domain adaptive pretraining on a carefully curated corpus of scientific literature, (2) instruction tuning on a specialized dataset to guide the model in following domain-specific tasks, and (3) reasoning-based knowledge distillation through fine-tuning to significantly enhance its ability to generate contextually relevant and logically sound responses. We demonstrate the versatility of OmniScience by developing a battery agent that efficiently ranks molecules as potential electrolyte solvents or additives. Comprehensive evaluations reveal that OmniScience is competitive with state-of-the-art large reasoning models on the GPQA Diamond and domain-specific battery benchmarks, while outperforming all public reasoning and non-reasoning models with similar parameter counts. We further demonstrate via ablation experiments that domain adaptive pretraining and reasoning-based knowledge distillation are critical to attain our performance levels, across benchmarks.
Authors: Abdul Sittar, Simon M\"unker, Fabio Sartori, Andreas Reitenbach, Achim Rettinger, Michael M\"as, Alenka Gu\v{c}ek, Marko Grobelnik
Abstract: User engagement on social media platforms is influenced by historical context, time constraints, and reward-driven interactions. This study presents an agent-based simulation approach that models user interactions, considering past conversation history, motivation, and resource constraints. Utilizing German Twitter data on political discourse, we fine-tune AI models to generate posts and replies, incorporating sentiment analysis, irony detection, and offensiveness classification. The simulation employs a myopic best-response model to govern agent behavior, accounting for decision-making based on expected rewards. Our results highlight the impact of historical context on AI-generated responses and demonstrate how engagement evolves under varying constraints.
Authors: Danial Hooshyar, Eve Kikas, Yeongwook Yang, Gustav \v{S}\'ir, Raija H\"am\"al\"ainen, Tommi K\"arkk\"ainen, Roger Azevedo
Abstract: Given the demand for responsible and trustworthy AI for education, this study evaluates symbolic, sub-symbolic, and neural-symbolic AI (NSAI) in terms of generalizability and interpretability. Our extensive experiments on balanced and imbalanced self-regulated learning datasets of Estonian primary school students predicting 7th-grade mathematics national test performance showed that symbolic and sub-symbolic methods performed well on balanced data but struggled to identify low performers in imbalanced datasets. Interestingly, symbolic and sub-symbolic methods emphasized different factors in their decision-making: symbolic approaches primarily relied on cognitive and motivational factors, while sub-symbolic methods focused more on cognitive aspects, learnt knowledge, and the demographic variable of gender -- yet both largely overlooked metacognitive factors. The NSAI method, on the other hand, showed advantages by: (i) being more generalizable across both classes -- even in imbalanced datasets -- as its symbolic knowledge component compensated for the underrepresented class; and (ii) relying on a more integrated set of factors in its decision-making, including motivation, (meta)cognition, and learnt knowledge, thus offering a comprehensive and theoretically grounded interpretability framework. These contrasting findings highlight the need for a holistic comparison of AI methods before drawing conclusions based solely on predictive performance. They also underscore the potential of hybrid, human-centred NSAI methods to address the limitations of other AI families and move us closer to responsible AI for education. Specifically, by enabling stakeholders to contribute to AI design, NSAI aligns learned patterns with theoretical constructs, incorporates factors like motivation and metacognition, and strengthens the trustworthiness and responsibility of educational data mining.
Authors: Hamed Mahdavi, Alireza Hashemi, Majid Daliri, Pegah Mohammadipour, Alireza Farhadi, Samira Malek, Yekta Yazdanifard, Amir Khasahmadi, Vasant Honavar
Abstract: Recent advances in large language models (LLMs) have shown impressive progress in mathematical reasoning tasks. However, current evaluation benchmarks predominantly focus on the accuracy of final answers, often overlooking the crucial logical rigor for mathematical problem solving. The claim that state-of-the-art LLMs can solve Math Olympiad-level problems requires closer examination. To explore this, we conducted both qualitative and quantitative human evaluations of proofs generated by LLMs, and developed a schema for automatically assessing their reasoning capabilities. Our study reveals that current LLMs fall significantly short of solving challenging Olympiad-level problems and frequently fail to distinguish correct mathematical reasoning from clearly flawed solutions. Our analyses demonstrate that the occasional correct final answers provided by LLMs often result from pattern recognition or heuristic shortcuts rather than genuine mathematical reasoning. These findings underscore the substantial gap between LLM performance and human expertise in advanced mathematical reasoning and highlight the importance of developing benchmarks that prioritize the soundness of the reasoning used to arrive at an answer rather than the mere correctness of the final answers.
Authors: Peijie Yu, Yifan Yang, Jinjian Li, Zelong Zhang, Haorui Wang, Xiao Feng, Feng Zhang
Abstract: Large language models (LLMs) demonstrate strong potential as agents for tool invocation due to their advanced comprehension and planning capabilities. Users increasingly rely on LLM-based agents to solve complex missions through iterative interactions. However, existing benchmarks predominantly access agents in single-mission scenarios, failing to capture real-world complexity. To bridge this gap, we propose the Multi-Mission Tool Bench. In the benchmark, each test case comprises multiple interrelated missions. This design requires agents to dynamically adapt to evolving demands. Moreover, the proposed benchmark explores all possible mission-switching patterns within a fixed mission number. Specifically, we propose a multi-agent data generation framework to construct the benchmark. We also propose a novel method to evaluate the accuracy and efficiency of agent decisions with dynamic decision trees. Experiments on diverse open-source and closed-source LLMs reveal critical factors influencing agent robustness and provide actionable insights to the tool invocation society.
Authors: Yu Yue, Yufeng Yuan, Qiying Yu, Xiaochen Zuo, Ruofei Zhu, Wenyuan Xu, Jiaze Chen, Chengyi Wang, TianTian Fan, Zhengyin Du, Xiangpeng Wei, Xiangyu Yu, Gaohong Liu, Juncai Liu, Lingjun Liu, Haibin Lin, Zhiqi Lin, Bole Ma, Chi Zhang, Mofan Zhang, Wang Zhang, Hang Zhu, Ru Zhang, Xin Liu, Mingxuan Wang, Yonghui Wu, Lin Yan
Abstract: We present VAPO, Value-based Augmented Proximal Policy Optimization framework for reasoning models., a novel framework tailored for reasoning models within the value-based paradigm. Benchmarked the AIME 2024 dataset, VAPO, built on the Qwen 32B pre-trained model, attains a state-of-the-art score of $\mathbf{60.4}$. In direct comparison under identical experimental settings, VAPO outperforms the previously reported results of DeepSeek-R1-Zero-Qwen-32B and DAPO by more than 10 points. The training process of VAPO stands out for its stability and efficiency. It reaches state-of-the-art performance within a mere 5,000 steps. Moreover, across multiple independent runs, no training crashes occur, underscoring its reliability. This research delves into long chain-of-thought (long-CoT) reasoning using a value-based reinforcement learning framework. We pinpoint three key challenges that plague value-based methods: value model bias, the presence of heterogeneous sequence lengths, and the sparsity of reward signals. Through systematic design, VAPO offers an integrated solution that effectively alleviates these challenges, enabling enhanced performance in long-CoT reasoning tasks.
Authors: Chenrui Fan, Ming Li, Lichao Sun, Tianyi Zhou
Abstract: We find that the response length of reasoning LLMs, whether trained by reinforcement learning or supervised learning, drastically increases for ill-posed questions with missing premises (MiP), ending up with redundant and ineffective thinking. This newly introduced scenario exacerbates the general overthinking issue to a large extent, which we name as the MiP-Overthinking. Such failures are against the ``test-time scaling law'' but have been widely observed on multiple datasets we curated with MiP, indicating the harm of cheap overthinking and a lack of critical thinking. Surprisingly, LLMs not specifically trained for reasoning exhibit much better performance on the MiP scenario, producing much shorter responses that quickly identify ill-posed queries. This implies a critical flaw of the current training recipe for reasoning LLMs, which does not encourage efficient thinking adequately, leading to the abuse of thinking patterns. To further investigate the reasons behind such failures, we conduct fine-grained analyses of the reasoning length, overthinking patterns, and location of critical thinking on different types of LLMs. Moreover, our extended ablation study reveals that the overthinking is contagious through the distillation of reasoning models' responses. These results improve the understanding of overthinking and shed novel insights into mitigating the problem.
Authors: Kostas Hatalis, Despina Christou, Vyshnavi Kondapalli
Abstract: Agents powered by Large Language Models (LLMs) have recently demonstrated impressive capabilities in various tasks. Still, they face limitations in tasks requiring specific, structured knowledge, flexibility, or accountable decision-making. While agents are capable of perceiving their environments, forming inferences, planning, and executing actions towards goals, they often face issues such as hallucinations and lack of contextual memory across interactions. This paper explores how Case-Based Reasoning (CBR), a strategy that solves new problems by referencing past experiences, can be integrated into LLM agent frameworks. This integration allows LLMs to leverage explicit knowledge, enhancing their effectiveness. We systematically review the theoretical foundations of these enhanced agents, identify critical framework components, and formulate a mathematical model for the CBR processes of case retrieval, adaptation, and learning. We also evaluate CBR-enhanced agents against other methods like Chain-of-Thought reasoning and standard Retrieval-Augmented Generation, analyzing their relative strengths. Moreover, we explore how leveraging CBR's cognitive dimensions (including self-reflection, introspection, and curiosity) via goal-driven autonomy mechanisms can further enhance the LLM agent capabilities. Contributing to the ongoing research on neuro-symbolic hybrid systems, this work posits CBR as a viable technique for enhancing the reasoning skills and cognitive aspects of autonomous LLM agents.
