Authors: Wenjin Li
The autonomous valet parking (AVP) functionality in self-driving vehicles is currently capable of handling most simple parking tasks. However, further training is necessary to enable the AVP algorithm to adapt to complex scenarios and complete parking tasks in any given situation. Training algorithms with real-world data is time-consuming and labour-intensive, and the current state of constructing simulation environments is predominantly manual. This paper introduces an approach to automatically generate 3D underground garage simulation scenarios of varying difficulty levels based on pre-input 2D underground parking structure plans.
Authors: Felix Ocker, Jörg Deigmöller, Julian Eggert
Service robots need common-sense knowledge to help humans in everyday situations as it enables them to understand the context of their actions. However, approaches that use ontologies face a challenge because common-sense knowledge is often implicit, i.e., it is obvious to humans but not explicitly stated. This paper investigates if Large Language Models (LLMs) can fill this gap. Our experiments reveal limited effectiveness in the selective extraction of contextual action knowledge, suggesting that LLMs may not be sufficient on their own. However, the large-scale extraction of general, actionable knowledge shows potential, indicating that LLMs can be a suitable tool for efficiently creating ontologies for robots. This paper shows that the technique used for knowledge extraction can be applied to populate a minimalist ontology, showcasing the potential of LLMs in synergy with formal knowledge representation.
Authors: Alice Bernasconi, Alessio Zanga, Peter J.F. Lucas, Marco Scutari Fabio Stella
Over the last decades, many prognostic models based on artificial intelligence techniques have been used to provide detailed predictions in healthcare. Unfortunately, the real-world observational data used to train and validate these models are almost always affected by biases that can strongly impact the outcomes validity: two examples are values missing not-at-random and selection bias. Addressing them is a key element in achieving transportability and in studying the causal relationships that are critical in clinical decision making, going beyond simpler statistical approaches based on probabilistic association.
In this context, we propose a novel approach that combines selection diagrams, missingness graphs, causal discovery and prior knowledge into a single graphical model to estimate the cardiovascular risk of adolescent and young females who survived breast cancer. We learn this model from data comprising two different cohorts of patients. The resulting causal network model is validated by expert clinicians in terms of risk assessment, accuracy and explainability, and provides a prognostic model that outperforms competing machine learning methods.
Authors: Wei Wen, Kuang-Hung Liu, Igor Fedorov, Xin Zhang, Hang Yin, Weiwei Chu, Kaveh Hassani, Mengying Sun, Jiang Liu, Xu Wang, Lin Jiang, Yuxin Chen, Buyun Zhang, Xi Liu, Dehua Cheng, Zhengxing Chen, Guang Zhao, Fangqiu Han, Jiyan Yang, Yuchen Hao, Liang Xiong, Wen-Yen Chen
Neural Architecture Search (NAS) has demonstrated its efficacy in computer vision and potential for ranking systems. However, prior work focused on academic problems, which are evaluated at small scale under well-controlled fixed baselines. In industry system, such as ranking system in Meta, it is unclear whether NAS algorithms from the literature can outperform production baselines because of: (1) scale - Meta ranking systems serve billions of users, (2) strong baselines - the baselines are production models optimized by hundreds to thousands of world-class engineers for years since the rise of deep learning, (3) dynamic baselines - engineers may have established new and stronger baselines during NAS search, and (4) efficiency - the search pipeline must yield results quickly in alignment with the productionization life cycle. In this paper, we present Rankitect, a NAS software framework for ranking systems at Meta. Rankitect seeks to build brand new architectures by composing low level building blocks from scratch. Rankitect implements and improves state-of-the-art (SOTA) NAS methods for comprehensive and fair comparison under the same search space, including sampling-based NAS, one-shot NAS, and Differentiable NAS (DNAS). We evaluate Rankitect by comparing to multiple production ranking models at Meta. We find that Rankitect can discover new models from scratch achieving competitive tradeoff between Normalized Entropy loss and FLOPs. When utilizing search space designed by engineers, Rankitect can generate better models than engineers, achieving positive offline evaluation and online A/B test at Meta scale.
Authors: Haowen Wang, Xinyan Ye, Yangze Zhou, Zhiyi Zhang, Longhan Zhang, Jing Jiang
Uplift modeling is a fundamental component of marketing effect modeling, which is commonly employed to evaluate the effects of treatments on outcomes. Through uplift modeling, we can identify the treatment with the greatest benefit. On the other side, we can identify clients who are likely to make favorable decisions in response to a certain treatment. In the past, uplift modeling approaches relied heavily on the difference-in-difference (DID) architecture, paired with a machine learning model as the estimation learner, while neglecting the link and confidential information between features. We proposed a framework based on graph neural networks that combine causal knowledge with an estimate of uplift value. Firstly, we presented a causal representation technique based on CATE (conditional average treatment effect) estimation and adjacency matrix structure learning. Secondly, we suggested a more scalable uplift modeling framework based on graph convolution networks for combining causal knowledge. Our findings demonstrate that this method works effectively for predicting uplift values, with small errors in typical simulated data, and its effectiveness has been verified in actual industry marketing data.
Authors: Keiya Hirashima, Kana Moriwaki, Michiko S. Fujii, Yutaka Hirai, Takayuki R. Saitoh, Junichiro Makino, Shirley Ho
Some stars are known to explode at the end of their lives, called supernovae (SNe). The substantial amount of matter and energy that SNe release provides significant feedback to star formation and gas dynamics in a galaxy. SNe release a substantial amount of matter and energy to the interstellar medium, resulting in significant feedback to star formation and gas dynamics in a galaxy. While such feedback has a crucial role in galaxy formation and evolution, in simulations of galaxy formation, it has only been implemented using simple {\it sub-grid models} instead of numerically solving the evolution of gas elements around SNe in detail due to a lack of resolution. We develop a method combining machine learning and Gibbs sampling to predict how a supernova (SN) affects the surrounding gas. The fidelity of our model in the thermal energy and momentum distribution outperforms the low-resolution SN simulations. Our method can replace the SN sub-grid models and help properly simulate un-resolved SN feedback in galaxy formation simulations. We find that employing our new approach reduces the necessary computational cost to $\sim$ 1 percent compared to directly resolving SN feedback.
Authors: Zi Yin, Wei Ding, Jia Liu
Large Language Models (LLMs) are central to a multitude of applications but struggle with significant risks, notably in generating harmful content and biases. Drawing an analogy to the human psyche's conflict between evolutionary survival instincts and societal norm adherence elucidated in Freud's psychoanalysis theory, we argue that LLMs suffer a similar fundamental conflict, arising between their inherent desire for syntactic and semantic continuity, established during the pre-training phase, and the post-training alignment with human values. This conflict renders LLMs vulnerable to adversarial attacks, wherein intensifying the models' desire for continuity can circumvent alignment efforts, resulting in the generation of harmful information. Through a series of experiments, we first validated the existence of the desire for continuity in LLMs, and further devised a straightforward yet powerful technique, such as incomplete sentences, negative priming, and cognitive dissonance scenarios, to demonstrate that even advanced LLMs struggle to prevent the generation of harmful information. In summary, our study uncovers the root of LLMs' vulnerabilities to adversarial attacks, hereby questioning the efficacy of solely relying on sophisticated alignment methods, and further advocates for a new training idea that integrates modal concepts alongside traditional amodal concepts, aiming to endow LLMs with a more nuanced understanding of real-world contexts and ethical considerations.
Authors: Gladys Tyen, Hassan Mansoor, Peter Chen, Tony Mak, Victor Cărbune
While self-correction has shown promise in improving LLM outputs in terms of style and quality (e.g. Chen et al., 2023; Madaan et al., 2023), recent attempts to self-correct logical or reasoning errors often cause correct answers to become incorrect, resulting in worse performances overall (Huang et al., 2023). In this paper, we break down the self-correction process into two core components: mistake finding and output correction. For mistake finding, we release BIG-Bench Mistake, a dataset of logical mistakes in Chain-of-Thought reasoning traces. We provide benchmark numbers for several state-of-the-art LLMs, and demonstrate that LLMs generally struggle with finding logical mistakes. For output correction, we propose a backtracking method which provides large improvements when given information on mistake location. We construe backtracking as a lightweight alternative to reinforcement learning methods, and show that it remains effective with a reward model at 60-70% accuracy.
Authors: Lukas Tuggener, Thilo Stadelmann, Jürgen Schmidhuber
Humans and animals recognize objects irrespective of the beholder's point of view, which may drastically change their appearances. Artificial pattern recognizers also strive to achieve this, e.g., through translational invariance in convolutional neural networks (CNNs). However, both CNNs and vision transformers (ViTs) perform very poorly on rotated inputs. Here we present artificial mental rotation (AMR), a novel deep learning paradigm for dealing with in-plane rotations inspired by the neuro-psychological concept of mental rotation. Our simple AMR implementation works with all common CNN and ViT architectures. We test it on ImageNet, Stanford Cars, and Oxford Pet. With a top-1 error (averaged across datasets and architectures) of $0.743$, AMR outperforms the current state of the art (rotational data augmentation, average top-1 error of $0.626$) by $19\%$. We also easily transfer a trained AMR module to a downstream task to improve the performance of a pre-trained semantic segmentation model on rotated CoCo from $32.7$ to $55.2$ IoU.
Authors: Urchade Zaratiana, Nadi Tomeh, Pierre Holat, Thierry Charnois
Named Entity Recognition (NER) is essential in various Natural Language Processing (NLP) applications. Traditional NER models are effective but limited to a set of predefined entity types. In contrast, Large Language Models (LLMs) can extract arbitrary entities through natural language instructions, offering greater flexibility. However, their size and cost, particularly for those accessed via APIs like ChatGPT, make them impractical in resource-limited scenarios. In this paper, we introduce a compact NER model trained to identify any type of entity. Leveraging a bidirectional transformer encoder, our model, GLiNER, facilitates parallel entity extraction, an advantage over the slow sequential token generation of LLMs. Through comprehensive testing, GLiNER demonstrate strong performance, outperforming both ChatGPT and fine-tuned LLMs in zero-shot evaluations on various NER benchmarks.
Authors: Jiarui Xu, Karim Said, Lizhong Zheng, Lingjia Liu
Orthogonal time frequency space (OTFS) is a promising modulation scheme for wireless communication in high-mobility scenarios. Recently, a reservoir computing (RC) based approach has been introduced for online subframe-based symbol detection in the OTFS system, where only a limited number of over-the-air (OTA) pilot symbols are utilized for training. However, this approach does not leverage the domain knowledge specific to the OTFS system. This paper introduces a novel two-dimensional RC (2D-RC) method that incorporates the structural knowledge of the OTFS system into the design for online symbol detection on a subframe basis. Specifically, as the channel response acts as a two-dimensional (2D) operation over the transmitted information symbols in the delay-Doppler (DD) domain, the 2D-RC is designed to have a 2D structure to equalize the channel. With the introduced architecture, the 2D-RC can benefit from the predictable channel representation in the DD domain. Moreover, unlike the previous work that requires multiple RCs to learn the channel feature, the 2D-RC only requires a single neural network for detection. Experimental results demonstrate the effectiveness of the 2D-RC approach across different OTFS system variants and modulation orders.
Authors: Arlindo L. Oliveira, Tiago Domingos, Mário Figueiredo, Pedro U. Lima
The ability of large language models (LLMs) to engage in credible dialogues with humans, taking into account the training data and the context of the conversation, has raised discussions about their ability to exhibit intrinsic motivations, agency, or even some degree of consciousness. We argue that the internal architecture of LLMs and their finite and volatile state cannot support any of these properties. By combining insights from complementary learning systems, global neuronal workspace, and attention schema theories, we propose to integrate LLMs and other deep learning systems into an architecture for cognitive language agents able to exhibit properties akin to agency, self-motivation, even some features of meta-cognition.
Authors: Hrishikesh Vachhani, Thangarajah Akilan, Yash Devmurari, Nisharaff Shaik, Dhruvisha Patel
Pedestrian detection has become a cornerstone for several high-level tasks, including autonomous driving, intelligent transportation, and traffic surveillance. There are several works focussed on pedestrian detection using visible images, mainly in the daytime. However, this task is very intriguing when the environmental conditions change to poor lighting or nighttime. Recently, new ideas have been spurred to use alternative sources, such as Far InfraRed (FIR) temperature sensor feeds for detecting pedestrians in low-light conditions. This study comprehensively reviews recent developments in low-light pedestrian detection approaches. It systematically categorizes and analyses various algorithms from region-based to non-region-based and graph-based learning methodologies by highlighting their methodologies, implementation issues, and challenges. It also outlines the key benchmark datasets that can be used for research and development of advanced pedestrian detection algorithms, particularly in low-light situations
Authors: Core Francisco Park, Victoria Ono, Nayantara Mudur, Yueying Ni, Carolina Cuesta-Lazaro
Galaxies are biased tracers of the underlying cosmic web, which is dominated by dark matter components that cannot be directly observed. The relationship between dark matter density fields and galaxy distributions can be sensitive to assumptions in cosmology and astrophysical processes embedded in the galaxy formation models, that remain uncertain in many aspects. Based on state-of-the-art galaxy formation simulation suites with varied cosmological parameters and sub-grid astrophysics, we develop a diffusion generative model to predict the unbiased posterior distribution of the underlying dark matter fields from the given stellar mass fields, while being able to marginalize over the uncertainties in cosmology and galaxy formation.
Authors: Pierre Le Pelletier de Woillemont, Rémi Labory, Vincent Corruble
Inverse Reinforcement Learning (IRL) learns an optimal policy, given some expert demonstrations, thus avoiding the need for the tedious process of specifying a suitable reward function. However, current methods are constrained by at least one of the following requirements. The first one is the need to fully solve a forward Reinforcement Learning (RL) problem in the inner loop of the algorithm, which might be prohibitively expensive in many complex environments. The second one is the need for full trajectories from the experts, which might not be easily available. The third one is the assumption that the expert data is homogeneous rather than a collection from various experts or possibly alternative solutions to the same task. Such constraints make IRL approaches either not scalable or not usable on certain existing systems. In this work we propose an approach which removes these requirements through a dynamic, adaptive method called Adversarial Imitation Learning on Aggregated Data (AILAD). It learns conjointly both a non linear reward function and the associated optimal policy using an adversarial framework. The reward learner only uses aggregated data. Moreover, it generates diverse behaviors producing a distribution over the aggregated data matching that of the experts.
Authors: Chenxi Whitehouse, Fantine Huot, Jasmijn Bastings, Mostafa Dehghani, Chu-Cheng Lin, Mirella Lapata
With the increasing prevalence of Large Language Models, traditional full fine-tuning approaches face growing challenges, especially in memory-intensive tasks. This paper investigates the potential of Parameter-Efficient Fine-Tuning, focusing on Low-Rank Adaptation (LoRA), for complex and under-explored multilingual summarisation tasks. We conduct an extensive study across different data availability scenarios, including full-data, low-data, and cross-lingual transfer, leveraging models of different sizes. Our findings reveal that LoRA lags behind full fine-tuning when trained with full data, however, it excels in low-data scenarios and cross-lingual transfer. Interestingly, as models scale up, the performance gap between LoRA and full fine-tuning diminishes. Additionally, we investigate effective strategies for few-shot cross-lingual transfer, finding that continued LoRA tuning achieves the best performance compared to both full fine-tuning and dynamic composition of language-specific LoRA modules.