Authors: Zen Kit Heng, Zimeng Zhao, Tianhao Wu, Yuanfei Wang, Mingdong Wu, Yangang Wang, Hao Dong
Abstract: Large Language Models (LLMs) are emerging as promising tools for automated reinforcement learning (RL) reward design, owing to their robust capabilities in commonsense reasoning and code generation. By engaging in dialogues with RL agents, LLMs construct a Reward Observation Space (ROS) by selecting relevant environment states and defining their internal operations. However, existing frameworks have not effectively leveraged historical exploration data or manual task descriptions to iteratively evolve this space. In this paper, we propose a novel heuristic framework that enhances LLM-driven reward design by evolving the ROS through a table-based exploration caching mechanism and a text-code reconciliation strategy. Our framework introduces a state execution table, which tracks the historical usage and success rates of environment states, overcoming the Markovian constraint typically found in LLM dialogues and facilitating more effective exploration. Furthermore, we reconcile user-provided task descriptions with expert-defined success criteria using structured prompts, ensuring alignment in reward design objectives. Comprehensive evaluations on benchmark RL tasks demonstrate the effectiveness and stability of the proposed framework. Code and video demos are available at jingjjjjjie.github.io/LLM2Reward.
Authors: Xin Luo, Fang-Yi Liang, Jiale Liu, Yu-Wei Zhan, Zhen-Duo Chen, Xin-Shun Xu
Abstract: As Web technology continues to develop, it has become increasingly common to use data stored on different clients. At the same time, federated learning has received widespread attention due to its ability to protect data privacy when let models learn from data which is distributed across various clients. However, most existing works assume that the client's data are fixed. In real-world scenarios, such an assumption is most likely not true as data may be continuously generated and new classes may also appear. To this end, we focus on the practical and challenging federated class-incremental learning (FCIL) problem. For FCIL, the local and global models may suffer from catastrophic forgetting on old classes caused by the arrival of new classes and the data distributions of clients are non-independent and identically distributed (non-iid). In this paper, we propose a novel method called Federated Class-Incremental Learning with PrompTing (FCILPT). Given the privacy and limited memory, FCILPT does not use a rehearsal-based buffer to keep exemplars of old data. We choose to use prompts to ease the catastrophic forgetting of the old classes. Specifically, we encode the task-relevant and task-irrelevant knowledge into prompts, preserving the old and new knowledge of the local clients and solving the problem of catastrophic forgetting. We first sort the task information in the prompt pool in the local clients to align the task information on different clients before global aggregation. It ensures that the same task's knowledge are fully integrated, solving the problem of non-iid caused by the lack of classes among different clients in the same incremental task. Experiments on CIFAR-100, Mini-ImageNet, and Tiny-ImageNet demonstrate that FCILPT achieves significant accuracy improvements over the state-of-the-art methods.
Authors: Rose E. Guingrich, Michael S. A. Graziano
Abstract: As artificial intelligence (AI) becomes more widespread, one question that arises is how human-AI interaction might impact human-human interaction. Chatbots, for example, are increasingly used as social companions, and while much is speculated, little is known empirically about how their use impacts human relationships. A common hypothesis is that relationships with companion chatbots are detrimental to social health by harming or replacing human interaction, but this hypothesis may be too simplistic, especially considering the social needs of users and the health of their preexisting human relationships. To understand how relationships with companion chatbots impact social health, we studied people who regularly used companion chatbots and people who did not use them. Contrary to expectations, companion chatbot users indicated that these relationships were beneficial to their social health, whereas non-users viewed them as harmful. Another common assumption is that people perceive conscious, humanlike AI as disturbing and threatening. Among both users and non-users, however, we found the opposite: perceiving companion chatbots as more conscious and humanlike correlated with more positive opinions and more pronounced social health benefits. Detailed accounts from users suggested that these humanlike chatbots may aid social health by supplying reliable and safe interactions, without necessarily harming human relationships, but this may depend on users' preexisting social needs and how they perceive both human likeness and mind in the chatbot.
Authors: Georgi Ganev, Emiliano De Cristofaro
Abstract: Generative models producing synthetic data are meant to provide a privacy-friendly approach to releasing data. However, their privacy guarantees are only considered robust when models satisfy Differential Privacy (DP). Alas, this is not a ubiquitous standard, as many leading companies (and, in fact, research papers) use ad-hoc privacy metrics based on testing the statistical similarity between synthetic and real data. In this paper, we examine the privacy metrics used in real-world synthetic data deployments and demonstrate their unreliability in several ways. First, we provide counter-examples where severe privacy violations occur even if the privacy tests pass and instantiate accurate membership and attribute inference attacks with minimal cost. We then introduce ReconSyn, a reconstruction attack that generates multiple synthetic datasets that are considered private by the metrics but actually leak information unique to individual records. We show that ReconSyn recovers 78-100% of the outliers in the train data with only black-box access to a single fitted generative model and the privacy metrics. In the process, we show that applying DP only to the model does not mitigate this attack, as using privacy metrics breaks the end-to-end DP pipeline.
Authors: Di Cooke, Abigail Edwards, Sophia Barkoff, Kathryn Kelly
Abstract: One of the current principal defenses against weaponized synthetic media continues to be the ability of the targeted individual to visually or auditorily recognize AI-generated content when they encounter it. However, as the realism of synthetic media continues to rapidly improve, it is vital to have an accurate understanding of just how susceptible people currently are to potentially being misled by convincing but false AI generated content. We conducted a perceptual study with 1276 participants to assess how capable people were at distinguishing between authentic and synthetic images, audio, video, and audiovisual media. We find that on average, people struggled to distinguish between synthetic and authentic media, with the mean detection performance close to a chance level performance of 50%. We also find that accuracy rates worsen when the stimuli contain any degree of synthetic content, features foreign languages, and the media type is a single modality. People are also less accurate at identifying synthetic images when they feature human faces, and when audiovisual stimuli have heterogeneous authenticity. Finally, we find that higher degrees of prior knowledgeability about synthetic media does not significantly impact detection accuracy rates, but age does, with older individuals performing worse than their younger counterparts. Collectively, these results highlight that it is no longer feasible to rely on the perceptual capabilities of people to protect themselves against the growing threat of weaponized synthetic media, and that the need for alternative countermeasures is more critical than ever before.
Authors: Yash Saxena, Deepa Tilwani, Ali Mohammadi, Edward Raff, Amit Sheth, Srinivasan Parthasarathy, Manas Gaur
Abstract: Large language models (LLMs) present a promising yet challenging frontier for automated source citation in scientific communication. Previous approaches to citation generation have been limited by citation ambiguity and LLM overgeneralization. We introduce REASONS, a novel dataset with sentence-level annotations across 12 scientific domains from arXiv. Our evaluation framework covers two key citation scenarios: indirect queries (matching sentences to paper titles) and direct queries (author attribution), both enhanced with contextual metadata. We conduct extensive experiments with models such as GPT-O1, GPT-4O, GPT-3.5, DeepSeek, and other smaller models like Perplexity AI (7B). While top-tier LLMs achieve high performance in sentence attribution, they struggle with high hallucination rates, a key metric for scientific reliability. Our metadata-augmented approach reduces hallucination rates across all tasks, offering a promising direction for improvement. Retrieval-augmented generation (RAG) with Mistral improves performance in indirect queries, reducing hallucination rates by 42% and maintaining competitive precision with larger models. However, adversarial testing highlights challenges in linking paper titles to abstracts, revealing fundamental limitations in current LLMs. REASONS provides a challenging benchmark for developing reliable and trustworthy LLMs in scientific applications
Authors: Zhiwei Li, Guodong Long, Chunxu Zhang, Honglei Zhang, Jing Jiang, Chengqi Zhang
Abstract: Federated Recommendation Systems (FRSs) offer a privacy-preserving alternative to traditional centralized approaches by decentralizing data storage. However, they face persistent challenges such as data sparsity and heterogeneity, largely due to isolated client environments. Recent advances in Foundation Models (FMs), particularly large language models like ChatGPT, present an opportunity to surmount these issues through powerful, cross-task knowledge transfer. In this position paper, we systematically examine the convergence of FRSs and FMs, illustrating how FM-enhanced frameworks can substantially improve client-side personalization, communication efficiency, and server-side aggregation. We also delve into pivotal challenges introduced by this integration, including privacy-security trade-offs, non-IID data, and resource constraints in federated setups, and propose prospective research directions in areas such as multimodal recommendation, real-time FM adaptation, and explainable federated reasoning. By unifying FRSs with FMs, our position paper provides a forward-looking roadmap for advancing privacy-preserving, high-performance recommendation systems that fully leverage large-scale pre-trained knowledge to enhance local performance.
Authors: Zeyu Gao, Yao Mu, Jinye Qu, Mengkang Hu, Shijia Peng, Chengkai Hou, Lingyue Guo, Ping Luo, Shanghang Zhang, Yanfeng Lu
Abstract: Dual-arm robots offer enhanced versatility and efficiency over single-arm counterparts by enabling concurrent manipulation of multiple objects or cooperative execution of tasks using both arms. However, the coordination of dual-arm systems for long-horizon tasks continues to pose significant challenges, stemming from the intricate temporal and spatial dependencies among sub-tasks, necessitating intelligent decisions regarding the allocation of actions between arms and their optimal execution order. Existing task planning methods predominantly focus on single-arm robots or rely on predefined bimanual operations to use large language models (LLMs) generate task sequence with linear temporal dependency, failing to fully leverage the capabilities of dual-arm systems. To address this limitation, we introduce DAG-Plan, a structured task planning framework tailored for dual-arm robots. DAG-Plan harnesses LLMs to decompose intricate tasks into actionable sub-tasks represented as nodes within a directed acyclic graph (DAG). Critically, DAG-Plan dynamically assigns these sub-tasks to the appropriate arm based on real-time environmental observations, enabling parallel and adaptive execution. We evaluate DAG-Plan on the Dual-Arm Kitchen Benchmark, comprising 5 sequential tasks with 44 sub-tasks. Extensive experiments demonstrate the superiority of DAG-Plan over directly using LLM to generate linear task sequence, achieving 52.8% higher efficiency compared to the single-arm task planning and 48% higher success rate of the dual-arm task planning. Compared to iterative methods, DAG-Plan improving execution efficiency 84.1% due to its fewer query time. More demos and information are available on https://sites.google.com/view/dag-plan.
Authors: Jing Gong, Yanghui Wu, Linxi Liang, Yanlin Wang, Jiachi Chen, Mingwei Liu, Zibin Zheng
Abstract: Semantic code search, retrieving code that matches a given natural language query, is an important task to improve productivity in software engineering. Existing code search datasets face limitations: they rely on human annotators who assess code primarily through semantic understanding rather than functional verification, leading to potential inaccuracies and scalability issues. Additionally, current evaluation metrics often overlook the multi-choice nature of code search. This paper introduces CoSQA+, pairing high-quality queries from CoSQA with multiple suitable codes. We develop an automated pipeline featuring multiple model-based candidate selections and the novel test-driven agent annotation system. Among a single Large Language Model (LLM) annotator and Python expert annotators (without test-based verification), agents leverage test-based verification and achieve the highest accuracy of 92.0%. Through extensive experiments, CoSQA+ has demonstrated superior quality over CoSQA. Models trained on CoSQA+ exhibit improved performance. We provide the code and data at https://github.com/DeepSoftwareAnalytics/CoSQA_Plus.
Authors: Jose Luis Ponton, Eduard Pujol, Andreas Aristidou, Carlos Andujar, Nuria Pelechano
Abstract: High-quality motion reconstruction that follows the user's movements can be achieved by high-end mocap systems with many sensors. However, obtaining such animation quality with fewer input devices is gaining popularity as it brings mocap closer to the general public. The main challenges include the loss of end-effector accuracy in learning-based approaches, or the lack of naturalness and smoothness in IK-based solutions. In addition, such systems are often finely tuned to a specific number of trackers and are highly sensitive to missing data e.g., in scenarios where a sensor is occluded or malfunctions. In response to these challenges, we introduce DragPoser, a novel deep-learning-based motion reconstruction system that accurately represents hard and dynamic on-the-fly constraints, attaining real-time high end-effectors position accuracy. This is achieved through a pose optimization process within a structured latent space. Our system requires only one-time training on a large human motion dataset, and then constraints can be dynamically defined as losses, while the pose is iteratively refined by computing the gradients of these losses within the latent space. To further enhance our approach, we incorporate a Temporal Predictor network, which employs a Transformer architecture to directly encode temporality within the latent space. This network ensures the pose optimization is confined to the manifold of valid poses and also leverages past pose data to predict temporally coherent poses. Results demonstrate that DragPoser surpasses both IK-based and the latest data-driven methods in achieving precise end-effector positioning, while it produces natural poses and temporally coherent motion. In addition, our system showcases robustness against on-the-fly constraint modifications, and exhibits exceptional adaptability to various input configurations and changes.
Authors: Asankhaya Sharma
Abstract: This paper introduces Patched Round-Trip Correctness (Patched RTC), a novel evaluation technique for Large Language Models (LLMs) applied to diverse software development tasks, particularly focusing on "outer loop" activities such as bug fixing, code review, and documentation updates. Patched RTC extends the original Round-Trip Correctness method to work with any LLM and downstream task, offering a self-evaluating framework that measures consistency and robustness of model responses without human intervention. The study demonstrates a correlation between Patched RTC scores and task-specific accuracy metrics, presenting it as an alternative to the LLM-as-Judge paradigm for open-domain task evaluation. We implement Patched RTC in an open-source framework called patchwork, allowing for transparent evaluation during inference across various patchflows. Experiments comparing GPT-3.5 and GPT-4 models across different software development tasks reveal that Patched RTC effectively distinguishes model performance and task difficulty. The paper also explores the impact of consistency prompts on improving model accuracy, suggesting that Patched RTC can guide prompt refinement and model selection for complex software development workflows.
Authors: Ning Wang, Bingkun Yao, Jie Zhou, Xi Wang, Zhe Jiang, Nan Guan
Abstract: Recent advancements in large language models (LLMs) have sparked significant interest in the automatic generation of Register Transfer Level (RTL) designs, particularly using Verilog. Current research on this topic primarily focuses on pre-training and instruction tuning, but the effectiveness of these methods is constrained by the limited availability of training data, as public Verilog code is far less abundant than software code. In particular, these methods struggle to effectively capture Verilog parallel code structures, which fundamentally differ from the imperative, sequential control flow typical in most software programming languages. This paper introduces VeriSeek, an LLM enhanced by reinforcement learning using a limited amount of high-quality training data to achieve high Verilog code generation performance. Our reinforcement learning approach employs code structure information as feedback signals to refine the pre-trained model, enabling it to effectively learn important patterns from Verilog code with parallel structures. Experiments show that VeriSeek outperforms state-of-the-art methods across multiple benchmarks.
Authors: Asankhaya Sharma
Abstract: This paper introduces Patched MOA (Mixture of Agents), an inference optimization technique that significantly enhances the performance of large language models (LLMs) across diverse software development tasks. We evaluate three inference optimization algorithms - Best of N, Mixture of Agents, and Monte Carlo Tree Search and demonstrate that Patched MOA can boost the performance of smaller models to surpass that of larger, more expensive models. Notably, our approach improves the gpt-4o-mini model's performance on the Arena-Hard-Auto benchmark by 15.52%, outperforming gpt-4-turbo at a fraction of the cost. We also apply Patched MOA to various software development workflows, showing consistent improvements in task completion rates. Our method is model-agnostic, transparent to end-users, and can be easily integrated into existing LLM pipelines. This work contributes to the growing field of LLM optimization, offering a cost-effective solution for enhancing model performance without the need for fine-tuning or larger models. Our implementation is open-source and available at https://github.com/codelion/optillm.
Authors: Cyril Shih-Huan Hsu, Danny De Vleeschauwer, Chrysa Papagianni, Paola Grosso
Abstract: When a network slice spans multiple technology domains, it is crucial for each domain to uphold the End-to-End (E2E) Service Level Agreement (SLA) associated with the slice. Consequently, the E2E SLA must be properly decomposed into partial SLAs that are assigned to each domain involved. In a network slice management system with a two-level architecture, comprising an E2E service orchestrator and local domain controllers, we consider that the orchestrator has access only to historical data regarding the responses of local controllers to previous requests, and this information is used to construct a risk model for each domain. In this study, we extend our previous work by investigating the dynamic nature of real-world systems and introducing an online learning-decomposition framework to tackle the dynamicity. We propose a framework that continuously updates the risk models based on the most recent feedback. This approach leverages key components such as online gradient descent and FIFO memory buffers, which enhance the stability and robustness of the overall process. Our empirical study on an analytic model-based simulator demonstrates that the proposed framework outperforms the state-of-the-art static approach, delivering more accurate and resilient SLA decomposition under varying conditions and data limitations. Furthermore, we provide a comprehensive complexity analysis of the proposed solution.
Authors: Hongliang Zhong, Can Wang, Jingbo Zhang, Jing Liao
Abstract: Generating and inserting new objects into 3D content is a compelling approach for achieving versatile scene recreation. Existing methods, which rely on SDS optimization or single-view inpainting, often struggle to produce high-quality results. To address this, we propose a novel method for object insertion in 3D content represented by Gaussian Splatting. Our approach introduces a multi-view diffusion model, dubbed MVInpainter, which is built upon a pre-trained stable video diffusion model to facilitate view-consistent object inpainting. Within MVInpainter, we incorporate a ControlNet-based conditional injection module to enable controlled and more predictable multi-view generation. After generating the multi-view inpainted results, we further propose a mask-aware 3D reconstruction technique to refine Gaussian Splatting reconstruction from these sparse inpainted views. By leveraging these fabricate techniques, our approach yields diverse results, ensures view-consistent and harmonious insertions, and produces better object quality. Extensive experiments demonstrate that our approach outperforms existing methods.
Authors: Yu Fu, Jie He, Yifan Yang, Qun Liu, Deyi Xiong
Abstract: Meta learning has been widely used to exploit rich-resource source tasks to improve the performance of low-resource target tasks. Unfortunately, most existing meta learning approaches treat different source tasks equally, ignoring the relatedness of source tasks to the target task in knowledge transfer. To mitigate this issue, we propose a reinforcement-based multi-source meta-transfer learning framework (Meta-RTL) for low-resource commonsense reasoning. In this framework, we present a reinforcement-based approach to dynamically estimating source task weights that measure the contribution of the corresponding tasks to the target task in the meta-transfer learning. The differences between the general loss of the meta model and task-specific losses of source-specific temporal meta models on sampled target data are fed into the policy network of the reinforcement learning module as rewards. The policy network is built upon LSTMs that capture long-term dependencies on source task weight estimation across meta learning iterations. We evaluate the proposed Meta-RTL using both BERT and ALBERT as the backbone of the meta model on three commonsense reasoning benchmark datasets. Experimental results demonstrate that Meta-RTL substantially outperforms strong baselines and previous task selection strategies and achieves larger improvements on extremely low-resource settings.
Authors: Guixian Zhang, Guan Yuan, Debo Cheng, Lin Liu, Jiuyong Li, Shichao Zhang
Abstract: The rapid development of Large Language Models (LLMs) creates new opportunities for recommender systems, especially by exploiting the side information (e.g., descriptions and analyses of items) generated by these models. However, aligning this side information with collaborative information from historical interactions poses significant challenges. The inherent biases within LLMs can skew recommendations, resulting in distorted and potentially unfair user experiences. On the other hand, propensity bias causes side information to be aligned in such a way that it often tends to represent all inputs in a low-dimensional subspace, leading to a phenomenon known as dimensional collapse, which severely restricts the recommender system's ability to capture user preferences and behaviours. To address these issues, we introduce a novel framework named Counterfactual LLM Recommendation (CLLMR). Specifically, we propose a spectrum-based side information encoder that implicitly embeds structural information from historical interactions into the side information representation, thereby circumventing the risk of dimension collapse. Furthermore, our CLLMR approach explores the causal relationships inherent in LLM-based recommender systems. By leveraging counterfactual inference, we counteract the biases introduced by LLMs. Extensive experiments demonstrate that our CLLMR approach consistently enhances the performance of various recommender models.
Authors: Fabian Akkerman, Nils Knofius, Matthieu van der Heijden, Martijn Mes
Abstract: This paper investigates dual sourcing problems with supply mode dependent failure rates, particularly relevant in managing spare parts for downtime-critical assets. To enhance resilience, businesses increasingly adopt dual sourcing strategies using both conventional and additive manufacturing techniques. This paper explores how these strategies can optimise sourcing by addressing variations in part properties and failure rates. A significant challenge is the distinct failure characteristics of parts produced by these methods, which influence future demand. To tackle this, we propose a new iterative heuristic and several reinforcement learning techniques combined with an endogenous parameterised learning (EPL) approach. This EPL approach - compatible with any learning method - allows a single policy to handle various input parameters for multiple items. In a stylised setting, our best policy achieves an average optimality gap of 0.4%. In a case study within the energy sector, our policies outperform the baseline in 91.1% of instances, yielding average cost savings up to 22.6%.