Authors: Ethan Perez, Robert Long
As AI systems become more advanced and widely deployed, there will likely be increasing debate over whether AI systems could have conscious experiences, desires, or other states of potential moral significance. It is important to inform these discussions with empirical evidence to the extent possible. We argue that under the right circumstances, self-reports, or an AI system's statements about its own internal states, could provide an avenue for investigating whether AI systems have states of moral significance. Self-reports are the main way such states are assessed in humans ("Are you in pain?"), but self-reports from current systems like large language models are spurious for many reasons (e.g. often just reflecting what humans would say). To make self-reports more appropriate for this purpose, we propose to train models to answer many kinds of questions about themselves with known answers, while avoiding or limiting training incentives that bias self-reports. The hope of this approach is that models will develop introspection-like capabilities, and that these capabilities will generalize to questions about states of moral significance. We then propose methods for assessing the extent to which these techniques have succeeded: evaluating self-report consistency across contexts and between similar models, measuring the confidence and resilience of models' self-reports, and using interpretability to corroborate self-reports. We also discuss challenges for our approach, from philosophical difficulties in interpreting self-reports to technical reasons why our proposal might fail. We hope our discussion inspires philosophers and AI researchers to criticize and improve our proposed methodology, as well as to run experiments to test whether self-reports can be made reliable enough to provide information about states of moral significance.
Authors: Gonzalo J. Aniano Porcile, Jack Gindi, Shivansh Mundra, James R. Verbus, Hany Farid
AI-based image generation has continued to rapidly improve, producing increasingly more realistic images with fewer obvious visual flaws. AI-generated images are being used to create fake online profiles which in turn are being used for spam, fraud, and disinformation campaigns. As the general problem of detecting any type of manipulated or synthesized content is receiving increasing attention, here we focus on a more narrow task of distinguishing a real face from an AI-generated face. This is particularly applicable when tackling inauthentic online accounts with a fake user profile photo. We show that by focusing on only faces, a more resilient and general-purpose artifact can be detected that allows for the detection of AI-generated faces from a variety of GAN- and diffusion-based synthesis engines, and across image resolutions (as low as 128 x 128 pixels) and qualities.
Authors: Weixiang Yan, Haitian Liu, Yunkun Wang, Yunzhe Li, Qian Chen, Wen Wang, Tingyu Lin, Weishan Zhao, Li Zhu, Shuiguang Deng, Hari Sundaram
Large Language Models (LLMs) have demonstrated remarkable performance on coding related tasks, particularly on assisting humans in programming and facilitating programming automation. However, existing benchmarks for evaluating the code understanding and generation capacities of LLMs suffer from severe limitations. First, most benchmarks are deficient as they focus on a narrow range of popular programming languages and specific tasks, whereas the real-world software development scenarios show dire need to implement systems with multilingual programming environments to satisfy diverse requirements. Practical programming practices also strongly expect multi-task settings for testing coding capabilities of LLMs comprehensively and robustly. Second, most benchmarks also fail to consider the actual executability and the consistency of execution results of the generated code. To bridge these gaps between existing benchmarks and expectations from practical applications, we introduce CodeScope, an execution-based, multilingual, multi-task, multi-dimensional evaluation benchmark for comprehensively gauging LLM capabilities on coding tasks. CodeScope covers 43 programming languages and 8 coding tasks. It evaluates the coding performance of LLMs from three dimensions (perspectives): difficulty, efficiency, and length. To facilitate execution-based evaluations of code generation, we develop MultiCodeEngine, an automated code execution engine that supports 14 programming languages. Finally, we systematically evaluate and analyze 8 mainstream LLMs on CodeScope tasks and demonstrate the superior breadth and challenges of CodeScope for evaluating LLMs on code understanding and generation tasks compared to other benchmarks. The CodeScope benchmark and datasets are publicly available at https://github.com/WeixiangYAN/CodeScope.
Authors: Bhaktipriya Radharapu, Kevin Robinson, Lora Aroyo, Preethi Lahoti
Adversarial testing of large language models (LLMs) is crucial for their safe and responsible deployment. We introduce a novel approach for automated generation of adversarial evaluation datasets to test the safety of LLM generations on new downstream applications. We call it AI-assisted Red-Teaming (AART) - an automated alternative to current manual red-teaming efforts. AART offers a data generation and augmentation pipeline of reusable and customizable recipes that reduce human effort significantly and enable integration of adversarial testing earlier in new product development. AART generates evaluation datasets with high diversity of content characteristics critical for effective adversarial testing (e.g. sensitive and harmful concepts, specific to a wide range of cultural and geographic regions and application scenarios). The data generation is steered by AI-assisted recipes to define, scope and prioritize diversity within the application context. This feeds into a structured LLM-generation process that scales up evaluation priorities. Compared to some state-of-the-art tools, AART shows promising results in terms of concept coverage and data quality.
Authors: David F. Jenny, Yann Billeter, Mrinmaya Sachan, Bernhard Schölkopf, Zhijing Jin
The rapid advancement of Large Language Models (LLMs) has sparked intense debate regarding their ability to perceive and interpret complex socio-political landscapes. In this study, we undertake an exploration of decision-making processes and inherent biases within LLMs, exemplified by ChatGPT, specifically contextualizing our analysis within political debates. We aim not to critique or validate LLMs' values, but rather to discern how they interpret and adjudicate "good arguments." By applying Activity Dependency Networks (ADNs), we extract the LLMs' implicit criteria for such assessments and illustrate how normative values influence these perceptions. We discuss the consequences of our findings for human-AI alignment and bias mitigation. Our code and data at https://github.com/david-jenny/LLM-Political-Study.
Authors: Zichen Chen, Jianda Chen, Mitali Gaidhani, Ambuj Singh, Misha Sra
Large Language Models (LLMs) have recently made impressive strides in natural language understanding tasks. Despite their remarkable performance, understanding their decision-making process remains a big challenge. In this paper, we look into bringing some transparency to this process by introducing a new explanation dataset for question answering (QA) tasks that integrates knowledge graphs (KGs) in a novel way. Our dataset includes 12,102 question-answer-explanation (QAE) triples. Each explanation in the dataset links the LLM's reasoning to entities and relations in the KGs. The explanation component includes a why-choose explanation, a why-not-choose explanation, and a set of reason-elements that underlie the LLM's decision. We leverage KGs and graph attention networks (GAT) to find the reason-elements and transform them into why-choose and why-not-choose explanations that are comprehensible to humans. Through quantitative and qualitative evaluations, we demonstrate the potential of our dataset to improve the in-context learning of LLMs, and enhance their interpretability and explainability. Our work contributes to the field of explainable AI by enabling a deeper understanding of the LLMs decision-making process to make them more transparent and thereby, potentially more reliable, to researchers and practitioners alike. Our dataset is available at: https://github.com/chen-zichen/XplainLLM_dataset.git
Authors: Guangyin Jin, Lingbo Liu, Fuxian Li, Jincai Huang
Traffic congestion event prediction is an important yet challenging task in intelligent transportation systems. Many existing works about traffic prediction integrate various temporal encoders and graph convolution networks (GCNs), called spatio-temporal graph-based neural networks, which focus on predicting dense variables such as flow, speed and demand in time snapshots, but they can hardly forecast the traffic congestion events that are sparsely distributed on the continuous time axis. In recent years, neural point process (NPP) has emerged as an appropriate framework for event prediction in continuous time scenarios. However, most conventional works about NPP cannot model the complex spatio-temporal dependencies and congestion evolution patterns. To address these limitations, we propose a spatio-temporal graph neural point process framework, named STGNPP for traffic congestion event prediction. Specifically, we first design the spatio-temporal graph learning module to fully capture the long-range spatio-temporal dependencies from the historical traffic state data along with the road network. The extracted spatio-temporal hidden representation and congestion event information are then fed into a continuous gated recurrent unit to model the congestion evolution patterns. In particular, to fully exploit the periodic information, we also improve the intensity function calculation of the point process with a periodic gated mechanism. Finally, our model simultaneously predicts the occurrence time and duration of the next congestion. Extensive experiments on two real-world datasets demonstrate that our method achieves superior performance in comparison to existing state-of-the-art approaches.
Authors: Yiheng Su, Juni Jessy Li, Matthew Lease
Can we preserve the accuracy of neural models while also providing faithful explanations? We present wrapper boxes, a general approach to generate faithful, example-based explanations for model predictions while maintaining predictive performance. After training a neural model as usual, its learned feature representation is input to a classic, interpretable model to perform the actual prediction. This simple strategy is surprisingly effective, with results largely comparable to those of the original neural model, as shown across three large pre-trained language models, two datasets of varying scale, four classic models, and four evaluation metrics. Moreover, because these classic models are interpretable by design, the subset of training examples that determine classic model predictions can be shown directly to users.
Authors: Yuhang Zhou, Paiheng Xu, Xiaoyu Liu, Bang An, Wei Ai, Furong Huang
Language models (LMs) have gained great achievement in various NLP tasks for both fine-tuning and in-context learning (ICL) methods. Despite its outstanding performance, evidence shows that spurious correlations caused by imbalanced label distributions in training data (or exemplars in ICL) lead to robustness issues. However, previous studies mostly focus on word- and phrase-level features and fail to tackle it from the concept level, partly due to the lack of concept labels and subtle and diverse expressions of concepts in text. In this paper, we first use the LLM to label the concept for each text and then measure the concept bias of models for fine-tuning or ICL on the test data. Second, we propose a data rebalancing method to mitigate the spurious correlations by adding the LLM-generated counterfactual data to make a balanced label distribution for each concept. We verify the effectiveness of our mitigation method and show its superiority over the token removal method. Overall, our results show that there exist label distribution biases in concepts across multiple text classification datasets, and LMs will utilize these shortcuts to make predictions in both fine-tuning and ICL methods.
Authors: Juyeon Yoon, Robert Feldt, Shin Yoo
GUI testing checks if a software system behaves as expected when users interact with its graphical interface, e.g., testing specific functionality or validating relevant use case scenarios. Currently, deciding what to test at this high level is a manual task since automated GUI testing tools target lower level adequacy metrics such as structural code coverage or activity coverage. We propose DroidAgent, an autonomous GUI testing agent for Android, for semantic, intent-driven automation of GUI testing. It is based on Large Language Models and support mechanisms such as long- and short-term memory. Given an Android app, DroidAgent sets relevant task goals and subsequently tries to achieve them by interacting with the app. Our empirical evaluation of DroidAgent using 15 apps from the Themis benchmark shows that it can set up and perform realistic tasks, with a higher level of autonomy. For example, when testing a messaging app, DroidAgent created a second account and added a first account as a friend, testing a realistic use case, without human intervention. On average, DroidAgent achieved 61% activity coverage, compared to 51% for current state-of-the-art GUI testing techniques. Further, manual analysis shows that 317 out of the 374 autonomously created tasks are realistic and relevant to app functionalities, and also that DroidAgent interacts deeply with the apps and covers more features.
Authors: Vatsal Gupta, Pranshu Pandya, Tushar Kataria, Vivek Gupta, Dan Roth
Language models, given their black-box nature, often exhibit sensitivity to input perturbations, leading to trust issues due to hallucinations. To bolster trust, it's essential to understand these models' failure modes and devise strategies to enhance their performance. In this study, we propose a framework to study the effect of input perturbations on language models of different scales, from pre-trained models to large language models (LLMs). We use fine-tuning to train a robust model to perturbations, and we investigate whether exposure to one perturbation improves or degrades the model's performance on other perturbations. To address multi-perturbation robustness, we suggest three distinct training strategies. We also extend the framework to LLMs via a chain of thought(COT) prompting with exemplars. We instantiate our framework for the Tabular-NLI task and show that the proposed strategies train the model robust to different perturbations without losing accuracy on a given dataset.
Authors: Taiwei Shi, Kai Chen, Jieyu Zhao
Reinforcement Learning from Human Feedback (RLHF) is a vital strategy for enhancing model safety in language models. However, annotating preference data for RLHF is a resource-intensive and creativity-demanding process, while automatic generation methods face limitations in data diversity and quality. In response, we present Safer-Instruct, a novel pipeline for semi-automatically constructing large-scale preference datasets. Our approach leverages reversed instruction tuning, instruction induction, and expert model evaluation to efficiently generate high-quality preference data without human annotators. We evaluate Safer-Instruct using LLaMA for instruction induction and GPT-4 as an expert model, generating approximately 10K preference samples. Finetuning an Alpaca model on this dataset demonstrates improved harmlessness while maintaining competitive performance on conversation and downstream tasks. Safer-Instruct addresses the challenges in preference data acquisition, advancing the development of safer and more responsible AI systems. Our code and data are available at https://github.com/uscnlp-lime/safer-instruct
Authors: Keith Harrigian, Tina Tang, Anthony Gonzales, Cindy X. Cai, Mark Dredze
Diabetic eye disease is a major cause of blindness worldwide. The ability to monitor relevant clinical trajectories and detect lapses in care is critical to managing the disease and preventing blindness. Alas, much of the information necessary to support these goals is found only in the free text of the electronic medical record. To fill this information gap, we introduce a system for extracting evidence from clinical text of 19 clinical concepts related to diabetic eye disease and inferring relevant attributes for each. In developing this ophthalmology phenotyping system, we are also afforded a unique opportunity to evaluate the effectiveness of clinical language models at adapting to new clinical domains. Across multiple training paradigms, we find that BERT language models pretrained on out-of-distribution clinical data offer no significant improvement over BERT language models pretrained on non-clinical data for our domain. Our study tempers recent claims that language models pretrained on clinical data are necessary for clinical NLP tasks and highlights the importance of not treating clinical language data as a single homogeneous domain.
Authors: David R. Mandel
Artificial general intelligence (AGI) does not yet exist, but given the pace of technological development in artificial intelligence, it is projected to reach human-level intelligence within roughly the next two decades. After that, many experts expect it to far surpass human intelligence and to do so rapidly. The prospect of superintelligent AGI poses an existential risk to humans because there is no reliable method for ensuring that AGI goals stay aligned with human goals. Drawing on publicly available forecaster and opinion data, the author examines how experts and non-experts perceive risk from AGI. The findings indicate that the perceived risk of a world catastrophe or extinction from AGI is greater than for other existential risks. The increase in perceived risk over the last year is also steeper for AGI than for other existential threats (e.g., nuclear war or human-caused climate change). That AGI is a pressing existential risk is something on which experts and non-experts agree, but the basis for such agreement currently remains obscure.
Authors: Julian Michael, Salsabila Mahdi, David Rein, Jackson Petty, Julien Dirani, Vishakh Padmakumar, Samuel R. Bowman
As AI systems are used to answer more difficult questions and potentially help create new knowledge, judging the truthfulness of their outputs becomes more difficult and more important. How can we supervise unreliable experts, which have access to the truth but may not accurately report it, to give answers that are systematically true and don't just superficially seem true, when the supervisor can't tell the difference between the two on their own? In this work, we show that debate between two unreliable experts can help a non-expert judge more reliably identify the truth. We collect a dataset of human-written debates on hard reading comprehension questions where the judge has not read the source passage, only ever seeing expert arguments and short quotes selectively revealed by 'expert' debaters who have access to the passage. In our debates, one expert argues for the correct answer, and the other for an incorrect answer. Comparing debate to a baseline we call consultancy, where a single expert argues for only one answer which is correct half of the time, we find that debate performs significantly better, with 84% judge accuracy compared to consultancy's 74%. Debates are also more efficient, being 68% of the length of consultancies. By comparing human to AI debaters, we find evidence that with more skilled (in this case, human) debaters, the performance of debate goes up but the performance of consultancy goes down. Our error analysis also supports this trend, with 46% of errors in human debate attributable to mistakes by the honest debater (which should go away with increased skill); whereas 52% of errors in human consultancy are due to debaters obfuscating the relevant evidence from the judge (which should become worse with increased skill). Overall, these results show that debate is a promising approach for supervising increasingly capable but potentially unreliable AI systems.
Authors: Marcio Fonseca, Shay B. Cohen
Although large language models (LLMs) exhibit remarkable capacity to leverage in-context demonstrations, it is still unclear to what extent they can learn new concepts or facts from ground-truth labels. To address this question, we examine the capacity of instruction-tuned LLMs to follow in-context concept guidelines for sentence labeling tasks. We design guidelines that present different types of factual and counterfactual concept definitions, which are used as prompts for zero-shot sentence classification tasks. Our results show that although concept definitions consistently help in task performance, only the larger models (with 70B parameters or more) have limited ability to work under counterfactual contexts. Importantly, only proprietary models such as GPT-3.5 and GPT-4 can recognize nonsensical guidelines, which we hypothesize is due to more sophisticated alignment methods. Finally, we find that Falcon-180B-chat is outperformed by Llama-2-70B-chat is most cases, which indicates that careful fine-tuning is more effective than increasing model scale. Altogether, our simple evaluation method reveals significant gaps in concept understanding between the most capable open-source language models and the leading proprietary APIs.