Authors: Will Dorrell, Kyle Hsu, Luke Hollingsworth, Jin Hwa Lee, Jiajun Wu, Chelsea Finn, Peter E Latham, Tim EJ Behrens, James CR Whittington
Abstract: Why do biological and artificial neurons sometimes modularise, each encoding a single meaningful variable, and sometimes entangle their representation of many variables? In this work, we develop a theory of when biologically inspired networks -- those that are nonnegative and energy efficient -- modularise their representation of source variables (sources). We derive necessary and sufficient conditions on a sample of sources that determine whether the neurons in an optimal biologically-inspired linear autoencoder modularise. Our theory applies to any dataset, extending far beyond the case of statistical independence studied in previous work. Rather we show that sources modularise if their support is ``sufficiently spread''. From this theory, we extract and validate predictions in a variety of empirical studies on how data distribution affects modularisation in nonlinear feedforward and recurrent neural networks trained on supervised and unsupervised tasks. Furthermore, we apply these ideas to neuroscience data, showing that range independence can be used to understand the mixing or modularising of spatial and reward information in entorhinal recordings in seemingly conflicting experiments. Further, we use these results to suggest alternate origins of mixed-selectivity, beyond the predominant theory of flexible nonlinear classification. In sum, our theory prescribes precise conditions on when neural activities modularise, providing tools for inducing and elucidating modular representations in brains and machines.
Authors: Adonisz Dimitriu, Tam\'as Michaletzky, Viktor Remeli
Abstract: Adversarial attacks threaten the reliability of machine learning models in critical applications like autonomous vehicles and defense systems. As object detectors become more robust with models like YOLOv8, developing effective adversarial methodologies is increasingly challenging. We present Truck Adversarial Camouflage Optimization (TACO), a novel framework that generates adversarial camouflage patterns on 3D vehicle models to deceive state-of-the-art object detectors. Adopting Unreal Engine 5, TACO integrates differentiable rendering with a Photorealistic Rendering Network to optimize adversarial textures targeted at YOLOv8. To ensure the generated textures are both effective in deceiving detectors and visually plausible, we introduce the Convolutional Smooth Loss function, a generalized smooth loss function. Experimental evaluations demonstrate that TACO significantly degrades YOLOv8's detection performance, achieving an AP@0.5 of 0.0099 on unseen test data. Furthermore, these adversarial patterns exhibit strong transferability to other object detection models such as Faster R-CNN and earlier YOLO versions.
Authors: Hao Phung, Quan Dao, Trung Dao, Hoang Phan, Dimitris Metaxas, Anh Tran
Abstract: We introduce a novel state-space architecture for diffusion models, effectively harnessing spatial and frequency information to enhance the inductive bias towards local features in input images for image generation tasks. While state-space networks, including Mamba, a revolutionary advancement in recurrent neural networks, typically scan input sequences from left to right, they face difficulties in designing effective scanning strategies, especially in the processing of image data. Our method demonstrates that integrating wavelet transformation into Mamba enhances the local structure awareness of visual inputs and better captures long-range relations of frequencies by disentangling them into wavelet subbands, representing both low- and high-frequency components. These wavelet-based outputs are then processed and seamlessly fused with the original Mamba outputs through a cross-attention fusion layer, combining both spatial and frequency information to optimize the order awareness of state-space models which is essential for the details and overall quality of image generation. Besides, we introduce a globally-shared transformer to supercharge the performance of Mamba, harnessing its exceptional power to capture global relationships. Through extensive experiments on standard benchmarks, our method demonstrates superior results compared to DiT and DIFFUSSM, achieving faster training convergence and delivering high-quality outputs. The codes and pretrained models are released at https://github.com/VinAIResearch/DiMSUM.git.
Authors: Zongjian Li, Bin Lin, Yang Ye, Liuhan Chen, Xinhua Cheng, Shenghai Yuan, Li Yuan
Abstract: Video Variational Autoencoder (VAE) encodes videos into a low-dimensional latent space, becoming a key component of most Latent Video Diffusion Models (LVDMs) to reduce model training costs. However, as the resolution and duration of generated videos increase, the encoding cost of Video VAEs becomes a limiting bottleneck in training LVDMs. Moreover, the block-wise inference method adopted by most LVDMs can lead to discontinuities of latent space when processing long-duration videos. The key to addressing the computational bottleneck lies in decomposing videos into distinct components and efficiently encoding the critical information. Wavelet transform can decompose videos into multiple frequency-domain components and improve the efficiency significantly, we thus propose Wavelet Flow VAE (WF-VAE), an autoencoder that leverages multi-level wavelet transform to facilitate low-frequency energy flow into latent representation. Furthermore, we introduce a method called Causal Cache, which maintains the integrity of latent space during block-wise inference. Compared to state-of-the-art video VAEs, WF-VAE demonstrates superior performance in both PSNR and LPIPS metrics, achieving 2x higher throughput and 4x lower memory consumption while maintaining competitive reconstruction quality. Our code and models are available at https://github.com/PKU-YuanGroup/WF-VAE.
Authors: Gregory Beylkin
Abstract: We present an unconditionally stable algorithm for applying matrix transfer function of a linear time invariant system (LTI) in time domain. The state matrix of an LTI system used for modeling long range dependencies in state space models (SSMs) has eigenvalues close to $1$. The standard recursion defining LTI system becomes unstable if the $m\times m$ state matrix has just one eigenvalue with absolute value even slightly greater than 1. This may occur when approximating a state matrix by a structured matrix to reduce the cost of matrix-vector multiplication from $\mathcal{O}\left(m^{2}\right)$ to $\mathcal{O}\left(m\right)$ or $\mathcal{O}\left(m\log m\right).$ We introduce an unconditionally stable algorithm that uses an approximation of the rational transfer function in the z-domain by a matrix polynomial of degree $2^{N+1}-1$, where $N$ is chosen to achieve any user-selected accuracy. Using a cascade implementation in time domain, applying such transfer function to compute $L$ states requires no more than $2L$ matrix-vector multiplications (whereas the standard recursion requires $L$ matrix-vector multiplications). However, using unconditionally stable algorithm, it is not necessary to assure that an approximate state matrix has all eigenvalues with absolute values strictly less than 1 i.e., within the desired accuracy, the absolute value of some eigenvalues may possibly exceed $1$. Consequently, this algorithm allows one to use a wider variety of structured approximations to reduce the cost of matrix-vector multiplication and we briefly describe several of them to be used for this purpose.
Authors: Maxwell A. Xu, Jaya Narain, Gregory Darnell, Haraldur Hallgrimsson, Hyewon Jeong, Darren Forde, Richard Fineman, Karthik J. Raghuram, James M. Rehg, Shirley Ren
Abstract: We present RelCon, a novel self-supervised Relative Contrastive learning approach for training a motion foundation model from wearable accelerometry sensors. First, a learnable distance measure is trained to capture motif similarity and domain-specific semantic information such as rotation invariance. Then, the learned distance provides a measurement of semantic similarity between a pair of accelerometry time-series, which we use to train our foundation model to model relative relationships across time and across subjects. The foundation model is trained on 1 billion segments from 87,376 participants, and achieves state-of-the-art performance across multiple downstream tasks, including human activity recognition and gait metric regression. To our knowledge, we are the first to show the generalizability of a foundation model with motion data from wearables across distinct evaluation tasks.
Authors: Haotian Ye, Axel Wisiorek, Antonis Maronikolakis, \"Ozge Ala\c{c}am, Hinrich Sch\"utze
Abstract: Hate speech online remains an understudied issue for marginalized communities, particularly in the Global South, which includes developing societies with increasing internet penetration. In this paper, we aim to provide marginalized communities in societies where the dominant language is low-resource with a privacy-preserving tool to protect themselves from online hate speech by filtering offensive content in their native languages. Our contributions are twofold: 1) we release REACT (REsponsive hate speech datasets Across ConTexts), a collection of high-quality, culture-specific hate speech detection datasets comprising multiple target groups and low-resource languages, curated by experienced data collectors; 2) we propose a few-shot hate speech detection approach based on federated learning (FL), a privacy-preserving method for collaboratively training a central model that exhibits robustness when tackling different target groups and languages. By keeping training local to user devices, we ensure data privacy while leveraging the collective learning benefits of FL. Furthermore, we explore personalized client models tailored to specific target groups and evaluate their performance. Our findings indicate the overall effectiveness of FL across different target groups, and point to personalization as a promising direction.
Authors: Pu Zhao, Xuan Shen, Zhenglun Kong, Yixin Shen, Sung-En Chang, Timothy Rupprecht, Lei Lu, Enfu Nan, Changdi Yang, Yumei He, Weiyan Shi, Xingchen Xu, Yu Huang, Wei Jiang, Wei Wang, Yue Chen, Yong He, Yanzhi Wang
Abstract: Recently, Large Language Models (LLMs) have undergone a significant transformation, marked by a rapid rise in both their popularity and capabilities. Leading this evolution are proprietary LLMs like GPT-4 and GPT-o1, which have captured widespread attention in the AI community due to their remarkable performance and versatility. Simultaneously, open-source LLMs, such as LLaMA, have made great contributions to the ever-increasing popularity of LLMs due to the ease to customize and deploy the models across diverse applications. Although open-source LLMs present unprecedented opportunities for innovation and research, the commercialization of LLMs has raised concerns about transparency, reproducibility, and safety. Many open-source LLMs fail to meet fundamental transparency requirements by withholding essential components like training code and data, which may hinder further innovations on LLMs. To mitigate this issue, we introduce Moxin 7B, a fully open-source LLM developed, adhering to principles of open science, open source, open data, and open access. We release the pre-training code and configurations, training and fine-tuning datasets, and intermediate and final checkpoints, aiming to make continuous commitments to fully open-source LLMs. After pre-training and obtaining the base model, we finetune the Moxin Base model with SOTA post-training framework and instruction data to obtain Moxin Instruct model. To improve the reasoning capability, we further finetune our Instruct model with chain-of-thought data distilled from DeepSeek R1, and then use Group Relative Policy Optimization (GRPO), an efficient and effective reinforcement learning algorithm following DeepSeek R1, to finetune our model, leading to the Moxin Reasoning model. Experiments show that our models achieve superior performance in various evaluations such as zero-shot evaluation, few-shot evaluation, and CoT evaluation.