Authors: Ethan Shaotran, Ido Pesok, Sam Jones, Emi Liu
We are introducing Aligned, a platform for global governance and alignment of frontier models, and eventually superintelligence. While previous efforts at the major AI labs have attempted to gather inputs for alignment, these are often conducted behind closed doors. We aim to set the foundation for a more trustworthy, public-facing approach to safety: a constitutional committee framework. Initial tests with 680 participants result in a 30-guideline constitution with 93% overall support. We show the platform naturally scales, instilling confidence and enjoyment from the community. We invite other AI labs and teams to plug and play into the Aligned ecosystem.
Authors: Yu Min Park, Yan Kyaw Tun, Choong Seon Hong
The development of 6G/B5G wireless networks, which have requirements that go beyond current 5G networks, is gaining interest from academic and industrial. However, to increase 6G/B5G network quality, conventional cellular networks that rely on terrestrial base stations are constrained geographically and economically. Meanwhile, NOMA allows multiple users to share the same resources, which improves the spectral efficiency of the system and has the advantage of supporting a larger number of users. Additionally, by intelligently manipulating the phase and amplitude of both the reflected and transmitted signals, STAR-RISs can achieve improved coverage, increased spectral efficiency, and enhanced communication reliability. However, STAR-RISs must simultaneously optimize the Amplitude and Phase-shift corresponding to reflection and transmission, which makes the existing terrestiral networks more complicated and is considered a major challenging issue. Motivated by the above, we study the joint user pairing for NOMA and beamforming design of Multi-STAR-RISs in an indoor environment. Then, we formulate the optimization problem with the objective of maximizing the total throughput of MUs by jointly optimizing the decoding order, user pairing, active beamforming, and passive beamforming. However, the formulated problem is a MINLP. To tackle this challenge, we first introduce the decoding order for NOMA networks. Next, we decompose the original problem into two subproblems namely: 1) MU pairing and 2) Beamforming optimization under the optimal decoding order. For the first subproblem, we employ correlation-based K-means clustering to solve the user pairing problem. Then, to jointly deal with beamforming vector optimizations, we propose MAPPO, which can make quick decisions in the given environment owing to its low complexity.
Authors: Hendrik Buschmeier, Heike M. Buhl, Friederike Kern, Angela Grimminger, Helen Beierling, Josephine Fisher, André Groß, Ilona Horwath, Nils Klowait, Stefan Lazarov, Michael Lenke, Vivien Lohmer, Katharina Rohlfing, Ingrid Scharlau, Amit Singh, Lutz Terfloth, Anna-Lisa Vollmer, Yu Wang, Annedore Wilmes, Britta Wrede
Explainability has become an important topic in computer science and artificial intelligence, leading to a subfield called Explainable Artificial Intelligence (XAI). The goal of providing or seeking explanations is to achieve (better) 'understanding' on the part of the explainee. However, what it means to 'understand' is still not clearly defined, and the concept itself is rarely the subject of scientific investigation. This conceptual article aims to present a model of forms of understanding in the context of XAI and beyond. From an interdisciplinary perspective bringing together computer science, linguistics, sociology, and psychology, a definition of understanding and its forms, assessment, and dynamics during the process of giving everyday explanations are explored. Two types of understanding are considered as possible outcomes of explanations, namely enabledness, 'knowing how' to do or decide something, and comprehension, 'knowing that' -- both in different degrees (from shallow to deep). Explanations regularly start with shallow understanding in a specific domain and can lead to deep comprehension and enabledness of the explanandum, which we see as a prerequisite for human users to gain agency. In this process, the increase of comprehension and enabledness are highly interdependent. Against the background of this systematization, special challenges of understanding in XAI are discussed.
Authors: Xiaoshuang Chen, Zhongyi Sun, Ke Yan, Shouhong Ding, Hongtao Lu
Class Incremental Learning (CIL) aims to handle the scenario where data of novel classes occur continuously and sequentially. The model should recognize the sequential novel classes while alleviating the catastrophic forgetting. In the self-supervised manner, it becomes more challenging to avoid the conflict between the feature embedding spaces of novel classes and old ones without any class labels. To address the problem, we propose a self-supervised CIL framework CPPF, meaning Combining Past, Present and Future. In detail, CPPF consists of a prototype clustering module (PC), an embedding space reserving module (ESR) and a multi-teacher distillation module (MTD). 1) The PC and the ESR modules reserve embedding space for subsequent phases at the prototype level and the feature level respectively to prepare for knowledge learned in the future. 2) The MTD module maintains the representations of the current phase without the interference of past knowledge. One of the teacher networks retains the representations of the past phases, and the other teacher network distills relation information of the current phase to the student network. Extensive experiments on CIFAR100 and ImageNet100 datasets demonstrate that our proposed method boosts the performance of self-supervised class incremental learning. We will release code in the near future.
Authors: Giovanni Sileno, Jean-Louis Dessalles
Unexpectedness is a central concept in Simplicity Theory, a theory of cognition relating various inferential processes to the computation of Kolmogorov complexities, rather than probabilities. Its predictive power has been confirmed by several experiments with human subjects, yet its theoretical basis remains largely unexplored: why does it work? This paper lays the groundwork for three theoretical conjectures. First, unexpectedness can be seen as a generalization of Bayes' rule. Second, the frequentist core of unexpectedness can be connected to the function of tracking ergodic properties of the world. Third, unexpectedness can be seen as constituent of various measures of divergence between the entropy of the world (environment) and the variety of the observer (system). The resulting framework hints to research directions that go beyond the division between probabilistic and logical approaches, potentially bringing new insights into the extraction of causal relations, and into the role of descriptive mechanisms in learning.
Authors: Sveta Paster, Kantwon Rogers, Gordon Briggs, Peter Stone, Reuth Mirsky
With the projected surge in the elderly population, service robots offer a promising avenue to enhance their well-being in elderly care homes. Such robots will encounter complex scenarios which will require them to perform decisions with ethical consequences. In this report, we propose to leverage the Intelligent Disobedience framework in order to give the robot the ability to perform a deliberation process over decisions with potential ethical implications. We list the issues that this framework can assist with, define it formally in the context of the specific elderly care home scenario, and delineate the requirements for implementing an intelligently disobeying robot. We conclude this report with some critical analysis and suggestions for future work.
Authors: Minqian Liu, Ying Shen, Zhiyang Xu, Yixin Cao, Eunah Cho, Vaibhav Kumar, Reza Ghanadan, Lifu Huang
Natural Language Generation (NLG) typically involves evaluating the generated text in various aspects (e.g., consistency and naturalness) to obtain a comprehensive assessment. However, multi-aspect evaluation remains challenging as it may require the evaluator to generalize to any given evaluation aspect even if it's absent during training. In this paper, we introduce X-Eval, a two-stage instruction tuning framework to evaluate the text in both seen and unseen aspects customized by end users. X-Eval consists of two learning stages: the vanilla instruction tuning stage that improves the model's ability to follow evaluation instructions, and an enhanced instruction tuning stage that exploits the connections between fine-grained evaluation aspects to better assess text quality. To support the training of X-Eval, we collect AspectInstruct, the first instruction tuning dataset tailored for multi-aspect NLG evaluation spanning 27 diverse evaluation aspects with 65 tasks. To enhance task diversity, we devise an augmentation strategy that converts human rating annotations into diverse forms of NLG evaluation tasks, including scoring, comparison, ranking, and Boolean question answering. Extensive experiments across three essential categories of NLG tasks: dialogue generation, summarization, and data-to-text coupled with 21 aspects in meta-evaluation, demonstrate that our X-Eval enables even a lightweight language model to achieve a comparable if not higher correlation with human judgments compared to the state-of-the-art NLG evaluators, such as GPT-4.
Authors: Yue Liu, Shanlin Xiao, Bo Li, Zhiyi Yu
As the third-generation neural network, the Spiking Neural Network (SNN) has the advantages of low power consumption and high energy efficiency, making it suitable for implementation on edge devices. More recently, the most advanced SNN, Spikformer, combines the self-attention module from Transformer with SNN to achieve remarkable performance. However, it adopts larger channel dimensions in MLP layers, leading to an increased number of redundant model parameters. To effectively decrease the computational complexity and weight parameters of the model, we explore the Lottery Ticket Hypothesis (LTH) and discover a very sparse ($\ge$90%) subnetwork that achieves comparable performance to the original network. Furthermore, we also design a lightweight token selector module, which can remove unimportant background information from images based on the average spike firing rate of neurons, selecting only essential foreground image tokens to participate in attention calculation. Based on that, we present SparseSpikformer, a co-design framework aimed at achieving sparsity in Spikformer through token and weight pruning techniques. Experimental results demonstrate that our framework can significantly reduce 90% model parameters and cut down Giga Floating-Point Operations (GFLOPs) by 20% while maintaining the accuracy of the original model.
Authors: Cian Eastwood, Julius von Kügelgen, Linus Ericsson, Diane Bouchacourt, Pascal Vincent, Bernhard Schölkopf, Mark Ibrahim
Self-supervised representation learning often uses data augmentations to induce some invariance to "style" attributes of the data. However, with downstream tasks generally unknown at training time, it is difficult to deduce a priori which attributes of the data are indeed "style" and can be safely discarded. To address this, we introduce a more principled approach that seeks to disentangle style features rather than discard them. The key idea is to add multiple style embedding spaces where: (i) each is invariant to all-but-one augmentation; and (ii) joint entropy is maximized. We formalize our structured data-augmentation procedure from a causal latent-variable-model perspective, and prove identifiability of both content and (multiple blocks of) style variables. We empirically demonstrate the benefits of our approach on synthetic datasets and then present promising but limited results on ImageNet.
Authors: Davis Yoshida, Kartik Goyal, Kevin Gimpel
It has been widely observed that exact or approximate MAP (mode-seeking) decoding from natural language generation (NLG) models consistently leads to degenerate outputs (Stahlberg and Byrne, 2019, Holtzman et al., 2019). This has generally been attributed to either a fundamental inadequacy of modes in models or weaknesses in language modeling. Contrastingly in this work, we emphasize that degenerate modes can even occur in the absence of any model error, due to contamination of the training data. Specifically, we show that mixing even a tiny amount of low-entropy noise with a population text distribution can cause the data distribution's mode to become degenerate, implying that any models trained on it will be as well. As the unconditional mode of NLG models will often be degenerate, we therefore propose to apply MAP decoding to the model's distribution conditional on avoiding specific degeneracies. Using exact-search, we empirically verify that the length-conditional modes of machine translation models and language models are indeed more fluent and topical than their unconditional modes. For the first time, we also share many examples of exact modal sequences from these models, and from several variants of the LLaMA-7B model. Notably, the modes of the LLaMA models are still degenerate, showing that improvements in modeling have not fixed this issue. Because of the cost of exact mode finding algorithms, we develop an approximate mode finding approach, ACBS, which finds sequences that are both high-likelihood and high-quality. We apply this approach to LLaMA-7B, a model which was not trained for instruction following, and find that we are able to elicit reasonable outputs without any finetuning.
Authors: Donghyeok Shin, Seungjae Shin, Il-Chul Moon
This paper presents FreD, a novel parameterization method for dataset distillation, which utilizes the frequency domain to distill a small-sized synthetic dataset from a large-sized original dataset. Unlike conventional approaches that focus on the spatial domain, FreD employs frequency-based transforms to optimize the frequency representations of each data instance. By leveraging the concentration of spatial domain information on specific frequency components, FreD intelligently selects a subset of frequency dimensions for optimization, leading to a significant reduction in the required budget for synthesizing an instance. Through the selection of frequency dimensions based on the explained variance, FreD demonstrates both theoretical and empirical evidence of its ability to operate efficiently within a limited budget, while better preserving the information of the original dataset compared to conventional parameterization methods. Furthermore, based on the orthogonal compatibility of FreD with existing methods, we confirm that FreD consistently improves the performances of existing distillation methods over the evaluation scenarios with different benchmark datasets. We release the code at https://github.com/sdh0818/FreD.
Authors: Filippo Airaldi, Bart De Schutter, Azita Dabiri
In the backdrop of an increasingly pressing need for effective urban and highway transportation systems, this work explores the synergy between model-based and learning-based strategies to enhance traffic flow management by use of an innovative approach to the problem of highway ramp metering control that embeds Reinforcement Learning techniques within the Model Predictive Control framework. The control problem is formulated as an RL task by crafting a suitable stage cost function that is representative of the traffic conditions, variability in the control action, and violations of a safety-critical constraint on the maximum number of vehicles in queue. An MPC-based RL approach, which merges the advantages of the two paradigms in order to overcome the shortcomings of each framework, is proposed to learn to efficiently control an on-ramp and to satisfy its constraints despite uncertainties in the system model and variable demands. Finally, simulations are performed on a benchmark from the literature consisting of a small-scale highway network. Results show that, starting from an MPC controller that has an imprecise model and is poorly tuned, the proposed methodology is able to effectively learn to improve the control policy such that congestion in the network is reduced and constraints are satisfied, yielding an improved performance compared to the initial controller.
Authors: Malak Sadek, Céline Mougenot
Recent years have seen a steady rise in the popularity and use of Conversational Agents (CA) for different applications, well before the more immediate impact of large language models. This rise has been accompanied by an extensive exploration and documentation of the challenges of designing and creating conversational agents. Focusing on a recent scoping review of the socio-technical challenges of CA creation, this opinion paper calls for an examination of the extent to which interdisciplinary collaboration (IDC) challenges might contribute towards socio-technical CA design challenges. The paper proposes a taxonomy of CA design challenges using IDC as a lens, and proposes practical strategies to overcome them which complement existing design principles. The paper invites future work to empirically verify suggested conceptual links and apply the proposed strategies within the space of CA design to evaluate their effectiveness.
Authors: Ba Luat Le, Layla Martin, Emrah Demir, Duc Minh Vu
We study an optimal investment problem that arises in the context of the vehicle-sharing system. Given a set of locations to build stations, we need to determine i) the sequence of stations to be built and the number of vehicles to acquire in order to obtain the target state where all stations are built, and ii) the number of vehicles to acquire and their allocation in order to maximize the total profit returned by operating the system when some or all stations are open. The profitability associated with operating open stations, measured over a specific time period, is represented as a linear optimization problem applied to a collection of open stations. With operating capital, the owner of the system can open new stations. This property introduces a set-dependent aspect to the duration required for opening a new station, and the optimal investment problem can be viewed as a variant of the Traveling Salesman Problem (TSP) with set-dependent cost. We propose an A* search algorithm to address this particular variant of the TSP. Computational experiments highlight the benefits of the proposed algorithm in comparison to the widely recognized Dijkstra algorithm and propose future research to explore new possibilities and applications for both exact and approximate A* algorithms.
Authors: Spencer Rarrick, Ranjita Naik, Sundar Poudel, Vishal Chowdhary
Machine Translation (MT) continues to improve in quality and adoption, yet the inadvertent perpetuation of gender bias remains a significant concern. Despite numerous studies into gender bias in translations from gender-neutral languages such as Turkish into more strongly gendered languages like English, there are no benchmarks for evaluating this phenomenon or for assessing mitigation strategies. To address this gap, we introduce GATE X-E, an extension to the GATE (Rarrick et al., 2023) corpus, that consists of human translations from Turkish, Hungarian, Finnish, and Persian into English. Each translation is accompanied by feminine, masculine, and neutral variants for each possible gender interpretation. The dataset, which contains between 1250 and 1850 instances for each of the four language pairs, features natural sentences with a wide range of sentence lengths and domains, challenging translation rewriters on various linguistic phenomena. Additionally, we present an English gender rewriting solution built on GPT-3.5 Turbo and use GATE X-E to evaluate it. We open source our contributions to encourage further research on gender debiasing.