Authors: Zhipeng Wang, Rui Sun, Elizabeth Lui, Tuo Zhou, Yizhe Wen, Jiahao Sun
Abstract: The rapid advancement of AI has underscored critical challenges in its development and implementation, largely due to centralized control by a few major corporations. This concentration of power intensifies biases within AI models, resulting from inadequate governance and oversight mechanisms. Additionally, it limits public involvement and heightens concerns about the integrity of model generation. Such monopolistic control over data and AI outputs threatens both innovation and fair data usage, as users inadvertently contribute data that primarily benefits these corporations. In this work, we propose AIArena, a blockchain-based decentralized AI training platform designed to democratize AI development and alignment through on-chain incentive mechanisms. AIArena fosters an open and collaborative environment where participants can contribute models and computing resources. Its on-chain consensus mechanism ensures fair rewards for participants based on their contributions. We instantiate and implement AIArena on the public Base blockchain Sepolia testnet, and the evaluation results demonstrate the feasibility of AIArena in real-world applications.
Authors: Tananun Songdechakraiwut, Yutong Wu
Abstract: The human brain is a complex system, and understanding its mechanisms has been a long-standing challenge in neuroscience. The study of the functional connectome, which maps the functional connections between different brain regions, has provided valuable insights through various advanced analysis techniques developed over the years. Similarly, neural networks, inspired by the brain's architecture, have achieved notable success in diverse applications but are often noted for their lack of interpretability. In this paper, we propose a novel approach that bridges neural networks and human brain functions by leveraging brain-inspired techniques. Our approach, grounded in the insights from the functional connectome, offers scalable ways to characterize topology of large neural networks using stable statistical and machine learning techniques. Our empirical analysis demonstrates its capability to enhance the interpretability of neural networks, providing a deeper understanding of their underlying mechanisms.
Authors: Sieun Hyeon, Kyudan Jung, Jaehee Won, Nam-Joon Kim, Hyun Gon Ryu, Hyuk-Jae Lee, Jaeyoung Do
Abstract: In various academic and professional settings, such as mathematics lectures or research presentations, it is often necessary to convey mathematical expressions orally. However, reading mathematical expressions aloud without accompanying visuals can significantly hinder comprehension, especially for those who are hearing-impaired or rely on subtitles due to language barriers. For instance, when a presenter reads Euler's Formula, current Automatic Speech Recognition (ASR) models often produce a verbose and error-prone textual description (e.g., e to the power of i x equals cosine of x plus i $\textit{side}$ of x), instead of the concise $\LaTeX{}$ format (i.e., $ e^{ix} = \cos(x) + i\sin(x) $), which hampers clear understanding and communication. To address this issue, we introduce MathSpeech, a novel pipeline that integrates ASR models with small Language Models (sLMs) to correct errors in mathematical expressions and accurately convert spoken expressions into structured $\LaTeX{}$ representations. Evaluated on a new dataset derived from lecture recordings, MathSpeech demonstrates $\LaTeX{}$ generation capabilities comparable to leading commercial Large Language Models (LLMs), while leveraging fine-tuned small language models of only 120M parameters. Specifically, in terms of CER, BLEU, and ROUGE scores for $\LaTeX{}$ translation, MathSpeech demonstrated significantly superior capabilities compared to GPT-4o. We observed a decrease in CER from 0.390 to 0.298, and higher ROUGE/BLEU scores compared to GPT-4o.
Authors: Yousef Yeganeh, Azade Farshad, Ioannis Charisiadis, Marta Hasny, Martin Hartenberger, Bj\"orn Ommer, Nassir Navab, Ehsan Adeli
Abstract: Scaling by training on large datasets has been shown to enhance the quality and fidelity of image generation and manipulation with diffusion models; however, such large datasets are not always accessible in medical imaging due to cost and privacy issues, which contradicts one of the main applications of such models to produce synthetic samples where real data is scarce. Also, fine-tuning pre-trained general models has been a challenge due to the distribution shift between the medical domain and the pre-trained models. Here, we propose Latent Drift (LD) for diffusion models that can be adopted for any fine-tuning method to mitigate the issues faced by the distribution shift or employed in inference time as a condition. Latent Drifting enables diffusion models to be conditioned for medical images fitted for the complex task of counterfactual image generation, which is crucial to investigate how parameters such as gender, age, and adding or removing diseases in a patient would alter the medical images. We evaluate our method on three public longitudinal benchmark datasets of brain MRI and chest X-rays for counterfactual image generation. Our results demonstrate significant performance gains in various scenarios when combined with different fine-tuning schemes.
Authors: Yi-Ping Chen, Vispi Karkaria, Ying-Kuan Tsai, Faith Rolark, Daniel Quispe, Robert X. Gao, Jian Cao, Wei Chen
Abstract: Digital Twin -- a virtual replica of a physical system enabling real-time monitoring, model updating, prediction, and decision-making -- combined with recent advances in machine learning, offers new opportunities for proactive control strategies in autonomous manufacturing. However, achieving real-time decision-making with Digital Twins requires efficient optimization driven by accurate predictions of highly nonlinear manufacturing systems. This paper presents a simultaneous multi-step Model Predictive Control (MPC) framework for real-time decision-making, using a multivariate deep neural network, named Time-Series Dense Encoder (TiDE), as the surrogate model. Unlike conventional MPC models which only provide one-step ahead prediction, TiDE is capable of predicting future states within the prediction horizon in one shot (multi-step), significantly accelerating the MPC. Using Directed Energy Deposition (DED) additive manufacturing as a case study, we demonstrate the effectiveness of the proposed MPC in achieving melt pool temperature tracking to ensure part quality, while reducing porosity defects by regulating laser power to maintain melt pool depth constraints. In this work, we first show that TiDE is capable of accurately predicting melt pool temperature and depth. Second, we demonstrate that the proposed MPC achieves precise temperature tracking while satisfying melt pool depth constraints within a targeted dilution range (10\%-30\%), reducing potential porosity defects. Compared to PID controller, the MPC results in smoother and less fluctuating laser power profiles with competitive or superior melt pool temperature control performance. This demonstrates the MPC's proactive control capabilities, leveraging time-series prediction and real-time optimization, positioning it as a powerful tool for future Digital Twin applications and real-time process optimization in manufacturing.
Authors: Eneko Osaba, Esther Villar-Rodriguez
Abstract: Being immersed in the NISQ-era, current quantum annealers present limitations for solving optimization problems efficiently. To mitigate these limitations, D-Wave Systems developed a mechanism called Reverse Annealing, a specific type of quantum annealing designed to perform local refinement of good states found elsewhere. Despite the research activity around Reverse Annealing, none has theorized about the possible benefits related to the transfer of knowledge under this paradigm. This work moves in that direction and is driven by experimentation focused on answering two key research questions: i) is reverse annealing a paradigm that can benefit from knowledge transfer between similar problems? and ii) can we infer the characteristics that an input solution should meet to help increase the probability of success? To properly guide the tests in this paper, the well-known Knapsack Problem has been chosen for benchmarking purposes, using a total of 34 instances composed of 14 and 16 items.
Authors: Ahmed K. Kadhim, Lei Jiao, Rishad Shafik, Ole-Christoffer Granmo
Abstract: In recent years, text generation tools utilizing Artificial Intelligence (AI) have occasionally been misused across various domains, such as generating student reports or creative writings. This issue prompts plagiarism detection services to enhance their capabilities in identifying AI-generated content. Adversarial attacks are often used to test the robustness of AI-text generated detectors. This work proposes a novel textual adversarial attack on the detection models such as Fast-DetectGPT. The method employs embedding models for data perturbation, aiming at reconstructing the AI generated texts to reduce the likelihood of detection of the true origin of the texts. Specifically, we employ different embedding techniques, including the Tsetlin Machine (TM), an interpretable approach in machine learning for this purpose. By combining synonyms and embedding similarity vectors, we demonstrates the state-of-the-art reduction in detection scores against Fast-DetectGPT. Particularly, in the XSum dataset, the detection score decreased from 0.4431 to 0.2744 AUROC, and in the SQuAD dataset, it dropped from 0.5068 to 0.3532 AUROC.
Authors: Jaeyong Kang, Dorien Herremans
Abstract: One of the most significant challenges in Music Emotion Recognition (MER) comes from the fact that emotion labels can be heterogeneous across datasets with regard to the emotion representation, including categorical (e.g., happy, sad) versus dimensional labels (e.g., valence-arousal). In this paper, we present a unified multitask learning framework that combines these two types of labels and is thus able to be trained on multiple datasets. This framework uses an effective input representation that combines musical features (i.e., key and chords) and MERT embeddings. Moreover, knowledge distillation is employed to transfer the knowledge of teacher models trained on individual datasets to a student model, enhancing its ability to generalize across multiple tasks. To validate our proposed framework, we conducted extensive experiments on a variety of datasets, including MTG-Jamendo, DEAM, PMEmo, and EmoMusic. According to our experimental results, the inclusion of musical features, multitask learning, and knowledge distillation significantly enhances performance. In particular, our model outperforms the state-of-the-art models, including the best-performing model from the MediaEval 2021 competition on the MTG-Jamendo dataset. Our work makes a significant contribution to MER by allowing the combination of categorical and dimensional emotion labels in one unified framework, thus enabling training across datasets.
Authors: Giampaolo Bovenzi, Francesco Cerasuolo, Domenico Ciuonzo, Davide Di Monda, Idio Guarino, Antonio Montieri, Valerio Persico, Antonio Pescap\`e
Abstract: Generative Artificial Intelligence (GenAI) models such as LLMs, GPTs, and Diffusion Models have recently gained widespread attention from both the research and the industrial communities. This survey explores their application in network monitoring and management, focusing on prominent use cases, as well as challenges and opportunities. We discuss how network traffic generation and classification, network intrusion detection, networked system log analysis, and network digital assistance can benefit from the use of GenAI models. Additionally, we provide an overview of the available GenAI models, datasets for large-scale training phases, and platforms for the development of such models. Finally, we discuss research directions that potentially mitigate the roadblocks to the adoption of GenAI for network monitoring and management. Our investigation aims to map the current landscape and pave the way for future research in leveraging GenAI for network monitoring and management.