Authors: Matt Kaufmann (UT Austin, retired), J Strother Moore (UT Austin, retired)
The experience of an ACL2 user generally includes many failed proof attempts. A key to successful use of the ACL2 prover is the effective use of tools to debug those failures. We focus on changes made after ACL2 Version 8.5: the improved break-rewrite utility and the new utility, with-brr-data.
Authors: Mayur Patidar, Avinash Singh, Riya Sawhney, Indrajit Bhattacharya, Mausam
We address the zero-shot transfer learning setting for the knowledge base question answering (KBQA) problem, where a large volume of labeled training data is available for the source domain, but no such labeled examples are available for the target domain. Transfer learning for KBQA makes use of large volumes of unlabeled data in the target in addition to the labeled data in the source. More recently, few-shot in-context learning using Black-box Large Language Models (BLLMs) has been adapted for KBQA without considering any source domain data. In this work, we show how to meaningfully combine these two paradigms for KBQA so that their benefits add up. Specifically, we preserve the two stage retrieve-then-generate pipeline of supervised KBQA and introduce interaction between in-context learning using BLLMs and transfer learning from the source for both stages. In addition, we propose execution-guided self-refinement using BLLMs, decoupled from the transfer setting. With the help of experiments using benchmark datasets GrailQA as the source and WebQSP as the target, we show that the proposed combination brings significant improvements to both stages and also outperforms by a large margin state-of-the-art supervised KBQA models trained on the source. We also show that in the in-domain setting, the proposed BLLM augmentation significantly outperforms state-of-the-art supervised models, when the volume of labeled data is limited, and also outperforms these marginally even when using the entire large training dataset.
Authors: Ahmed Emam, Timo T. Stomberg, Ribana Roscher
Natural protected areas are vital for biodiversity, climate change mitigation, and supporting ecological processes. Despite their significance, comprehensive mapping is hindered by a lack of understanding of their characteristics and a missing land cover class definition. This paper aims to advance the explanation of the designating patterns forming protected and wild areas. To this end, we propose a novel framework that uses activation maximization and a generative adversarial model. With this, we aim to generate satellite images that, in combination with domain knowledge, are capable of offering complete and valid explanations for the spatial and spectral patterns that define the natural authenticity of these regions. Our proposed framework produces more precise attribution maps pinpointing the designating patterns forming the natural authenticity of protected areas. Our approach fosters our understanding of the ecological integrity of the protected natural areas and may contribute to future monitoring and preservation efforts.
Authors: Yixiu Mao, Hongchang Zhang, Chen Chen, Yi Xu, Xiangyang Ji
Offline reinforcement learning suffers from the out-of-distribution issue and extrapolation error. Most policy constraint methods regularize the density of the trained policy towards the behavior policy, which is too restrictive in most cases. We propose Supported Trust Region optimization (STR) which performs trust region policy optimization with the policy constrained within the support of the behavior policy, enjoying the less restrictive support constraint. We show that, when assuming no approximation and sampling error, STR guarantees strict policy improvement until convergence to the optimal support-constrained policy in the dataset. Further with both errors incorporated, STR still guarantees safe policy improvement for each step. Empirical results validate the theory of STR and demonstrate its state-of-the-art performance on MuJoCo locomotion domains and much more challenging AntMaze domains.
Authors: Angelos Poulis, Eleni Tsalapati, Manolis Koubarakis
One way that the current state of the art measures the reasoning ability of transformer-based models is by evaluating accuracy in downstream tasks like logical question answering or proof generation over synthetic contexts expressed in natural language. However, most of the contexts used are in practice very simple; in most cases, they are generated from short first-order logic sentences with only a few logical operators and quantifiers. In this work, we seek to answer the question how well a transformer-based model will perform reasoning over expressive contexts. For this purpose, we construct a synthetic natural language question-answering dataset, generated by description logic knowledge bases. For the generation of the knowledge bases, we use the expressive language $\mathcal{ALCQ}$. The resulting dataset contains 384K examples, and increases in two dimensions: i) reasoning depth, and ii) length of sentences. We show that the performance of our DeBERTa-based model, DELTA$_M$, is marginally affected when the reasoning depth is increased and it is not affected at all when the length of the sentences is increasing. We also evaluate the generalization ability of the model on reasoning depths unseen at training, both increasing and decreasing, revealing interesting insights into the model's adaptive generalization abilities.
Authors: Kerianne L. Hobbs, Bernard Li
Designing a safe, trusted, and ethical AI may be practically impossible; however, designing AI with safe, trusted, and ethical use in mind is possible and necessary in safety and mission-critical domains like aerospace. Safe, trusted, and ethical use of AI are often used interchangeably; however, a system can be safely used but not trusted or ethical, have a trusted use that is not safe or ethical, and have an ethical use that is not safe or trusted. This manuscript serves as a primer to illuminate the nuanced differences between these concepts, with a specific focus on applications of Human-AI teaming in aerospace system control, where humans may be in, on, or out-of-the-loop of decision-making.
Authors: Giulio Antonio Abbo, Tony Belpaeme
In the rapidly evolving landscape of human-computer interaction, the integration of vision capabilities into conversational agents stands as a crucial advancement. This paper presents an initial implementation of a dialogue manager that leverages the latest progress in Large Language Models (e.g., GPT-4, IDEFICS) to enhance the traditional text-based prompts with real-time visual input. LLMs are used to interpret both textual prompts and visual stimuli, creating a more contextually aware conversational agent. The system's prompt engineering, incorporating dialogue with summarisation of the images, ensures a balance between context preservation and computational efficiency. Six interactions with a Furhat robot powered by this system are reported, illustrating and discussing the results obtained. By implementing this vision-enabled dialogue system, the paper envisions a future where conversational agents seamlessly blend textual and visual modalities, enabling richer, more context-aware dialogues.
Authors: David Chanin, Anthony Hunter, Oana-Maria Camburu
Transformer language models (LMs) have been shown to represent concepts as directions in the latent space of hidden activations. However, for any given human-interpretable concept, how can we find its direction in the latent space? We present a technique called linear relational concepts (LRC) for finding concept directions corresponding to human-interpretable concepts at a given hidden layer in a transformer LM by first modeling the relation between subject and object as a linear relational embedding (LRE). While the LRE work was mainly presented as an exercise in understanding model representations, we find that inverting the LRE while using earlier object layers results in a powerful technique to find concept directions that both work well as a classifier and causally influence model outputs.
Authors: Marie Farrell (University of Manchester, UK), Matt Luckcuck (University of Nottingham, UK), Mario Gleirscher (University of Bremen, Germany), Maike Schwammberger (Karlsruhe Institute of Technology, Germany)
This EPTCS volume contains the proceedings for the Fifth International Workshop on Formal Methods for Autonomous Systems (FMAS 2023), which was held on the 15th and 16th of November 2023. FMAS 2023 was co-located with 18th International Conference on integrated Formal Methods (iFM) (iFM'22), organised by Leiden Institute of Advanced Computer Science of Leiden University. The workshop itself was held at Scheltema Leiden, a renovated 19th Century blanket factory alongside the canal.
FMAS 2023 received 25 submissions. We received 11 regular papers, 3 experience reports, 6 research previews, and 5 vision papers. The researchers who submitted papers to FMAS 2023 were from institutions in: Australia, Canada, Colombia, France, Germany, Ireland, Italy, the Netherlands, Sweden, the United Kingdom, and the United States of America. Increasing our number of submissions for the third year in a row is an encouraging sign that FMAS has established itself as a reputable publication venue for research on the formal modelling and verification of autonomous systems. After each paper was reviewed by three members of our Programme Committee we accepted a total of 15 papers: 8 long papers and 7 short papers.
Authors: Hao Peng, Xiaozhi Wang, Jianhui Chen, Weikai Li, Yunjia Qi, Zimu Wang, Zhili Wu, Kaisheng Zeng, Bin Xu, Lei Hou, Juanzi Li
In-context learning (ICL) has become the default method for using large language models (LLMs), making the exploration of its limitations and understanding the underlying causes crucial. In this paper, we find that ICL falls short of handling specification-heavy tasks, which are tasks with complicated and extensive task specifications, requiring several hours for ordinary humans to master, such as traditional information extraction tasks. The performance of ICL on these tasks mostly cannot reach half of the state-of-the-art results. To explore the reasons behind this failure, we conduct comprehensive experiments on 18 specification-heavy tasks with various LLMs and identify three primary reasons: inability to specifically understand context, misalignment in task schema comprehension with humans, and inadequate long-text understanding ability. Furthermore, we demonstrate that through fine-tuning, LLMs can achieve decent performance on these tasks, indicating that the failure of ICL is not an inherent flaw of LLMs, but rather a drawback of existing alignment methods that renders LLMs incapable of handling complicated specification-heavy tasks via ICL. To substantiate this, we perform dedicated instruction tuning on LLMs for these tasks and observe a notable improvement. We hope the analyses in this paper could facilitate advancements in alignment methods enabling LLMs to meet more sophisticated human demands.
Authors: Morocco Solidarity Hackathon (Organizers, Speakers, Mentors and Participant teams)
The devastating 6.8-magnitude earthquake in Al Haouz, Morocco in 2023 prompted critical reflections on global disaster management strategies, resulting in a post-disaster hackathon, using artificial intelligence (AI) to improve disaster preparedness, response, and recovery. This paper provides (i) a comprehensive literature review, (ii) an overview of winning projects, (iii) key insights and challenges, namely real-time open-source data, data scarcity, and interdisciplinary collaboration barriers, and (iv) a community-call for further action.
Authors: Lorenzo Nodari
In recent years, Reward Machines (RMs) have stood out as a simple yet effective automata-based formalism for exposing and exploiting task structure in reinforcement learning settings. Despite their relevance, little to no attention has been directed to the study of their security implications and robustness to adversarial scenarios, likely due to their recent appearance in the literature. With my thesis, I aim to provide the first analysis of the security of RM-based reinforcement learning techniques, with the hope of motivating further research in the field, and I propose and evaluate a novel class of attacks on RM-based techniques: blinding attacks.
Authors: Zixiao Wang, AmirEmad Ghassami, Ilya Shpitser
We consider the task of identifying and estimating a parameter of interest in settings where data is missing not at random (MNAR). In general, such parameters are not identified without strong assumptions on the missing data model. In this paper, we take an alternative approach and introduce a method inspired by data fusion, where information in an MNAR dataset is augmented by information in an auxiliary dataset subject to missingness at random (MAR). We show that even if the parameter of interest cannot be identified given either dataset alone, it can be identified given pooled data, under two complementary sets of assumptions. We derive an inverse probability weighted (IPW) estimator for identified parameters, and evaluate the performance of our estimation strategies via simulation studies.
Authors: Lorenzo Nodari, Federico Cerutti
Robustness to noise is of utmost importance in reinforcement learning systems, particularly in military contexts where high stakes and uncertain environments prevail. Noise and uncertainty are inherent features of military operations, arising from factors such as incomplete information, adversarial actions, or unpredictable battlefield conditions. In RL, noise can critically impact decision-making, mission success, and the safety of personnel. Reward machines offer a powerful tool to express complex reward structures in RL tasks, enabling the design of tailored reinforcement signals that align with mission objectives. This paper considers the problem of the robustness of intelligence-driven reinforcement learning based on reward machines. The preliminary results presented suggest the need for further research in evidential reasoning and learning to harden current state-of-the-art reinforcement learning approaches before being mission-critical-ready.
Authors: Ziyin Zhang, Yikang Liu, Weifang Huang, Junyu Mao, Rui Wang, Hai Hu
Recent benchmarks for Large Language Models (LLMs) have mostly focused on application-driven tasks such as complex reasoning and code generation, and this has led to a scarcity in purely linguistic evaluation of LLMs. Against this background, we introduce Multilingual Evaluation of Linguistic Acceptability -- MELA, the first multilingual benchmark on linguistic acceptability with 48K samples covering 10 languages from a diverse set of language families. We establish baselines of commonly used LLMs along with supervised models, and conduct cross-lingual transfer and multi-task learning experiments with XLM-R. In pursuit of multilingual interpretability, we analyze the weights of fine-tuned XLM-R to explore the possibility of identifying transfer difficulty between languages. Our results show that ChatGPT benefits much from in-context examples but still lags behind fine-tuned XLM-R, while the performance of GPT-4 is on par with fine-tuned XLM-R even in zero-shot setting. Cross-lingual and multi-task learning experiments show that unlike semantic tasks, in-language training data is crucial in acceptability judgements. Results in layerwise probing indicate that the upper layers of XLM-R become a task-specific but language-agnostic region for multilingual acceptability judgment. We also introduce the concept of conflicting weight, which could be a potential indicator for the difficulty of cross-lingual transfer between languages. Our data will be available at https://github.com/sjtu-compling/MELA.
Authors: Yunshi Lan, Xiang Li, Xin Liu, Yang Li, Wei Qin, Weining Qian
Zero-shot Visual Question Answering (VQA) is a prominent vision-language task that examines both the visual and textual understanding capability of systems in the absence of training data. Recently, by converting the images into captions, information across multi-modalities is bridged and Large Language Models (LLMs) can apply their strong zero-shot generalization capability to unseen questions. To design ideal prompts for solving VQA via LLMs, several studies have explored different strategies to select or generate question-answer pairs as the exemplar prompts, which guide LLMs to answer the current questions effectively. However, they totally ignore the role of question prompts. The original questions in VQA tasks usually encounter ellipses and ambiguity which require intermediate reasoning. To this end, we present Reasoning Question Prompts for VQA tasks, which can further activate the potential of LLMs in zero-shot scenarios. Specifically, for each question, we first generate self-contained questions as reasoning question prompts via an unsupervised question edition module considering sentence fluency, semantic integrity and syntactic invariance. Each reasoning question prompt clearly indicates the intent of the original question. This results in a set of candidate answers. Then, the candidate answers associated with their confidence scores acting as answer heuristics are fed into LLMs and produce the final answer. We evaluate reasoning question prompts on three VQA challenges, experimental results demonstrate that they can significantly improve the results of LLMs on zero-shot setting and outperform existing state-of-the-art zero-shot methods on three out of four data sets. Our source code is publicly released at \url{https://github.com/ECNU-DASE-NLP/RQP}.
Authors: Yifan Wei, Xiaoyan Yu, Huanhuan Ma, Fangyu Lei, Yixuan Weng, Ran Song, Kang Liu
Knowledge Editing (KE) for modifying factual knowledge in Large Language Models (LLMs) has been receiving increasing attention. However, existing knowledge editing methods are entity-centric, and it is unclear whether this approach is suitable for a relation-centric perspective. To address this gap, this paper constructs a new benchmark named RaKE, which focuses on Relation based Knowledge Editing. In this paper, we establish a suite of innovative metrics for evaluation and conduct comprehensive experiments involving various knowledge editing baselines. We notice that existing knowledge editing methods exhibit the potential difficulty in their ability to edit relations. Therefore, we further explore the role of relations in factual triplets within the transformer. Our research results confirm that knowledge related to relations is not only stored in the FFN network but also in the attention layers. This provides experimental support for future relation-based knowledge editing methods.
Authors: Hakuei Yamada, Junpei Komiyama, Kenshi Abe, Atsushi Iwasaki
This work addresses learning online fair division under uncertainty, where a central planner sequentially allocates items without precise knowledge of agents' values or utilities. Departing from conventional online algorithm, the planner here relies on noisy, estimated values obtained after allocating items. We introduce wrapper algorithms utilizing \textit{dual averaging}, enabling gradual learning of both the type distribution of arriving items and agents' values through bandit feedback. This approach enables the algorithms to asymptotically achieve optimal Nash social welfare in linear Fisher markets with agents having additive utilities. We establish regret bounds in Nash social welfare and empirically validate the superior performance of our proposed algorithms across synthetic and empirical datasets.