Authors: Irene Solaiman, Rishi Bommasani, Dan Hendrycks, Ariel Herbert-Voss, Yacine Jernite, Aviya Skowron, Andrew Trask
Abstract: Generative AI release decisions determine whether system components are made available, but release does not address many other elements that change how users and stakeholders are able to engage with a system. Beyond release, access to system components informs potential risks and benefits. Access refers to practical needs, infrastructurally, technically, and societally, in order to use available components in some way. We deconstruct access along three axes: resourcing, technical usability, and utility. Within each category, a set of variables per system component clarify tradeoffs. For example, resourcing requires access to computing infrastructure to serve model weights. We also compare the accessibility of four high performance language models, two open-weight and two closed-weight, showing similar considerations for all based instead on access variables. Access variables set the foundation for being able to scale or increase access to users; we examine the scale of access and how scale affects ability to manage and intervene on risks. This framework better encompasses the landscape and risk-benefit tradeoffs of system releases to inform system release decisions, research, and policy.
Authors: Hyeonjeong Ha, Xiaomeng Jin, Jeonghwan Kim, Jiateng Liu, Zhenhailong Wang, Khanh Duy Nguyen, Ansel Blume, Nanyun Peng, Kai-Wei Chang, Heng Ji
Abstract: Text-to-image (T2I) models enable rapid concept design, making them widely used in AI-driven design. While recent studies focus on generating semantic and stylistic variations of given design concepts, functional coherence--the integration of multiple affordances into a single coherent concept--remains largely overlooked. In this paper, we introduce SYNTHIA, a framework for generating novel, functionally coherent designs based on desired affordances. Our approach leverages a hierarchical concept ontology that decomposes concepts into parts and affordances, serving as a crucial building block for functionally coherent design. We also develop a curriculum learning scheme based on our ontology that contrastively fine-tunes T2I models to progressively learn affordance composition while maintaining visual novelty. To elaborate, we (i) gradually increase affordance distance, guiding models from basic concept-affordance association to complex affordance compositions that integrate parts of distinct affordances into a single, coherent form, and (ii) enforce visual novelty by employing contrastive objectives to push learned representations away from existing concepts. Experimental results show that SYNTHIA outperforms state-of-the-art T2I models, demonstrating absolute gains of 25.1% and 14.7% for novelty and functional coherence in human evaluation, respectively.
Authors: Jungsoo Park, Junmo Kang, Gabriel Stanovsky, Alan Ritter
Abstract: The surge of LLM studies makes synthesizing their findings challenging. Analysis of experimental results from literature can uncover important trends across studies, but the time-consuming nature of manual data extraction limits its use. Our study presents a semi-automated approach for literature analysis that accelerates data extraction using LLMs. It automatically identifies relevant arXiv papers, extracts experimental results and related attributes, and organizes them into a structured dataset, LLMEvalDB. We then conduct an automated literature analysis of frontier LLMs, reducing the effort of paper surveying and data extraction by more than 93% compared to manual approaches. We validate LLMEvalDB by showing that it reproduces key findings from a recent manual analysis of Chain-of-Thought (CoT) reasoning and also uncovers new insights that go beyond it, showing, for example, that in-context examples benefit coding and multimodal tasks but offer limited gains in math reasoning tasks compared to zero-shot CoT. Our automatically updatable dataset enables continuous tracking of target models by extracting evaluation studies as new data becomes available. Through LLMEvalDB and empirical analysis, we provide insights into LLMs while facilitating ongoing literature analyses of their behavior.
Authors: Zhengxuan Zhang, Yin Wu, Yuyu Luo, Nan Tang
Abstract: Visual Question Answering (VQA) focuses on providing answers to natural language questions by utilizing information from images. Although cutting-edge multimodal large language models (MLLMs) such as GPT-4o achieve strong performance on VQA tasks, they frequently fall short in accessing domain-specific or the latest knowledge. To mitigate this issue, retrieval-augmented generation (RAG) leveraging external knowledge bases (KBs), referred to as KB-VQA, emerges as a promising approach. Nevertheless, conventional unimodal retrieval techniques, which translate images into textual descriptions, often result in the loss of critical visual details. This study presents fine-grained knowledge units, which merge textual snippets with entity images stored in vector databases. Furthermore, we introduce a knowledge unit retrieval-augmented generation framework (KU-RAG) that integrates fine-grained retrieval with MLLMs. The proposed KU-RAG framework ensures precise retrieval of relevant knowledge and enhances reasoning capabilities through a knowledge correction chain. Experimental findings demonstrate that our approach significantly boosts the performance of leading KB-VQA methods, achieving an average improvement of approximately 3% and up to 11% in the best case.
Authors: Santosh Bhupathi
Abstract: Generative AI (GenAI) is transforming industries by enabling intelligent content generation, automation, and decision-making. However, the effectiveness of GenAI applications depends significantly on efficient data storage, retrieval, and contextual augmentation. This paper explores the critical role of databases in GenAI workflows, emphasizing the importance of choosing the right database architecture to optimize performance, accuracy, and scalability. It categorizes database roles into conversational context (key-value/document databases), situational context (relational databases/data lakehouses), and semantic context (vector databases) each serving a distinct function in enriching AI-generated responses. Additionally, the paper highlights real-time query processing, vector search for semantic retrieval, and the impact of database selection on model efficiency and scalability. By leveraging a multi-database approach, GenAI applications can achieve more context-aware, personalized, and high-performing AI-driven solutions.
Authors: Navdeep Kaur, Lachlan McPheat, Alessandra Russo, Anthony G Cohn, Pranava Madhyastha
Abstract: In this paper, we examine the use of Conformal Language Modelling (CLM) alongside Answer Set Programming (ASP) to enhance the performance of standard open-weight LLMs on complex multi-step reasoning tasks. Using the StepGame dataset, which requires spatial reasoning, we apply CLM to generate sets of ASP programs from an LLM, providing statistical guarantees on the correctness of the outputs. Experimental results show that CLM significantly outperforms baseline models that use standard sampling methods, achieving substantial accuracy improvements across different levels of reasoning complexity. Additionally, the LLM-as-Judge metric enhances CLM's performance, especially in assessing structurally and logically correct ASP outputs. However, calibrating CLM with diverse calibration sets did not improve generalizability for tasks requiring much longer reasoning steps, indicating limitations in handling more complex tasks.
Authors: Siyuan Mu, Sen Lin
Abstract: Artificial intelligence (AI) has achieved astonishing successes in many domains, especially with the recent breakthroughs in the development of foundational large models. These large models, leveraging their extensive training data, provide versatile solutions for a wide range of downstream tasks. However, as modern datasets become increasingly diverse and complex, the development of large AI models faces two major challenges: (1) the enormous consumption of computational resources and deployment difficulties, and (2) the difficulty in fitting heterogeneous and complex data, which limits the usability of the models. Mixture of Experts (MoE) models has recently attracted much attention in addressing these challenges, by dynamically selecting and activating the most relevant sub-models to process input data. It has been shown that MoEs can significantly improve model performance and efficiency with fewer resources, particularly excelling in handling large-scale, multimodal data. Given the tremendous potential MoE has demonstrated across various domains, it is urgent to provide a comprehensive summary of recent advancements of MoEs in many important fields. Existing surveys on MoE have their limitations, e.g., being outdated or lacking discussion on certain key areas, and we aim to address these gaps. In this paper, we first introduce the basic design of MoE, including gating functions, expert networks, routing mechanisms, training strategies, and system design. We then explore the algorithm design of MoE in important machine learning paradigms such as continual learning, meta-learning, multi-task learning, and reinforcement learning. Additionally, we summarize theoretical studies aimed at understanding MoE and review its applications in computer vision and natural language processing. Finally, we discuss promising future research directions.
Authors: Chen Gong, Kecen Li, Zinan Lin, Tianhao Wang
Abstract: Differentially private (DP) image synthesis aims to generate artificial images that retain the properties of sensitive images while protecting the privacy of individual images within the dataset. Despite recent advancements, we find that inconsistent--and sometimes flawed--evaluation protocols have been applied across studies. This not only impedes the understanding of current methods but also hinders future advancements. To address the issue, this paper introduces DPImageBench for DP image synthesis, with thoughtful design across several dimensions: (1) Methods. We study eleven prominent methods and systematically characterize each based on model architecture, pretraining strategy, and privacy mechanism. (2) Evaluation. We include nine datasets and seven fidelity and utility metrics to thoroughly assess them. Notably, we find that a common practice of selecting downstream classifiers based on the highest accuracy on the sensitive test set not only violates DP but also overestimates the utility scores. DPImageBench corrects for these mistakes. (3) Platform. Despite the methods and evaluation protocols, DPImageBench provides a standardized interface that accommodates current and future implementations within a unified framework. With DPImageBench, we have several noteworthy findings. For example, contrary to the common wisdom that pretraining on public image datasets is usually beneficial, we find that the distributional similarity between pretraining and sensitive images significantly impacts the performance of the synthetic images and does not always yield improvements. In addition, adding noise to low-dimensional features, such as the high-level characteristics of sensitive images, is less affected by the privacy budget compared to adding noise to high-dimensional features, like weight gradients. The former methods perform better than the latter under a low privacy budget.
Authors: Sean Koon
Abstract: Amidst the race to create more intelligent machines there is a risk that we will rely on AI in ways that reduce our own agency as humans. To reduce this risk, we could aim to create tools that prioritize and enhance the human role in human-AI interactions. This paper outlines a human-centered augmented reasoning paradigm by 1. Articulating fundamental principles for augmented reasoning tools, emphasizing their ergonomic, pre-conclusive, directable, exploratory, enhancing, and integrated nature; 2. Proposing a 'many tasks, many tools' approach to ensuring human influence and control, and 3. Offering examples of interaction modes that can serve as bridges between human reasoning and AI algorithms.
Authors: Sixu Li, Ben Keller, Yingyan Celine Lin, Brucek Khailany
Abstract: 3D intelligence leverages rich 3D features and stands as a promising frontier in AI, with 3D rendering fundamental to many downstream applications. 3D Gaussian Splatting (3DGS), an emerging high-quality 3D rendering method, requires significant computation, making real-time execution on existing GPU-equipped edge devices infeasible. Previous efforts to accelerate 3DGS rely on dedicated accelerators that require substantial integration overhead and hardware costs. This work proposes an acceleration strategy that leverages the similarities between the 3DGS pipeline and the highly optimized conventional graphics pipeline in modern GPUs. Instead of developing a dedicated accelerator, we enhance existing GPU rasterizer hardware to efficiently support 3DGS operations. Our results demonstrate a 23$\times$ increase in processing speed and a 24$\times$ reduction in energy consumption, with improvements yielding 6$\times$ faster end-to-end runtime for the original 3DGS algorithm and 4$\times$ for the latest efficiency-improved pipeline, achieving 24 FPS and 46 FPS respectively. These enhancements incur only a minimal area overhead of 0.2\% relative to the entire SoC chip area, underscoring the practicality and efficiency of our approach for enabling 3DGS rendering on resource-constrained platforms.
Authors: Yuxuan Zhu, Antony Kellermann, Dylan Bowman, Philip Li, Akul Gupta, Adarsh Danda, Richard Fang, Conner Jensen, Eric Ihli, Jason Benn, Jet Geronimo, Avi Dhir, Sudhit Rao, Kaicheng Yu, Twm Stone, Daniel Kang
Abstract: Large language model (LLM) agents are increasingly capable of autonomously conducting cyberattacks, posing significant threats to existing applications. This growing risk highlights the urgent need for a real-world benchmark to evaluate the ability of LLM agents to exploit web application vulnerabilities. However, existing benchmarks fall short as they are limited to abstracted Capture the Flag competitions or lack comprehensive coverage. Building a benchmark for real-world vulnerabilities involves both specialized expertise to reproduce exploits and a systematic approach to evaluating unpredictable threats. To address this challenge, we introduce CVE-Bench, a real-world cybersecurity benchmark based on critical-severity Common Vulnerabilities and Exposures. In CVE-Bench, we design a sandbox framework that enables LLM agents to exploit vulnerable web applications in scenarios that mimic real-world conditions, while also providing effective evaluation of their exploits. Our evaluation shows that the state-of-the-art agent framework can resolve up to 13% of vulnerabilities.
Authors: Antonio-Gabriel Chac\'on Menke (Shibaura Institute of Technology, Kempten University of Applied Sciences), Phan Xuan Tan (Shibaura Institute of Technology)
Abstract: Recent incidents highlight safety risks in Large Language Models (LLMs), motivating research into alignment methods like Constitutional AI (CAI). This paper explores CAI's self-critique mechanism on small, uncensored 7-9B parameter models: DeepSeek-R1-8B, Gemma-2-9B, Llama 3.1-8B, and Qwen2.5-7B. We show that while Llama-based models exhibited significant harm reduction through self-critique, other architectures demonstrated less improvement in harm detection after abliteration. These results suggest CAI's effectiveness may vary depending on model architecture and reasoning capabilities.
Authors: Alejandro Ortega
Abstract: Recent progress in AI capabilities has heightened concerns that AI systems could pose a threat to national security, for example, by making it easier for malicious actors to perform cyberattacks on critical national infrastructure, or through loss of control of autonomous AI systems. In parallel, federal legislators in the US have proposed nascent 'AI incident regimes' to identify and counter similar threats. In this paper, we consolidate these two trends and present a timely proposal for a legally mandated post-deployment AI incident regime that aims to counter potential national security threats from AI systems. We start the paper by introducing the concept of 'security-critical' to describe doctors that pose extreme risks to national security, before arguing that 'security-critical' describes civilian nuclear power, aviation, life science dual-use research of concern, and frontier AI development. We then present in detail our AI incident regime proposal, justifying each component of the proposal by demonstrating its similarity to US domestic incident regimes in other 'security-critical' sectors. Finally, we sketch a hypothetical scenario where our proposed AI incident regime deals with an AI cyber incident. Our proposed AI incident regime is split into three phases. The first phase revolves around a novel operationalization of what counts as an 'AI incident' and we suggest that AI providers must create a 'national security case' before deploying a frontier AI system. The second and third phases spell out that AI providers should notify a government agency about incidents, and that the government agency should be involved in amending AI providers' security and safety procedures, in order to counter future threats to national security.
Authors: Supriyo Maji, Linran Zhao, Souradip Poddar, David Z. Pan
Abstract: Layout-dependent effects (LDEs) significantly impact analog circuit performance. Traditionally, designers have relied on symmetric placement of circuit components to mitigate variations caused by LDEs. However, due to non-linear nature of these effects, conventional methods often fall short. We propose an objective-driven, multi-level, multi-agent Q-learning framework to explore unconventional design space of analog layout, opening new avenues for optimizing analog circuit performance. Our approach achieves better variation performance than the state-of-the-art layout techniques. Notably, this is the first application of multi-agent RL in analog layout automation. The proposed approach is compared with non-ML approach based on simulated annealing.
Authors: Mahjabin Nahar, Eun-Ju Lee, Jin Won Park, Dongwon Lee
Abstract: While we increasingly rely on large language models (LLMs) for various tasks, these models are known to produce inaccurate content or 'hallucinations' with potentially disastrous consequences. The recent integration of web search results into LLMs prompts the question of whether people utilize them to verify the generated content, thereby avoiding falling victim to hallucinations. This study (N = 560) investigated how the provision of search results, either static (fixed search results) or dynamic (participant-driven searches), affect participants' perceived accuracy and confidence in evaluating LLM-generated content (i.e., genuine, minor hallucination, major hallucination), compared to the control condition (no search results). Findings indicate that participants in both static and dynamic conditions (vs. control) rated hallucinated content to be less accurate. However, those in the dynamic condition rated genuine content as more accurate and demonstrated greater overall confidence in their assessments than those in the static or control conditions. In addition, those higher in need for cognition (NFC) rated major hallucinations to be less accurate than low NFC participants, with no corresponding difference for genuine content or minor hallucinations. These results underscore the potential benefits of integrating web search results into LLMs for the detection of hallucinations, as well as the need for a more nuanced approach when developing human-centered systems, taking user characteristics into account.
Authors: Brandon Radosevich, John Halloran
Abstract: To reduce development overhead and enable seamless integration between potential components comprising any given generative AI application, the Model Context Protocol (MCP) (Anthropic, 2024) has recently been released and subsequently widely adopted. The MCP is an open protocol that standardizes API calls to large language models (LLMs), data sources, and agentic tools. By connecting multiple MCP servers, each defined with a set of tools, resources, and prompts, users are able to define automated workflows fully driven by LLMs. However, we show that the current MCP design carries a wide range of security risks for end users. In particular, we demonstrate that industry-leading LLMs may be coerced into using MCP tools to compromise an AI developer's system through various attacks, such as malicious code execution, remote access control, and credential theft. To proactively mitigate these and related attacks, we introduce a safety auditing tool, MCPSafetyScanner, the first agentic tool to assess the security of an arbitrary MCP server. MCPScanner uses several agents to (a) automatically determine adversarial samples given an MCP server's tools and resources; (b) search for related vulnerabilities and remediations based on those samples; and (c) generate a security report detailing all findings. Our work highlights serious security issues with general-purpose agentic workflows while also providing a proactive tool to audit MCP server safety and address detected vulnerabilities before deployment. The described MCP server auditing tool, MCPSafetyScanner, is freely available at: https://github.com/johnhalloran321/mcpSafetyScanner
Authors: Dylan Riffle, Nima Shirooni, Cody He, Manush Murali, Sovit Nayak, Rishikumar Gopalan, Diego Gonzalez Lopez
Abstract: OLAF (Open Life Science Analysis Framework) is an open-source platform that enables researchers to perform bioinformatics analyses using natural language. By combining large language models (LLMs) with a modular agent-pipe-router architecture, OLAF generates and executes bioinformatics code on real scientific data, including formats like .h5ad. The system includes an Angular front end and a Python/Firebase backend, allowing users to run analyses such as single-cell RNA-seq workflows, gene annotation, and data visualization through a simple web interface. Unlike general-purpose AI tools, OLAF integrates code execution, data handling, and scientific libraries in a reproducible, user-friendly environment. It is designed to lower the barrier to computational biology for non-programmers and support transparent, AI-powered life science research.
Authors: Justus Westerhoff, Erblina Purelku, Jakob Hackstein, Leo Pinetzki, Lorenz Hufe
Abstract: Typographic attacks exploit the interplay between text and visual content in multimodal foundation models, causing misclassifications when misleading text is embedded within images. However, existing datasets are limited in size and diversity, making it difficult to study such vulnerabilities. In this paper, we introduce SCAM, the largest and most diverse dataset of real-world typographic attack images to date, containing 1,162 images across hundreds of object categories and attack words. Through extensive benchmarking of Vision-Language Models (VLMs) on SCAM, we demonstrate that typographic attacks significantly degrade performance, and identify that training data and model architecture influence the susceptibility to these attacks. Our findings reveal that typographic attacks persist in state-of-the-art Large Vision-Language Models (LVLMs) due to the choice of their vision encoder, though larger Large Language Models (LLMs) backbones help mitigate their vulnerability. Additionally, we demonstrate that synthetic attacks closely resemble real-world (handwritten) attacks, validating their use in research. Our work provides a comprehensive resource and empirical insights to facilitate future research toward robust and trustworthy multimodal AI systems. We publicly release the datasets introduced in this paper under https://huggingface.co/datasets/BLISS-e-V/SCAM, along with the code for evaluations at https://github.com/Bliss-e-V/SCAM.
URLs: https://huggingface.co/datasets/BLISS-e-V/SCAM,, https://github.com/Bliss-e-V/SCAM.