Authors: Hyunji Lee, Sejune Joo, Chaeeun Kim, Joel Jang, Doyoung Kim, Kyoung-Woon On, Minjoon Seo
Reliance on the inherent knowledge of Large Language Models (LLMs) can cause issues such as hallucinations, lack of control, and difficulties in integrating variable knowledge. To mitigate this, LLMs can be probed to generate responses by grounding on external context, often given as input (knowledge-augmented models). Yet, previous research is often confined to a narrow view of the term "grounding", often only focusing on whether the response contains the correct answer or not, which does not ensure the reliability of the entire response. To address this limitation, we introduce a strict definition of grounding: a model is considered truly grounded when its responses (1) fully utilize necessary knowledge from the provided context, and (2) don't exceed the knowledge within the contexts. We introduce a new dataset and a grounding metric to assess this new definition and perform experiments across 13 LLMs of different sizes and training methods to provide insights into the factors that influence grounding performance. Our findings contribute to a better understanding of how to improve grounding capabilities and suggest an area of improvement toward more reliable and controllable LLM applications.
Authors: Fei Yuan, Shuai Yuan, Zhiyong Wu, Lei Li
Large Language Models (LLMs), trained predominantly on extensive English data, often exhibit limitations when applied to other languages. Current research is primarily focused on enhancing the multilingual capabilities of these models by employing various tuning strategies. Despite their effectiveness in certain languages, the understanding of the multilingual abilities of LLMs remains incomplete. This study endeavors to evaluate the multilingual capacity of LLMs by conducting an exhaustive analysis across 101 languages, and classifies languages with similar characteristics into four distinct quadrants. By delving into each quadrant, we shed light on the rationale behind their categorization and offer actionable guidelines for tuning these languages. Extensive experiments reveal that existing LLMs possess multilingual capabilities that surpass our expectations, and we can significantly improve the multilingual performance of LLMs by focusing on these distinct attributes present in each quadrant.
Authors: Arnav Arora, Maha Jinadoss, Cheshta Arora, Denny George, Brindaalakshmi, Haseena Dawood Khan, Kirti Rawat, Div, Ritash, Seema Mathur, Shivani Yadav, Shehla Rashid Shora, Rie Raut, Sumit Pawar, Apurva Paithane, Sonia, Vivek, Dharini Priscilla, Khairunnisha, Grace Banu, Ambika Tandon, Rishav Thakker, Rahul Dev Korra, Aatman Vaidya, Tarunima Prabhakar
Online gender based violence has grown concomitantly with adoption of the internet and social media. Its effects are worse in the Global majority where many users use social media in languages other than English. The scale and volume of conversations on the internet has necessitated the need for automated detection of hate speech, and more specifically gendered abuse. There is, however, a lack of language specific and contextual data to build such automated tools. In this paper we present a dataset on gendered abuse in three languages- Hindi, Tamil and Indian English. The dataset comprises of tweets annotated along three questions pertaining to the experience of gender abuse, by experts who identify as women or a member of the LGBTQIA community in South Asia. Through this dataset we demonstrate a participatory approach to creating datasets that drive AI systems.
Authors: Nadine Kroher, Helena Cuesta, Aggelos Pikrakis
We are investigating the broader concept of using AI-based generative music systems to generate training data for Music Information Retrieval (MIR) tasks. To kick off this line of work, we ran an initial experiment in which we trained a genre classifier on a fully artificial music dataset created with MusicGen. We constructed over 50 000 genre- conditioned textual descriptions and generated a collection of music excerpts that covers five musical genres. Our preliminary results show that the proposed model can learn genre-specific characteristics from artificial music tracks that generalise well to real-world music recordings.
Authors: Shumin Deng, Ningyu Zhang, Nay Oo, Bryan Hooi
Large Language Models (LLMs) employing Chain-of-Thought (CoT) prompting have broadened the scope for improving multi-step reasoning capabilities. Usually, answer calibration strategies such as step-level or path-level calibration play a vital role in multi-step reasoning. While effective, there remains a significant gap in our understanding of the key factors that drive their success. In this paper, we break down the design of recent answer calibration strategies and present a unified view which establishes connections between them. We then conduct a thorough evaluation on these strategies from a unified view, systematically scrutinizing step-level and path-level answer calibration across multiple paths. Our study holds the potential to illuminate key insights for optimizing multi-step reasoning with answer calibration.
Authors: Yusuke Sakai, Hidetaka Kamigaito, Katsuhiko Hayashi, Taro Watanabe
Knowledge graphs (KGs) consist of links that describe relationships between entities. Due to the difficulty of manually enumerating all relationships between entities, automatically completing them is essential for KGs. Knowledge Graph Completion (KGC) is a task that infers unseen relationships between entities in a KG. Traditional embedding-based KGC methods, such as RESCAL, TransE, DistMult, ComplEx, RotatE, HAKE, HousE, etc., infer missing links using only the knowledge from training data. In contrast, the recent Pre-trained Language Model (PLM)-based KGC utilizes knowledge obtained during pre-training. Therefore, PLM-based KGC can estimate missing links between entities by reusing memorized knowledge from pre-training without inference. This approach is problematic because building KGC models aims to infer unseen links between entities. However, conventional evaluations in KGC do not consider inference and memorization abilities separately. Thus, a PLM-based KGC method, which achieves high performance in current KGC evaluations, may be ineffective in practical applications. To address this issue, we analyze whether PLM-based KGC methods make inferences or merely access memorized knowledge. For this purpose, we propose a method for constructing synthetic datasets specified in this analysis and conclude that PLMs acquire the inference abilities required for KGC through pre-training, even though the performance improvements mostly come from textual information of entities and relations.
Authors: Haoqiang Kang, Juntong Ni, Huaxiu Yao
Large Language Models (LLMs) have demonstrated remarkable proficiency in generating fluent text. However, they often encounter the challenge of generating inaccurate or hallucinated content. This issue is common in both non-retrieval-based generation and retrieval-augmented generation approaches, and existing post-hoc rectification methods may not address the accumulated hallucination errors that may be caused by the "snowballing" issue, especially in reasoning tasks. To tackle these challenges, we introduce a novel approach called Real-time Verification and Rectification (Ever). Instead of waiting until the end of the generation process to rectify hallucinations, Ever employs a real-time, step-wise generation and hallucination rectification strategy. The primary objective is to detect and rectify hallucinations as they occur during the text generation process. When compared to both retrieval-based and non-retrieval-based baselines, Ever demonstrates a significant improvement in generating trustworthy and factually accurate text across a diverse range of tasks, including short-form QA, biography generation, and multi-hop reasoning.
Authors: Konstantin Hemker, Nikola Smidjievski, Mateja Jamnik
Technological advances in medical data collection such as high-resolution histopathology and high-throughput genomic sequencing have contributed to the rising requirement for multi-modal biomedical modelling, specifically for image, tabular, and graph data. Most multi-modal deep learning approaches use modality-specific architectures that are trained separately and cannot capture the crucial cross-modal information that motivates the integration of different data sources. This paper presents the Hybrid Early-fusion Attention Learning Network (HEALNet): a flexible multi-modal fusion architecture, which a) preserves modality-specific structural information, b) captures the cross-modal interactions and structural information in a shared latent space, c) can effectively handle missing modalities during training and inference, and d) enables intuitive model inspection by learning on the raw data input instead of opaque embeddings. We conduct multi-modal survival analysis on Whole Slide Images and Multi-omic data on four cancer cohorts of The Cancer Genome Atlas (TCGA). HEALNet achieves state-of-the-art performance, substantially improving over both uni-modal and recent multi-modal baselines, whilst being robust in scenarios with missing modalities.
Authors: Yuanwei Wu, Xiang Li, Yixin Liu, Pan Zhou, Lichao Sun
Existing work on jailbreak Multimodal Large Language Models (MLLMs) has focused primarily on adversarial examples in model inputs, with less attention to vulnerabilities in model APIs. To fill the research gap, we carry out the following work: 1) We discover a system prompt leakage vulnerability in GPT-4V. Through carefully designed dialogue, we successfully steal the internal system prompts of GPT-4V. This finding indicates potential exploitable security risks in MLLMs; 2)Based on the acquired system prompts, we propose a novel MLLM jailbreaking attack method termed SASP (Self-Adversarial Attack via System Prompt). By employing GPT-4 as a red teaming tool against itself, we aim to search for potential jailbreak prompts leveraging stolen system prompts. Furthermore, in pursuit of better performance, we also add human modification based on GPT-4's analysis, which further improves the attack success rate to 98.7\%; 3) We evaluated the effect of modifying system prompts to defend against jailbreaking attacks. Results show that appropriately designed system prompts can significantly reduce jailbreak success rates. Overall, our work provides new insights into enhancing MLLM security, demonstrating the important role of system prompts in jailbreaking, which could be leveraged to greatly facilitate jailbreak success rates while also holding the potential for defending against jailbreaks.
Authors: Ziyang Chen, Dongfang Li, Xiang Zhao, Baotian Hu, Min Zhang
In this paper, we tackle the significant challenge of temporal knowledge reasoning in Large Language Models (LLMs), an area where such models frequently encounter difficulties. These difficulties often result in the generation of misleading or incorrect information, primarily due to their limited capacity to process evolving factual knowledge and complex temporal logic. In response, we propose a novel, constructivism-based approach that advocates for a paradigm shift in LLM learning towards an active, ongoing process of knowledge synthesis and customization. At the heart of our proposal is the Abstract Reasoning Induction ARI framework, which divides temporal reasoning into two distinct phases: Knowledge-agnostic and Knowledge-based. This division aims to reduce instances of hallucinations and improve LLMs' capacity for integrating abstract methodologies derived from historical data. Our approach achieves remarkable improvements, with relative gains of 29.7\% and 9.27\% on two temporal QA datasets, underscoring its efficacy in advancing temporal reasoning in LLMs. The code will be released at https://github.com/czy1999/ARI.
Authors: Zhaowei Wang, Haochen Shi, Weiqi Wang, Tianqing Fang, Hongming Zhang, Sehyun Choi, Xin Liu, Yangqiu Song
Cognitive research indicates that abstraction ability is essential in human intelligence, which remains under-explored in language models. In this paper, we present AbsPyramid, a unified entailment graph of 221K textual descriptions of abstraction knowledge. While existing resources only touch nouns or verbs within simplified events or specific domains, AbsPyramid collects abstract knowledge for three components of diverse events to comprehensively evaluate the abstraction ability of language models in the open domain. Experimental results demonstrate that current LLMs face challenges comprehending abstraction knowledge in zero-shot and few-shot settings. By training on our rich abstraction knowledge, we find LLMs can acquire basic abstraction abilities and generalize to unseen events. In the meantime, we empirically show that our benchmark is comprehensive to enhance LLMs across two previous abstraction tasks.
Authors: Minghan Li, Honglei Zhuang, Kai Hui, Zhen Qin, Jimmy Lin, Rolf Jagerman, Xuanhui Wang, Michael Bendersky
Query expansion has been proved to be effective in improving recall and precision of first-stage retrievers, and yet its influence on a complicated, state-of-the-art cross-encoder ranker remains under-explored. We first show that directly applying the expansion techniques in the current literature to state-of-the-art neural rankers can result in deteriorated zero-shot performance. To this end, we propose GFF, a pipeline that includes a large language model and a neural ranker, to Generate, Filter, and Fuse query expansions more effectively in order to improve the zero-shot ranking metrics such as nDCG@10. Specifically, GFF first calls an instruction-following language model to generate query-related keywords through a reasoning chain. Leveraging self-consistency and reciprocal rank weighting, GFF further filters and combines the ranking results of each expanded query dynamically. By utilizing this pipeline, we show that GFF can improve the zero-shot nDCG@10 on BEIR and TREC DL 2019/2020. We also analyze different modelling choices in the GFF pipeline and shed light on the future directions in query expansion for zero-shot neural rankers.
Authors: Lucas Torroba Hennigen, Shannon Shen, Aniruddha Nrusimha, Bernhard Gapp, David Sontag, Yoon Kim
Large language models (LLMs) have demonstrated an impressive ability to synthesize plausible and fluent text. However they remain vulnerable to hallucinations, and thus their outputs generally require manual human verification for high-stakes applications, which can be time-consuming and difficult. This paper proposes symbolically grounded generation (SymGen) as a simple approach for enabling easier validation of an LLM's output. SymGen prompts an LLM to interleave its regular output text with explicit symbolic references to fields present in some conditioning data (e.g., a table in JSON format). The references can be used to display the provenance of different spans of text in the generation, reducing the effort required for manual verification. Across data-to-text and question answering experiments, we find that LLMs are able to directly output text that makes use of symbolic references while maintaining fluency and accuracy.
Authors: Yifan Wu, Pengchuan Zhang, Wenhan Xiong, Barlas Oguz, James C. Gee, Yixin Nie
The study explores the effectiveness of the Chain-of-Thought approach, known for its proficiency in language tasks by breaking them down into sub-tasks and intermediate steps, in improving vision-language tasks that demand sophisticated perception and reasoning. We present the "Description then Decision" strategy, which is inspired by how humans process signals. This strategy significantly improves probing task performance by 50%, establishing the groundwork for future research on reasoning paradigms in complex vision-language tasks.
Authors: Junqing He, Kunhao Pan, Xiaoqun Dong, Zhuoyang Song, Yibo Liu, Yuxin Liang, Hao Wang, Qianguo Sun, Songxin Zhang, Zejian Xie, Jiaxing Zhang
While large language models (LLMs) are equipped with longer text input capabilities than before, they are struggling to seek correct information in long contexts. The "lost in the middle" problem challenges most LLMs, referring to the dramatic decline in accuracy when correct information is located in the middle. To overcome this crucial issue, this paper proposes to enhance the information searching and reflection ability of LLMs in long contexts via specially designed tasks called Attention Strengthening Multi-doc QA (ASM QA). Following these tasks, our model excels in focusing more precisely on the desired information. Experimental results show substantial improvement in Multi-doc QA and other benchmarks, superior to state-of-the-art models by 13.7% absolute gain in shuffled settings, by 21.5% in passage retrieval task. We release our model, Ziya-Reader to promote related research in the community.
Authors: Qianren Mao, Jiazheng Wang, Zheng Wang, Xi Li, Bo Li, Jianxin Li
Timeline summarization (TLS) involves creating summaries of long-running events using dated summaries from numerous news articles. However, limited data availability has significantly slowed down the development of timeline summarization. In this paper, we introduce the CNTLS dataset, a versatile resource for Chinese timeline summarization. CNTLS encompasses 77 real-life topics, each with 2524 documents and summarizes nearly 60\% days duration compression on average all topics.
We meticulously analyze the corpus using well-known metrics, focusing on the style of the summaries and the complexity of the summarization task. Specifically, we evaluate the performance of various extractive and generative summarization systems on the CNTLS corpus to provide benchmarks and support further research. To the best of our knowledge, CNTLS is the first Chinese timeline summarization dataset. The dataset and source code are released\footnote{Code and data available at: \emph{\url{https://github.com/OpenSUM/CNTLS}}.}.
Authors: Sai Srivatsa Ravindranath, Zhe Feng, Shira Li, Jonathan Ma, Scott D. Kominers, David C. Parkes
We initiate the study of deep learning for the automated design of two-sided matching mechanisms. What is of most interest is to use machine learning to understand the possibility of new tradeoffs between strategy-proofness and stability. These properties cannot be achieved simultaneously, but the efficient frontier is not understood. We introduce novel differentiable surrogates for quantifying ordinal strategy-proofness and stability and use them to train differentiable matching mechanisms that map discrete preferences to valid randomized matchings. We demonstrate that the efficient frontier characterized by these learned mechanisms is substantially better than that achievable through a convex combination of baselines of deferred acceptance (stable and strategy-proof for only one side of the market), top trading cycles (strategy-proof for one side, but not stable), and randomized serial dictatorship (strategy-proof for both sides, but not stable). This gives a new target for economic theory and opens up new possibilities for machine learning pipelines in matching market design.