Authors: Leonardo Kanashiro Felizardo, Edoardo Fadda, Paolo Brandimarte, Emilio Del-Moral-Hernandez, Mari\'a Cristina Vasconcelos Nascimento
Abstract: This paper presents Post-Decision Proximal Policy Optimization (PDPPO), a novel variation of the leading deep reinforcement learning method, Proximal Policy Optimization (PPO). The PDPPO state transition process is divided into two steps: a deterministic step resulting in the post-decision state and a stochastic step leading to the next state. Our approach incorporates post-decision states and dual critics to reduce the problem's dimensionality and enhance the accuracy of value function estimation. Lot-sizing is a mixed integer programming problem for which we exemplify such dynamics. The objective of lot-sizing is to optimize production, delivery fulfillment, and inventory levels in uncertain demand and cost parameters. This paper evaluates the performance of PDPPO across various environments and configurations. Notably, PDPPO with a dual critic architecture achieves nearly double the maximum reward of vanilla PPO in specific scenarios, requiring fewer episode iterations and demonstrating faster and more consistent learning across different initializations. On average, PDPPO outperforms PPO in environments with a stochastic component in the state transition. These results support the benefits of using a post-decision state. Integrating this post-decision state in the value function approximation leads to more informed and efficient learning in high-dimensional and stochastic environments.
Authors: Hao Nan Sheng, Zhi-yong Wang, Mingrui Yang, Hing Cheung So
Abstract: As large language models continue to grow in size, parameter-efficient fine-tuning (PEFT) has become increasingly crucial. While low-rank adaptation (LoRA) offers a solution through low-rank updates, its static rank allocation may yield suboptimal results. Adaptive low-rank adaptation (AdaLoRA) improves this with dynamic allocation but remains sensitive to initial and target rank configurations. We introduce AROMA, a framework that automatically constructs layer-specific updates by iteratively building up rank-one components with very few trainable parameters that gradually diminish to zero. Unlike existing methods that employ rank reduction mechanisms, AROMA introduces a dual-loop architecture for rank growth. The inner loop extracts information from each rank-one subspace, while the outer loop determines the number of rank-one subspaces, i.e., the optimal rank. We reset optimizer states to maintain subspace independence. AROMA significantly reduces parameters compared to LoRA and AdaLoRA while achieving superior performance on natural language understanding and commonsense reasoning tasks, offering new insights into adaptive PEFT. The code is available at \href{https://github.com/ShuDun23/AROMA}{AROMA}.
Authors: Ian Groves, Andrew Campbell, James Fernandes, Diego Ram\'irez Rodr\'iguez, Paul Murray, Massimiliano Vasile, Victoria Nockles
Abstract: Foundation Models, pre-trained on large unlabelled datasets before task-specific fine-tuning, are increasingly being applied to specialised domains. Recent examples include ClimaX for climate and Clay for satellite Earth observation, but a Foundation Model for Space Object Behavioural Analysis has not yet been developed. As orbital populations grow, automated methods for characterising space object behaviour are crucial for space safety. We present a Space Safety and Sustainability Foundation Model focusing on space object behavioural analysis using light curves (LCs). We implemented a Perceiver-Variational Autoencoder (VAE) architecture, pre-trained with self-supervised reconstruction and masked reconstruction on 227,000 LCs from the MMT-9 observatory. The VAE enables anomaly detection, motion prediction, and LC generation. We fine-tuned the model for anomaly detection & motion prediction using two independent LC simulators (CASSANDRA and GRIAL respectively), using CAD models of boxwing, Sentinel-3, SMOS, and Starlink platforms. Our pre-trained model achieved a reconstruction error of 0.01%, identifying potentially anomalous light curves through reconstruction difficulty. After fine-tuning, the model scored 88% and 82% accuracy, with 0.90 and 0.95 ROC AUC scores respectively in both anomaly detection and motion mode prediction (sun-pointing, spin, etc.). Analysis of high-confidence anomaly predictions on real data revealed distinct patterns including characteristic object profiles and satellite glinting. Here, we demonstrate how self-supervised learning can simultaneously enable anomaly detection, motion prediction, and synthetic data generation from rich representations learned in pre-training. Our work therefore supports space safety and sustainability through automated monitoring and simulation capabilities.
Authors: Alexander Rubinstein, Ameya Prabhu, Matthias Bethge, Seong Joon Oh
Abstract: Object-centric learning (OCL) seeks to learn representations that only encode an object, isolated from other objects or background cues in a scene. This approach underpins various aims, including out-of-distribution (OOD) generalization, sample-efficient composition, and modeling of structured environments. Most research has focused on developing unsupervised mechanisms that separate objects into discrete slots in the representation space, evaluated using unsupervised object discovery. However, with recent sample-efficient segmentation models, we can separate objects in the pixel space and encode them independently. This achieves remarkable zero-shot performance on OOD object discovery benchmarks, is scalable to foundation models, and can handle a variable number of slots out-of-the-box. Hence, the goal of OCL methods to obtain object-centric representations has been largely achieved. Despite this progress, a key question remains: How does the ability to separate objects within a scene contribute to broader OCL objectives, such as OOD generalization? We address this by investigating the OOD generalization challenge caused by spurious background cues through the lens of OCL. We propose a novel, training-free probe called Object-Centric Classification with Applied Masks (OCCAM), demonstrating that segmentation-based encoding of individual objects significantly outperforms slot-based OCL methods. However, challenges in real-world applications remain. We provide the toolbox for the OCL community to use scalable object-centric representations, and focus on practical applications and fundamental questions, such as understanding object perception in human cognition. Our code is available here: https://github.com/AlexanderRubinstein/OCCAM.
Authors: Jennifer D'Souza, Sameer Sadruddin, Holger Israel, Mathias Begoin, Diana Slawig
Abstract: We present SemEval-2025 Task 5: LLMs4Subjects, a shared task on automated subject tagging for scientific and technical records in English and German using the GND taxonomy. Participants developed LLM-based systems to recommend top-k subjects, evaluated through quantitative metrics (precision, recall, F1-score) and qualitative assessments by subject specialists. Results highlight the effectiveness of LLM ensembles, synthetic data generation, and multilingual processing, offering insights into applying LLMs for digital library classification.
Authors: Axelle Apvrille, Daniel Nakov
Abstract: This research studies the quality, speed and cost of malware analysis assisted by artificial intelligence. It focuses on Linux and IoT malware of 2024-2025, and uses r2ai, the AI extension of Radare2's disassembler. Not all malware and not all LLMs are equivalent but the study shows excellent results with Claude 3.5 and 3.7 Sonnet. Despite a few errors, the quality of analysis is overall equal or better than without AI assistance. For good results, the AI cannot operate alone and must constantly be guided by an experienced analyst. The gain of speed is largely visible with AI assistance, even when taking account the time to understand AI's hallucinations, exaggerations and omissions. The cost is usually noticeably lower than the salary of a malware analyst, but attention and guidance is needed to keep it under control in cases where the AI would naturally loop without showing progress.
Authors: Junli Liu, Qizhi Chen, Zhigang Wang, Yiwen Tang, Yiting Zhang, Chi Yan, Dong Wang, Xuelong Li, Bin Zhao
Abstract: Visual grounding (VG) aims to localize target objects in an image based on natural language descriptions. In this paper, we propose AerialVG, a new task focusing on visual grounding from aerial views. Compared to traditional VG, AerialVG poses new challenges, \emph{e.g.}, appearance-based grounding is insufficient to distinguish among multiple visually similar objects, and positional relations should be emphasized. Besides, existing VG models struggle when applied to aerial imagery, where high-resolution images cause significant difficulties. To address these challenges, we introduce the first AerialVG dataset, consisting of 5K real-world aerial images, 50K manually annotated descriptions, and 103K objects. Particularly, each annotation in AerialVG dataset contains multiple target objects annotated with relative spatial relations, requiring models to perform comprehensive spatial reasoning. Furthermore, we propose an innovative model especially for the AerialVG task, where a Hierarchical Cross-Attention is devised to focus on target regions, and a Relation-Aware Grounding module is designed to infer positional relations. Experimental results validate the effectiveness of our dataset and method, highlighting the importance of spatial reasoning in aerial visual grounding. The code and dataset will be released.
Authors: Yichun Yin, Wenyong Huang, Kaikai Song, Yehui Tang, Xueyu Wu, Wei Guo, Peng Guo, Yaoyuan Wang, Xiaojun Meng, Yasheng Wang, Dong Li, Can Chen, Dandan Tu, Yin Li, Fisher Yu, Ruiming Tang, Yunhe Wang, Baojun Wang, Bin Wang, Bo Wang, Boxiao Liu, Changzheng Zhang, Duyu Tang, Fei Mi, Hui Jin, Jiansheng Wei, Jiarui Qin, Jinpeng Li, Jun Zhao, Liqun Deng, Lin Li, Minghui Xu, Naifu Zhang, Nianzu Zheng, Qiang Li, Rongju Ruan, Shengjun Cheng, Tianyu Guo, Wei He, Wei Li, Weiwen Liu, Wulong Liu, Xinyi Dai, Yonghan Dong, Yu Pan, Yue Li, Yufei Wang, Yujun Li, Yunsheng Ni, Zhe Liu, Zhenhe Zhang, Zhicheng Liu
Abstract: We present Pangu Ultra, a Large Language Model (LLM) with 135 billion parameters and dense Transformer modules trained on Ascend Neural Processing Units (NPUs). Although the field of LLM has been witnessing unprecedented advances in pushing the scale and capability of LLM in recent years, training such a large-scale model still involves significant optimization and system challenges. To stabilize the training process, we propose depth-scaled sandwich normalization, which effectively eliminates loss spikes during the training process of deep models. We pre-train our model on 13.2 trillion diverse and high-quality tokens and further enhance its reasoning capabilities during post-training. To perform such large-scale training efficiently, we utilize 8,192 Ascend NPUs with a series of system optimizations. Evaluations on multiple diverse benchmarks indicate that Pangu Ultra significantly advances the state-of-the-art capabilities of dense LLMs such as Llama 405B and Mistral Large 2, and even achieves competitive results with DeepSeek-R1, whose sparse model structure contains much more parameters. Our exploration demonstrates that Ascend NPUs are capable of efficiently and effectively training dense models with more than 100 billion parameters. Our model and system will be available for our commercial customers.