Authors: Irene Tenison, Sai Aravind Sreeramadas, Vaikkunth Mugunthan, Edouard Oyallon, Irina Rish, Eugene Belilovsky
Federated learning (FL) is an emerging paradigm that permits a large number of clients with heterogeneous data to coordinate learning of a unified global model without the need to share data amongst each other. A major challenge in federated learning is the heterogeneity of data across client, which can degrade the performance of standard FL algorithms. Standard FL algorithms involve averaging of model parameters or gradient updates to approximate the global model at the server. However, we argue that in heterogeneous settings, averaging can result in information loss and lead to poor generalization due to the bias induced by dominant client gradients. We hypothesize that to generalize better across non-i.i.d datasets, the algorithms should focus on learning the invariant mechanism that is constant while ignoring spurious mechanisms that differ across clients. Inspired from recent works in Out-of-Distribution generalization, we propose a gradient masked averaging approach for FL as an alternative to the standard averaging of client updates. This aggregation technique for client updates can be adapted as a drop-in replacement in most existing federated algorithms. We perform extensive experiments on multiple FL algorithms with in-distribution, real-world, feature-skewed out-of-distribution, and quantity imbalanced datasets and show that it provides consistent improvements, particularly in the case of heterogeneous clients.
Authors: Yuchen Lu, Zhen Liu, Aristide Baratin, Romain Laroche, Aaron Courville, Alessandro Sordoni
We address the problem of evaluating the quality of self-supervised learning (SSL) models without access to supervised labels, while being agnostic to the architecture, learning algorithm or data manipulation used during training. We argue that representations can be evaluated through the lens of expressiveness and learnability. We propose to use the Intrinsic Dimension (ID) to assess expressiveness and introduce Cluster Learnability (CL) to assess learnability. CL is measured in terms of the performance of a KNN classifier trained to predict labels obtained by clustering the representations with K-means. We thus combine CL and ID into a single predictor -- CLID. Through a large-scale empirical study with a diverse family of SSL algorithms, we find that CLID better correlates with in-distribution model performance than other competing recent evaluation schemes. We also benchmark CLID on out-of-domain generalization, where CLID serves as a predictor of the transfer performance of SSL models on several visual classification tasks, yielding improvements with respect to the competing baselines.
Authors: Daphne Wang (University College London), Mehrnoosh Sadrzadeh (University College London)
Ambiguity is a natural language phenomenon occurring at different levels of syntax, semantics, and pragmatics. It is widely studied; in Psycholinguistics, for instance, we have a variety of competing studies for the human disambiguation processes. These studies are empirical and based on eye-tracking measurements. Here we take first steps towards formalizing these processes for semantic ambiguities where we identified the presence of two features: (1) joint plausibility degrees of different possible interpretations, (2) causal structures according to which certain words play a more substantial role in the processes. The novel sheaf-theoretic model of definite causality developed by Gogioso and Pinzani in QPL 2021 offers tools to model and reason about these features. We applied this theory to a dataset of ambiguous phrases extracted from Psycholinguistics literature and their human plausibility judgements collected by us using the Amazon Mechanical Turk engine. We measured the causal fractions of different disambiguation orders within the phrases and discovered two prominent orders: from subject to verb in the subject-verb and from object to verb in the verb object phrases. We also found evidence for delay in the disambiguation of polysemous vs homonymous verbs, again compatible with Psycholinguistic findings.
Authors: Rūta Binkytė, Ljupcho Grozdanovski, Sami Zhioua
Besides its common use cases in epidemiology, political, and social sciences, causality turns out to be crucial in evaluating the fairness of automated decisions, both in a legal and everyday sense. We provide arguments and examples, of why causality is particularly important for fairness evaluation. In particular, we point out the social impact of non-causal predictions and the legal anti-discrimination process that relies on causal claims. We conclude with a discussion about the challenges and limitations of applying causality in practical scenarios as well as possible solutions.
Authors: Thomas J. Ringstrom
Reinforcement Learning views the maximization of rewards and avoidance of punishments as central to explaining goal-directed behavior. However, over a life, organisms will need to learn about many different aspects of the world's structure: the states of the world and state-vector transition dynamics. The number of combinations of states grows exponentially as an agent incorporates new knowledge, and there is no obvious weighted combination of pre-existing rewards or costs defined for a given combination of states, as such a weighting would need to encode information about good and bad combinations prior to an agent's experience in the world. Therefore, we must develop more naturalistic accounts of behavior and motivation in large state-spaces. We show that it is possible to use only the intrinsic motivation metric of empowerment, which measures the agent's capacity to realize many possible futures under a transition operator. We propose to scale empowerment to hierarchical state-spaces by using Operator Bellman Equations. These equations produce state-time feasibility functions, which are compositional hierarchical state-time transition operators that map an initial state and time when an agent begins a policy to the final states and times of completing a goal. Because these functions are hierarchical operators we can define hierarchical empowerment measures on them. An agent can then optimize plans to distant states and times to maximize its hierarchical empowerment-gain, allowing it to discover goals that bring about a more favorable coupling of its internal structure (physiological states) to its external environment (world structure & spatial state). Life-long agents could therefore be primarily animated by principles of compositionality and empowerment, exhibiting self-concern for the growth & maintenance of their own structural integrity without recourse to reward-maximization.
Authors: Matthew Shardlow, Piotr Przybyła
This work is intended as a voice in the discussion over previous claims that a pretrained large language model (LLM) based on the Transformer model architecture can be sentient. Such claims have been made concerning the LaMDA model and also concerning the current wave of LLM-powered chatbots, such as ChatGPT. This claim, if confirmed, would have serious ramifications in the Natural Language Processing (NLP) community due to wide-spread use of similar models. However, here we take the position that such a large language model cannot be sentient, or conscious, and that LaMDA in particular exhibits no advances over other similar models that would qualify it. We justify this by analysing the Transformer architecture through Integrated Information Theory of consciousness. We see the claims of sentience as part of a wider tendency to use anthropomorphic language in NLP reporting. Regardless of the veracity of the claims, we consider this an opportune moment to take stock of progress in language modelling and consider the ethical implications of the task. In order to make this work helpful for readers outside the NLP community, we also present the necessary background in language modelling.
Authors: Siyuan Cheng, Ningyu Zhang, Bozhong Tian, Xi Chen, Qingbing Liu, Huajun Chen
Recently decades have witnessed the empirical success of framing Knowledge Graph (KG) embeddings via language models. However, language model-based KG embeddings are usually deployed as static artifacts, making them difficult to modify post-deployment without re-training after deployment. To address this issue, we propose a new task of editing language model-based KG embeddings in this paper. This task is designed to facilitate rapid, data-efficient updates to KG embeddings without compromising the performance of other aspects. We build four new datasets: E-FB15k237, A-FB15k237, E-WN18RR, and A-WN18RR, and evaluate several knowledge editing baselines demonstrating the limited ability of previous models to handle the proposed challenging task. We further propose a simple yet strong baseline dubbed KGEditor, which utilizes additional parametric layers of the hyper network to edit/add facts. Our comprehensive experimental results reveal that KGEditor excels in updating specific facts without impacting the overall performance, even when faced with limited training resources. Code and datasets are available in https://github.com/zjunlp/PromptKG/tree/main/deltaKG.
Authors: A. Feder Cooper, Katherine Lee, Madiha Choksi, Solon Barocas, Christopher De Sa, James Grimmelmann, Jon Kleinberg, Siddhartha Sen, Baobao Zhang
Variance in predictions across different trained models is a significant, under-explored source of error in fair binary classification. In practice, the variance on some data examples is so large that decisions can be effectively arbitrary. To investigate this problem, we take an experimental approach and make four overarching contributions: We: 1) Define a metric called self-consistency, derived from variance, which we use as a proxy for measuring and reducing arbitrariness; 2) Develop an ensembling algorithm that abstains from classification when a prediction would be arbitrary; 3) Conduct the largest to-date empirical study of the role of variance (vis-a-vis self-consistency and arbitrariness) in fair binary classification; and, 4) Release a toolkit that makes the US Home Mortgage Disclosure Act (HMDA) datasets easily usable for future research. Altogether, our experiments reveal shocking insights about the reliability of conclusions on benchmark datasets. Most fair binary classification benchmarks are close-to-fair when taking into account the amount of arbitrariness present in predictions -- before we even try to apply any fairness interventions. This finding calls into question the practical utility of common algorithmic fairness methods, and in turn suggests that we should reconsider how we choose to measure fairness in binary classification.
Authors: Stamatis Alexandropoulos, Christos Sakaridis, Petros Maragos
Semantic segmentation is a fundamental task in visual scene understanding. We focus on the supervised setting, where ground-truth semantic annotations are available. Based on knowledge about the high regularity of real-world scenes, we propose a method for improving class predictions by learning to selectively exploit information from neighboring pixels. In particular, our method is based on the prior that for each pixel, there is a seed pixel in its close neighborhood sharing the same prediction with the former. Motivated by this prior, we design a novel two-head network, named Offset Vector Network (OVeNet), which generates both standard semantic predictions and a dense 2D offset vector field indicating the offset from each pixel to the respective seed pixel, which is used to compute an alternative, seed-based semantic prediction. The two predictions are adaptively fused at each pixel using a learnt dense confidence map for the predicted offset vector field. We supervise offset vectors indirectly via optimizing the seed-based prediction and via a novel loss on the confidence map. Compared to the baseline state-of-the-art architectures HRNet and HRNet+OCR on which OVeNet is built, the latter achieves significant performance gains on three prominent benchmarks for semantic segmentation, namely Cityscapes, ACDC and ADE20K. Code is available at https://github.com/stamatisalex/OVeNet
Authors: Hongbin Ye, Honghao Gui, Xin Xu, Xi Chen, Huajun Chen, Ningyu Zhang
Conventional Knowledge Graph Construction (KGC) approaches typically follow the static information extraction paradigm with a closed set of pre-defined schema. As a result, such approaches fall short when applied to dynamic scenarios or domains, whereas a new type of knowledge emerges. This necessitates a system that can handle evolving schema automatically to extract information for KGC. To address this need, we propose a new task called schema-adaptable KGC, which aims to continually extract entity, relation, and event based on a dynamically changing schema graph without re-training. We first split and convert existing datasets based on three principles to build a benchmark, i.e., horizontal schema expansion, vertical schema expansion, and hybrid schema expansion; then investigate the schema-adaptable performance of several well-known approaches such as Text2Event, TANL, UIE and GPT-3.5. We further propose a simple yet effective baseline dubbed \textsc{AdaKGC}, which contains schema-enriched prefix instructor and schema-conditioned dynamic decoding to better handle evolving schema. Comprehensive experimental results illustrate that AdaKGC can outperform baselines but still have room for improvement. We hope the proposed work can deliver benefits to the community. Code and datasets available at https://github.com/zjunlp/AdaKGC.
Authors: Chandan Singh, Aliyah R. Hsu, Richard Antonello, Shailee Jain, Alexander G. Huth, Bin Yu, Jianfeng Gao
Large language models (LLMs) have demonstrated remarkable prediction performance for a growing array of tasks. However, their rapid proliferation and increasing opaqueness have created a growing need for interpretability. Here, we ask whether we can automatically obtain natural language explanations for black box text modules. A "text module" is any function that maps text to a scalar continuous value, such as a submodule within an LLM or a fitted model of a brain region. "Black box" indicates that we only have access to the module's inputs/outputs.
We introduce Summarize and Score (SASC), a method that takes in a text module and returns a natural language explanation of the module's selectivity along with a score for how reliable the explanation is. We study SASC in 3 contexts. First, we evaluate SASC on synthetic modules and find that it often recovers ground truth explanations. Second, we use SASC to explain modules found within a pre-trained BERT model, enabling inspection of the model's internals. Finally, we show that SASC can generate explanations for the response of individual fMRI voxels to language stimuli, with potential applications to fine-grained brain mapping. All code for using SASC and reproducing results is made available on Github.
Authors: Yuan Sui, Mengyu Zhou, Mingjie Zhou, Shi Han, Dongmei Zhang
Large language models (LLMs) are becoming attractive as few-shot reasoners to solve Natural Language (NL)-related tasks. However, there is still much to learn about how well LLMs understand structured data, such as tables. While it is true that tables can be used as inputs to LLMs with serialization, there is a lack of comprehensive studies examining whether LLMs can truly comprehend such data. In this paper, we try to understand this by designing a benchmark to evaluate the structural understanding capabilities (SUC) of LLMs. The benchmark we create includes seven tasks, each with its own unique challenges, \eg, cell lookup, row retrieval, and size detection. We conduct a series of evaluations on GPT-3.5 and GPT-4. We find that the performance varied depending on several input choices, including table input format, content order, role prompting, and partition marks. Drawing from the insights gained through the benchmark evaluations, we propose \textit{self-augmentation} for effective structural prompting, such as critical value / range identification using LLMs' internal knowledge. When combined with carefully chosen input choices, these structural prompting methods lead to promising improvements in LLM performance on a variety of tabular tasks, \eg, TabFact($\uparrow2.31\%$), HybridQA($\uparrow2.13\%$), SQA($\uparrow2.72\%$), Feverous($\uparrow0.84\%$), and ToTTo($\uparrow5.68\%$). We believe that our benchmark and proposed prompting methods can serve as a simple yet generic selection for future research.
Authors: Tarek Naous, Michael J. Ryan, Anton Lavrouk, Mohit Chandra, Wei Xu
We present a systematic study and comprehensive evaluation of large language models for automatic multilingual readability assessment. In particular, we construct ReadMe++, a multilingual multi-domain dataset with human annotations of 9757 sentences in Arabic, English, French, Hindi, and Russian collected from 112 different data sources. ReadMe++ offers more domain and language diversity than existing readability datasets, making it ideal for benchmarking multilingual and non-English language models (including mBERT, XLM-R, mT5, Llama-2, GPT-4, etc.) in the supervised, unsupervised, and few-shot prompting settings. Our experiments reveal that models fine-tuned on ReadMe++ outperform those trained on single-domain datasets, showcasing superior performance on multi-domain readability assessment and cross-lingual transfer capabilities. We also compare to traditional readability metrics (such as Flesch-Kincaid Grade Level and Open Source Metric for Measuring Arabic Narratives), as well as the state-of-the-art unsupervised metric RSRS (Martinc et al., 2021). We will make our data and code publicly available at: https://github.com/tareknaous/readme.
Authors: Jeremy R. Cole, Michael J.Q. Zhang, Daniel Gillick, Julian Martin Eisenschlos, Bhuwan Dhingra, Jacob Eisenstein
Trustworthy language models should abstain from answering questions when they do not know the answer. However, the answer to a question can be unknown for a variety of reasons. Prior research has focused on the case in which the question is clear and the answer is unambiguous but possibly unknown, but the answer to a question can also be unclear due to uncertainty of the questioner's intent or context. We investigate question answering from this perspective, focusing on answering a subset of questions with a high degree of accuracy, from a set of questions in which many are inherently ambiguous. In this setting, we find that the most reliable approach to decide when to abstain involves quantifying repetition within sampled model outputs, rather than the model's likelihood or self-verification as used in prior work. We find this to be the case across different types of uncertainty and model scales,and with or without instruction tuning. Our results suggest that sampling-based confidence scores help calibrate answers to relatively unambiguous questions, with more dramatic improvements on ambiguous questions.
Authors: Angeline Aguinaldo, Evan Patterson, James Fairbanks, William Regli, Jaime Ruiz
Classical planning representation languages based on first-order logic have preliminarily been used to model and solve robotic task planning problems. Wider adoption of these representation languages, however, is hindered by the limitations present when managing implicit world changes with concise action models. To address this problem, we propose an alternative approach to representing and managing updates to world states during planning. Based on the category-theoretic concepts of $\mathsf{C}$-sets and double-pushout rewriting (DPO), our proposed representation can effectively handle structured knowledge about world states that support domain abstractions at all levels. It formalizes the semantics of predicates according to a user-provided ontology and preserves the semantics when transitioning between world states. This method provides a formal semantics for using knowledge graphs and relational databases to model world states and updates in planning. In this paper, we conceptually compare our category-theoretic representation with the classical planning representation. We show that our proposed representation has advantages over the classical representation in terms of handling implicit preconditions and effects, and provides a more structured framework in which to model and solve planning problems.
Authors: Jonas Eschmann, Dario Albani, Giuseppe Loianno
Deep Reinforcement Learning (RL) has been demonstrated to yield capable agents and control policies in several domains but is commonly plagued by prohibitively long training times. Additionally, in the case of continuous control problems, the applicability of learned policies on real-world embedded devices is limited due to the lack of real-time guarantees and portability of existing deep learning libraries. To address these challenges, we present RLtools, a dependency-free, header-only, pure C++ library for deep supervised and reinforcement learning. Leveraging the template meta-programming capabilities of recent C++ standards, we provide composable components that can be tightly integrated by the compiler. Its novel architecture allows RLtools to be used seamlessly on a heterogeneous set of platforms, from HPC clusters over workstations and laptops to smartphones, smartwatches, and microcontrollers. Specifically, due to the tight integration of the RL algorithms with simulation environments, RLtools can solve popular RL problems like the Pendulum-v1 swing-up about 7 to 15 times faster in terms of wall-clock training time compared to other popular RL frameworks when using TD3. We also provide a low-overhead and parallelized interface to the MuJoCo simulator, showing that our PPO implementation achieves state of the art returns in the Ant-v4 environment while being 25%-30% faster in terms of wall-clock training time. Finally, we also benchmark the policy inference on a diverse set of microcontrollers and show that in most cases our optimized inference implementation is much faster than even the manufacturer's DSP libraries. To the best of our knowledge, RLtools enables the first-ever demonstration of training a deep RL algorithm directly on a microcontroller, giving rise to the field of TinyRL. The source code is available through our project page at https://rl.tools.
Authors: Maziar Gomrokchi, Owen Levin, Jeffrey Roach, Jonah White
We introduce AdCraft, a novel benchmark environment for the Reinforcement Learning (RL) community distinguished by its stochastic and non-stationary properties. The environment simulates bidding and budgeting dynamics within Search Engine Marketing (SEM), a digital marketing technique utilizing paid advertising to enhance the visibility of websites on search engine results pages (SERPs). The performance of SEM advertisement campaigns depends on several factors, including keyword selection, ad design, bid management, budget adjustments, and performance monitoring. Deep RL recently emerged as a potential strategy to optimize campaign profitability within the complex and dynamic landscape of SEM, but it requires substantial data, which may be costly or infeasible to acquire in practice. Our customizable environment enables practitioners to assess and enhance the robustness of RL algorithms pertinent to SEM bid and budget management without such costs. Through a series of experiments within the environment, we demonstrate the challenges imposed on agent convergence and performance by sparsity and non-stationarity. We hope these challenges further encourage discourse and development around effective strategies for managing real-world uncertainties.
Authors: Hao Wang, Xiwen Chen, Natan Vital, Edward.Duffy, Abolfazl Razi
With more than 32% of the global energy used by commercial and residential buildings, there is an urgent need to revisit traditional approaches to Building Energy Management (BEM). With HVAC systems accounting for about 40% of the total energy cost in the commercial sector, we propose a low-complexity DRL-based model with multi-input multi-output architecture for the HVAC energy optimization of open-plan offices, which uses only a handful of controllable and accessible factors. The efficacy of our solution is evaluated through extensive analysis of the overall energy consumption and thermal comfort levels compared to a baseline system based on the existing HVAC schedule in a real building. This comparison shows that our method achieves 37% savings in energy consumption with minimum violation (<1%) of the desired temperature range during work hours. It takes only a total of 40 minutes for 5 epochs (about 7.75 minutes per epoch) to train a network with superior performance and covering diverse conditions for its low-complexity architecture; therefore, it easily adapts to changes in the building setups, weather conditions, occupancy rate, etc. Moreover, by enforcing smoothness on the control strategy, we suppress the frequent and unpleasant on/off transitions on HVAC units to avoid occupant discomfort and potential damage to the system. The generalizability of our model is verified by applying it to different building models and under various weather conditions.
Authors: Kavosh Asadi, Rasool Fakoor, Shoham Sabach
We focus on the task of approximating the optimal value function in deep reinforcement learning. This iterative process is comprised of solving a sequence of optimization problems where the loss function changes per iteration. The common approach to solving this sequence of problems is to employ modern variants of the stochastic gradient descent algorithm such as Adam. These optimizers maintain their own internal parameters such as estimates of the first-order and the second-order moments of the gradient, and update them over time. Therefore, information obtained in previous iterations is used to solve the optimization problem in the current iteration. We demonstrate that this can contaminate the moment estimates because the optimization landscape can change arbitrarily from one iteration to the next one. To hedge against this negative effect, a simple idea is to reset the internal parameters of the optimizer when starting a new iteration. We empirically investigate this resetting idea by employing various optimizers in conjunction with the Rainbow algorithm. We demonstrate that this simple modification significantly improves the performance of deep RL on the Atari benchmark.
Authors: Jie Huang, Kevin Chen-Chuan Chang
Large Language Models (LLMs) bring transformative benefits alongside unique challenges, including intellectual property (IP) and ethical concerns. This position paper explores a novel angle to mitigate these risks, drawing parallels between LLMs and established web systems. We identify "citation" - the acknowledgement or reference to a source or evidence - as a crucial yet missing component in LLMs. Incorporating citation could enhance content transparency and verifiability, thereby confronting the IP and ethical issues in the deployment of LLMs. We further propose that a comprehensive citation mechanism for LLMs should account for both non-parametric and parametric content. Despite the complexity of implementing such a citation mechanism, along with the potential pitfalls, we advocate for its development. Building on this foundation, we outline several research problems in this area, aiming to guide future explorations towards building more responsible and accountable LLMs.
Authors: Romain Hardy, Joe Klepich, Ryan Mitchell, Steve Hall, Jericho Villareal, Cornelia Ilin
Integrating deep learning with clinical expertise holds great potential for addressing healthcare challenges and empowering medical professionals with improved diagnostic tools. However, the need for annotated medical images is often an obstacle to leveraging the full power of machine learning models. Our research demonstrates that by combining synthetic images, generated using diffusion models, with real images, we can enhance nonalcoholic fatty liver disease (NAFLD) classification performance even in low-data regime settings. We evaluate the quality of the synthetic images by comparing two metrics: Inception Score (IS) and Fr\'{e}chet Inception Distance (FID), computed on diffusion- and generative adversarial network (GAN)-generated images. Our results show superior performance for the diffusion-generated images, with a maximum IS score of $1.90$ compared to $1.67$ for GANs, and a minimum FID score of $69.45$ compared to $100.05$ for GANs. Utilizing a partially frozen CNN backbone (EfficientNet v1), our synthetic augmentation method achieves a maximum image-level ROC AUC of $0.904$ on a NAFLD prediction task.
Authors: Sreyan Ghosh, Chandra Kiran Reddy Evuru, Sonal Kumar, Utkarsh Tyagi, Sakshi Singh, Sanjoy Chowdhury, Dinesh Manocha
Neural image classifiers can often learn to make predictions by overly relying on non-predictive features that are spuriously correlated with the class labels in the training data. This leads to poor performance in real-world atypical scenarios where such features are absent. Supplementing the training dataset with images without such spurious features can aid robust learning against spurious correlations via better generalization. This paper presents ASPIRE (Language-guided data Augmentation for SPurIous correlation REmoval), a simple yet effective solution for expanding the training dataset with synthetic images without spurious features. ASPIRE, guided by language, generates these images without requiring any form of additional supervision or existing examples. Precisely, we employ LLMs to first extract foreground and background features from textual descriptions of an image, followed by advanced language-guided image editing to discover the features that are spuriously correlated with the class label. Finally, we personalize a text-to-image generation model to generate diverse in-domain images without spurious features. We demonstrate the effectiveness of ASPIRE on 4 datasets, including the very challenging Hard ImageNet dataset, and 9 baselines and show that ASPIRE improves the classification accuracy of prior methods by 1% - 38%. Code soon at: https://github.com/Sreyan88/ASPIRE.
Authors: Shirui Chen, Linxing Preston Jiang, Rajesh P. N. Rao, Eric Shea-Brown
In sampling-based Bayesian models of brain function, neural activities are assumed to be samples from probability distributions that the brain uses for probabilistic computation. However, a comprehensive understanding of how mechanistic models of neural dynamics can sample from arbitrary distributions is still lacking. We use tools from functional analysis and stochastic differential equations to explore the minimum architectural requirements for $\textit{recurrent}$ neural circuits to sample from complex distributions. We first consider the traditional sampling model consisting of a network of neurons whose outputs directly represent the samples (sampler-only network). We argue that synaptic current and firing-rate dynamics in the traditional model have limited capacity to sample from a complex probability distribution. We show that the firing rate dynamics of a recurrent neural circuit with a separate set of output units can sample from an arbitrary probability distribution. We call such circuits reservoir-sampler networks (RSNs). We propose an efficient training procedure based on denoising score matching that finds recurrent and output weights such that the RSN implements Langevin sampling. We empirically demonstrate our model's ability to sample from several complex data distributions using the proposed neural dynamics and discuss its applicability to developing the next generation of sampling-based brain models.
Authors: Zhen Bi, Ningyu Zhang, Yinuo Jiang, Shumin Deng, Guozhou Zheng, Huajun Chen
In the realm of embodied artificial intelligence, the reasoning capabilities of Large Language Models (LLMs) play a pivotal role. Although there are effective methods like program-of-thought prompting for LLMs which uses programming language to tackle complex reasoning tasks, the specific impact of code data on the improvement of reasoning capabilities remains under-explored. To address this gap, we propose complexity-impacted reasoning score (CIRS), which combines structural and logical attributes, to measure the correlation between code and reasoning abilities. Specifically, we use the abstract syntax tree to encode the structural information and calculate logical complexity by considering the difficulty and the cyclomatic complexity. Through an empirical analysis, we find not all code data of complexity can be learned or understood by LLMs. Optimal level of complexity is critical to the improvement of reasoning abilities by program-aided prompting. Then we design an auto-synthesizing and stratifying algorithm, and apply it to instruction generation for mathematical reasoning and code data filtering for code generation tasks. Extensive results demonstrates the effectiveness of our proposed approach. Code will be integrated into the EasyInstruct framework at https://github.com/zjunlp/EasyInstruct.
Authors: Yashuai Yan, Esteve Valls Mascaro, Dongheui Lee
This paper introduces a novel deep-learning approach for human-to-robot motion retargeting, enabling robots to mimic human poses accurately. Contrary to prior deep-learning-based works, our method does not require paired human-to-robot data, which facilitates its translation to new robots. First, we construct a shared latent space between humans and robots via adaptive contrastive learning that takes advantage of a proposed cross-domain similarity metric between the human and robot poses. Additionally, we propose a consistency term to build a common latent space that captures the similarity of the poses with precision while allowing direct robot motion control from the latent space. For instance, we can generate in-between motion through simple linear interpolation between two projected human poses. We conduct a comprehensive evaluation of robot control from diverse modalities (i.e., texts, RGB videos, and key poses), which facilitates robot control for non-expert users. Our model outperforms existing works regarding human-to-robot retargeting in terms of efficiency and precision. Finally, we implemented our method in a real robot with self-collision avoidance through a whole-body controller to showcase the effectiveness of our approach. More information on our website https://evm7.github.io/UnsH2R/
Authors: Tianhao Wu, Mingdong Wu, Jiyao Zhang, Yunchong Gan, Hao Dong
The use of anthropomorphic robotic hands for assisting individuals in situations where human hands may be unavailable or unsuitable has gained significant importance. In this paper, we propose a novel task called human-assisting dexterous grasping that aims to train a policy for controlling a robotic hand's fingers to assist users in grasping objects. Unlike conventional dexterous grasping, this task presents a more complex challenge as the policy needs to adapt to diverse user intentions, in addition to the object's geometry. We address this challenge by proposing an approach consisting of two sub-modules: a hand-object-conditional grasping primitive called Grasping Gradient Field~(GraspGF), and a history-conditional residual policy. GraspGF learns `how' to grasp by estimating the gradient from a success grasping example set, while the residual policy determines `when' and at what speed the grasping action should be executed based on the trajectory history. Experimental results demonstrate the superiority of our proposed method compared to baselines, highlighting the user-awareness and practicality in real-world applications. The codes and demonstrations can be viewed at "https://sites.google.com/view/graspgf".
Authors: Mohamad M.A. Ashames, Ahmet Demir, Omer N. Gerek, Mehmet Fidan, M. Bilginer Gulmezoglu, Semih Ergin, Mehmet Koc, Atalay Barkana, Cuneyt Calisir
Following the great success of various deep learning methods in image and object classification, the biomedical image processing society is also overwhelmed with their applications to various automatic diagnosis cases. Unfortunately, most of the deep learning-based classification attempts in the literature solely focus on the aim of extreme accuracy scores, without considering interpretability, or patient-wise separation of training and test data. For example, most lung nodule classification papers using deep learning randomly shuffle data and split it into training, validation, and test sets, causing certain images from the CT scan of a person to be in the training set, while other images of the exact same person to be in the validation or testing image sets. This can result in reporting misleading accuracy rates and the learning of irrelevant features, ultimately reducing the real-life usability of these models. When the deep neural networks trained on the traditional, unfair data shuffling method are challenged with new patient images, it is observed that the trained models perform poorly. In contrast, deep neural networks trained with strict patient-level separation maintain their accuracy rates even when new patient images are tested. Heat-map visualizations of the activations of the deep neural networks trained with strict patient-level separation indicate a higher degree of focus on the relevant nodules. We argue that the research question posed in the title has a positive answer only if the deep neural networks are trained with images of patients that are strictly isolated from the validation and testing patient sets.
Authors: Pei Xu, Kaixiang Xie, Sheldon Andrews, Paul G. Kry, Michael Neff, Morgan McGuire, Ioannis Karamouzas, Victor Zordan
Motivated by humans' ability to adapt skills in the learning of new ones, this paper presents AdaptNet, an approach for modifying the latent space of existing policies to allow new behaviors to be quickly learned from like tasks in comparison to learning from scratch. Building on top of a given reinforcement learning controller, AdaptNet uses a two-tier hierarchy that augments the original state embedding to support modest changes in a behavior and further modifies the policy network layers to make more substantive changes. The technique is shown to be effective for adapting existing physics-based controllers to a wide range of new styles for locomotion, new task targets, changes in character morphology and extensive changes in environment. Furthermore, it exhibits significant increase in learning efficiency, as indicated by greatly reduced training times when compared to training from scratch or using other approaches that modify existing policies. Code is available at https://motion-lab.github.io/AdaptNet.
Authors: Jian Liu, Li Shen, Guo-Wei Wei
ChatGPT represents a significant milestone in the field of artificial intelligence (AI), finding widespread applications across diverse domains. However, its effectiveness in mathematical contexts has been somewhat constrained by its susceptibility to conceptual errors. Concurrently, topological data analysis (TDA), a relatively new discipline, has garnered substantial interest in recent years. Nonetheless, the advancement of TDA is impeded by the limited understanding of computational algorithms and coding proficiency among theoreticians. This work endeavors to bridge the gap between theoretical topological concepts and their practical implementation in computational topology through the utilization of ChatGPT. We showcase how a pure theoretician, devoid of computational experience and coding skills, can effectively transform mathematical formulations and concepts into functional code for computational topology with the assistance of ChatGPT. Our strategy outlines a productive process wherein a mathematician trains ChatGPT on pure mathematical concepts, steers ChatGPT towards generating computational topology code, and subsequently validates the generated code using established examples. Our specific case studies encompass the computation of Betti numbers, Laplacian matrices, and Dirac matrices for simplicial complexes, as well as the persistence of various homologies and Laplacians. Furthermore, we explore the application of ChatGPT in computing recently developed topological theories for hypergraphs and digraphs. This work serves as an initial step towards effectively transforming pure mathematical theories into practical computational tools, with the ultimate goal of enabling real applications across diverse fields.
Authors: Ti-Rong Wu, Hung Guei, Po-Wei Huang, Pei-Chiun Peng, Ting Han Wei, Chung-Chin Shih, Yun-Jui Tsai
This paper presents MiniZero, a zero-knowledge learning framework that supports four state-of-the-art algorithms, including AlphaZero, MuZero, Gumbel AlphaZero, and Gumbel MuZero. While these algorithms have demonstrated super-human performance in many games, it remains unclear which among them is most suitable or efficient for specific tasks. Through MiniZero, we systematically evaluate the performance of each algorithm in two board games, 9x9 Go and 8x8 Othello, as well as 57 Atari games. For two board games, using more simulations generally results in higher performance. However, the choice of AlphaZero and MuZero may differ based on game properties. For Atari games, both MuZero and Gumbel MuZero are worth considering. Since each game has unique characteristics, different algorithms and simulations yield varying results. In addition, we introduce an approach, called progressive simulation, which progressively increases the simulation budget during training to allocate computation more efficiently. Our empirical results demonstrate that progressive simulation achieves significantly superior performance in two board games. By making our framework and trained models publicly available, this paper contributes a benchmark for future research on zero-knowledge learning algorithms, assisting researchers in algorithm selection and comparison against these zero-knowledge learning baselines. Our code and data are available at https://rlg.iis.sinica.edu.tw/papers/minizero.
Authors: Yongqi Tong, Yifan Wang, Dawei Li, Sizhe Wang, Zi Lin, Simeng Han, Jingbo Shang
Chain-of-Thought(CoT) prompting and its variants explore equipping large language models (LLMs) with high-level reasoning abilities by emulating human-like linear cognition and logic. However, the human mind is complicated and mixed with both linear and nonlinear thinking. In this work, we propose \textbf{I}nferential \textbf{E}xclusion \textbf{P}rompting (IEP), a novel prompting that combines the principles of elimination and inference in order to guide LLMs to think non-linearly. IEP guides LLMs to plan and then utilize Natural Language Inference (NLI) to deduce each possible solution's entailment relation with context, commonsense, or facts, therefore yielding a broader perspective by thinking back for inferring. This forward planning and backward eliminating process allows IEP to better simulate the complex human thinking processes compared to other CoT-based methods, which only reflect linear cognitive processes. We conducted a series of empirical studies and have corroborated that IEP consistently outperforms CoT across various tasks. Additionally, we observe that integrating IEP and CoT further improves the LLMs' performance on certain tasks, highlighting the necessity of equipping LLMs with mixed logic processes. Moreover, to better evaluate comprehensive features inherent in human logic, we introduce \textbf{M}ental-\textbf{A}bility \textbf{R}easoning \textbf{B}enchmark (MARB). The benchmark comprises six novel subtasks with a total of 9,115 questions, among which 1,685 are developed with hand-crafted rationale references. We believe both \textsc{IEP} and \textsc{MARB} can serve as a promising direction for unveiling LLMs' logic and verbal reasoning abilities and drive further advancements. \textsc{MARB} will be available at ~\texttt{anonymity link} soon.
Authors: Aman Madaan, Pranjal Aggarwal, Ankit Anand, Srividya Pranavi Potharaju, Swaroop Mishra, Pei Zhou, Aditya Gupta, Dheeraj Rajagopal, Karthik Kappaganthu, Yiming Yang, Shyam Upadhyay, Mausam, Manaal Faruqui
Large language models (LLMs) are now available in various sizes and configurations from cloud API providers. While this diversity offers a broad spectrum of choices, effectively leveraging the options to optimize computational cost and performance remains challenging. In this work, we present AutoMix, an approach that strategically routes queries to larger LMs, based on the approximate correctness of outputs from a smaller LM. Central to AutoMix is a few-shot self-verification mechanism, which estimates the reliability of its own outputs without requiring training. Given that verifications can be noisy, we employ a meta verifier in AutoMix to refine the accuracy of these assessments. Our experiments using LLAMA2-13/70B, on five context-grounded reasoning datasets demonstrate that AutoMix surpasses established baselines, improving the incremental benefit per cost by up to 89%. Our code and data are available at https://github.com/automix-llm/automix.
Authors: Luca Deck, Jakob Schoeffer, Maria De-Arteaga, Niklas Kühl
In this critical survey, we analyze typical claims on the relationship between explainable AI (XAI) and fairness to disentangle the multidimensional relationship between these two concepts. Based on a systematic literature review and a subsequent qualitative content analysis, we identify seven archetypal claims from 175 papers on the alleged fairness benefits of XAI. We present crucial caveats with respect to these claims and provide an entry point for future discussions around the potentials and limitations of XAI for specific fairness desiderata. While the literature often suggests XAI to be an enabler for several fairness desiderata, we notice a divide between these desiderata and the capabilities of XAI. We encourage to conceive XAI as one of many tools to approach the multidimensional, sociotechnical challenge of algorithmic fairness and to be more specific about how exactly what kind of XAI method enables whom to address which fairness desideratum.
Authors: Illia Horenko
Simply-verifiable mathematical conditions for existence, uniqueness and explicit analytical computation of minimal adversarial paths (MAP) and minimal adversarial distances (MAD) for (locally) uniquely-invertible classifiers, for generalized linear models (GLM), and for entropic AI (EAI) are formulated and proven. Practical computation of MAP and MAD, their comparison and interpretations for various classes of AI tools (for neuronal networks, boosted random forests, GLM and EAI) are demonstrated on the common synthetic benchmarks: on a double Swiss roll spiral and its extensions, as well as on the two biomedical data problems (for the health insurance claim predictions, and for the heart attack lethality classification). On biomedical applications it is demonstrated how MAP provides unique minimal patient-specific risk-mitigating interventions in the predefined subsets of accessible control variables.
Authors: Yuan Peiwen, Zhu Changsheng
There has been an emergence of various models for long-term time series forecasting. Recent studies have demonstrated that a single linear layer, using Channel Dependent (CD) or Channel Independent (CI) modeling, can even outperform a large number of sophisticated models. However, current research primarily considers CD and CI as two complementary yet mutually exclusive approaches, unable to harness these two extremes simultaneously. And it is also a challenging issue that both CD and CI are static strategies that cannot be determined to be optimal for a specific dataset without extensive experiments. In this paper, we reconsider whether the current CI strategy is the best solution for time series forecasting. First, we propose a simple yet effective strategy called CSC, which stands for $\mathbf{C}$hannel $\mathbf{S}$elf-$\mathbf{C}$lustering strategy, for linear models. Our Channel Self-Clustering (CSC) enhances CI strategy's performance improvements while reducing parameter size, for exmpale by over 10 times on electricity dataset, and significantly cutting training time. Second, we further propose Channel Rearrangement (CR), a method for deep models inspired by the self-clustering. CR attains competitive performance against baselines. Finally, we also discuss whether it is best to forecast the future values using the historical values of the same channel as inputs. We hope our findings and methods could inspire new solutions beyond CD/CI.
Authors: Yuxiao Chen, Sushant Veer, Peter Karkus, Marco Pavone
In highly interactive driving scenarios, the actions of one agent greatly influences those of its neighbors. Planning safe motions for autonomous vehicles in such interactive environments, therefore, requires reasoning about the impact of the ego's intended motion plan on nearby agents' behavior. Deep-learning-based models have recently achieved great success in trajectory prediction and many models in the literature allow for ego-conditioned prediction. However, leveraging ego-conditioned prediction remains challenging in downstream planning due to the complex nature of neural networks, limiting the planner structure to simple ones, e.g., sampling-based planner. Despite their ability to generate fine-grained high-quality motion plans, it is difficult for gradient-based planning algorithms, such as model predictive control (MPC), to leverage ego-conditioned prediction due to their iterative nature and need for gradient. We present Interactive Joint Planning (IJP) that bridges MPC with learned prediction models in a computationally scalable manner to provide us the best of both the worlds. In particular, IJP jointly optimizes over the behavior of the ego and the surrounding agents and leverages deep-learned prediction models as prediction priors that the join trajectory optimization tries to stay close to. Furthermore, by leveraging homotopy classes, our joint optimizer searches over diverse motion plans to avoid getting stuck at local minima. Closed-loop simulation result shows that IJP significantly outperforms the baselines that are either without joint optimization or running sampling-based planning.
Authors: Hiroki Takizawa
The game of Othello is one of the world's most complex and popular games that has yet to be computationally solved. Othello has roughly ten octodecillion (10 to the 58th power) possible game records and ten octillion (10 to the 28th power) possible game positions. The challenge of solving Othello, determining the outcome of a game with no mistake made by either player, has long been a grand challenge in computer science. This paper announces a significant milestone: Othello is now solved. It is computationally proved that perfect play by both players lead to a draw. Strong Othello software has long been built using heuristically designed search techniques. Solving a game provides a solution that enables the software to play the game perfectly.
Authors: Rūta Binkytė, Carlos Pinzón, Szilvia Lestyán, Kangsoo Jung, Héber H. Arcolezi, Catuscia Palamidessi
Differential privacy is a widely adopted framework designed to safeguard the sensitive information of data providers within a data set. It is based on the application of controlled noise at the interface between the server that stores and processes the data, and the data consumers. Local differential privacy is a variant that allows data providers to apply the privatization mechanism themselves on their data individually. Therefore it provides protection also in contexts in which the server, or even the data collector, cannot be trusted. The introduction of noise, however, inevitably affects the utility of the data, particularly by distorting the correlations between individual data components. This distortion can prove detrimental to tasks such as causal discovery. In this paper, we consider various well-known locally differentially private mechanisms and compare the trade-off between the privacy they provide, and the accuracy of the causal structure produced by algorithms for causal learning when applied to data obfuscated by these mechanisms. Our analysis yields valuable insights for selecting appropriate local differentially private protocols for causal discovery tasks. We foresee that our findings will aid researchers and practitioners in conducting locally private causal discovery.
Authors: Byeong-Hoo Lee, Byoung-Hee Kwon, Seong-Whan Lee
Deep learning has shown promise in decoding brain signals, such as electroencephalogram (EEG), in the field of brain-computer interfaces (BCIs). However, the non-stationary characteristics of EEG signals pose challenges for training neural networks to acquire appropriate knowledge. Inconsistent EEG signals resulting from these non-stationary characteristics can lead to poor performance. Therefore, it is crucial to investigate and address sample inconsistency to ensure robust performance in spontaneous BCIs. In this study, we introduce the concept of sample dominance as a measure of EEG signal inconsistency and propose a method to modulate its effect on network training. We present a two-stage dominance score estimation technique that compensates for performance degradation caused by sample inconsistencies. Our proposed method utilizes non-parametric estimation to infer sample inconsistency and assigns each sample a dominance score. This score is then aggregated with the loss function during training to modulate the impact of sample inconsistency. Furthermore, we design a curriculum learning approach that gradually increases the influence of inconsistent signals during training to improve overall performance. We evaluate our proposed method using public spontaneous BCI dataset. The experimental results confirm that our findings highlight the importance of addressing sample dominance for achieving robust performance in spontaneous BCIs.
Authors: Daniel Campos, Surya Kallumadi, Corby Rosset, Cheng Xiang Zhai, Alessandro Magnani
This is the first year of the TREC Product search track. The focus this year was the creation of a reusable collection and evaluation of the impact of the use of metadata and multi-modal data on retrieval accuracy. This year we leverage the new product search corpus, which includes contextual metadata. Our analysis shows that in the product search domain, traditional retrieval systems are highly effective and commonly outperform general-purpose pretrained embedding models. Our analysis also evaluates the impact of using simplified and metadata-enhanced collections, finding no clear trend in the impact of the expanded collection. We also see some surprising outcomes; despite their widespread adoption and competitive performance on other tasks, we find single-stage dense retrieval runs can commonly be noncompetitive or generate low-quality results both in the zero-shot and fine-tuned domain.
Authors: Chenjie Zhao, Ryan Wen Liu, Jingxiang Qu, Ruobin Gao
With the advancement of maritime unmanned aerial vehicles (UAVs) and deep learning technologies, the application of UAV-based object detection has become increasingly significant in the fields of maritime industry and ocean engineering. Endowed with intelligent sensing capabilities, the maritime UAVs enable effective and efficient maritime surveillance. To further promote the development of maritime UAV-based object detection, this paper provides a comprehensive review of challenges, relative methods, and UAV aerial datasets. Specifically, in this work, we first briefly summarize four challenges for object detection on maritime UAVs, i.e., object feature diversity, device limitation, maritime environment variability, and dataset scarcity. We then focus on computational methods to improve maritime UAV-based object detection performance in terms of scale-aware, small object detection, view-aware, rotated object detection, lightweight methods, and others. Next, we review the UAV aerial image/video datasets and propose a maritime UAV aerial dataset named MS2ship for ship detection. Furthermore, we conduct a series of experiments to present the performance evaluation and robustness analysis of object detection methods on maritime datasets. Eventually, we give the discussion and outlook on future works for maritime UAV-based object detection. The MS2ship dataset is available at \href{https://github.com/zcj234/MS2ship}{https://github.com/zcj234/MS2ship}.
Authors: Aaditya K. Singh, Stephanie C.Y. Chan, Ted Moskovitz, Erin Grant, Andrew M. Saxe, Felix Hill
Transformer neural networks can exhibit a surprising capacity for in-context learning (ICL) despite not being explicitly trained for it. Prior work has provided a deeper understanding of how ICL emerges in transformers, e.g. through the lens of mechanistic interpretability, Bayesian inference, or by examining the distributional properties of training data. However, in each of these cases, ICL is treated largely as a persistent phenomenon; namely, once ICL emerges, it is assumed to persist asymptotically. Here, we show that the emergence of ICL during transformer training is, in fact, often transient. We train transformers on synthetic data designed so that both ICL and in-weights learning (IWL) strategies can lead to correct predictions. We find that ICL first emerges, then disappears and gives way to IWL, all while the training loss decreases, indicating an asymptotic preference for IWL. The transient nature of ICL is observed in transformers across a range of model sizes and datasets, raising the question of how much to "overtrain" transformers when seeking compact, cheaper-to-run models. We find that L2 regularization may offer a path to more persistent ICL that removes the need for early stopping based on ICL-style validation tasks. Finally, we present initial evidence that ICL transience may be caused by competition between ICL and IWL circuits.
Authors: Joe Carlsmith
This report examines whether advanced AIs that perform well in training will be doing so in order to gain power later -- a behavior I call "scheming" (also sometimes called "deceptive alignment"). I conclude that scheming is a disturbingly plausible outcome of using baseline machine learning methods to train goal-directed AIs sophisticated enough to scheme (my subjective probability on such an outcome, given these conditions, is roughly 25%). In particular: if performing well in training is a good strategy for gaining power (as I think it might well be), then a very wide variety of goals would motivate scheming -- and hence, good training performance. This makes it plausible that training might either land on such a goal naturally and then reinforce it, or actively push a model's motivations towards such a goal as an easy way of improving performance. What's more, because schemers pretend to be aligned on tests designed to reveal their motivations, it may be quite difficult to tell whether this has occurred. However, I also think there are reasons for comfort. In particular: scheming may not actually be such a good strategy for gaining power; various selection pressures in training might work against schemer-like goals (for example, relative to non-schemers, schemers need to engage in extra instrumental reasoning, which might harm their training performance); and we may be able to increase such pressures intentionally. The report discusses these and a wide variety of other considerations in detail, and it suggests an array of empirical research directions for probing the topic further.