Mohammad Beigi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Mohammad Beigi has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
- Lifu Huang
University of California, Davis, Davis, California, USA
5 shared publications Ming Jin
3 shared publicationsQifan Wang
3 shared publicationsJunshan Zhang
2 shared publicationsYing Shen
2 shared publicationsZichao Wang
Adobe Research
2 shared publications
Latest Publications
- Open Access
IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking
Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking - models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the objectives internalized during RLHF remain opaque, making hacking behaviors difficult to detect or correct. We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models. We propose Contrastive Inverse Reinforcement Learning (C-IRL), which reconstructs the implicit reward function by contrasting paired responses from post-alignment and baseline policies to explain behavioral shifts during RLHF. We then decompose the reconstructed reward via sparse autoencoders into interpretable features, enabling identification of hacking signatures through contribution analysis. Finally, we propose mitigation strategies - clean reward optimization, adversarial shaping, constrained optimization, and feature-guided distillation - that target problematic features while preserving beneficial alignment. Experiments across multiple reward model configurations show that IR3 achieves 0.89 correlation with ground-truth rewards, identifies hacking features with over 90% precision, and significantly reduces hacking behaviors while maintaining capabilities within 3% of the original model.
Open AccessIR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking
Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking - models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the objectives internalized during RLHF remain opaque, making hacking behaviors difficult to detect or correct. We introduce IR3 (Interpretable Reward Reconstruction and Rectification), a framework that reverse-engineers, interprets, and surgically repairs the implicit objectives driving RLHF-tuned models. We propose Contrastive Inverse Reinforcement Learning (C-IRL), which reconstructs the implicit reward function by contrasting paired responses from post-alignment and baseline policies to explain behavioral shifts during RLHF. We then decompose the reconstructed reward via sparse autoencoders into interpretable features, enabling identification of hacking signatures through contribution analysis. Finally, we propose mitigation strategies - clean reward optimization, adversarial shaping, constrained optimization, and feature-guided distillation - that target problematic features while preserving beneficial alignment. Experiments across multiple reward model configurations show that IR3 achieves 0.89 correlation with ground-truth rewards, identifies hacking features with over 90% precision, and significantly reduces hacking behaviors while maintaining capabilities within 3% of the original model.
- Open Access
Adversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking
Reinforcement Learning from Human Feedback (RLHF) remains vulnerable to reward hacking, where models exploit spurious correlations in learned reward models to achieve high scores while violating human intent. Existing mitigations rely on static defenses that cannot adapt to novel exploitation strategies. We propose Adversarial Reward Auditing (ARA), a framework that reconceptualizes reward hacking as a dynamic, competitive game. ARA operates in two stages: first, a Hacker policy discovers reward model vulnerabilities while an Auditor learns to detect exploitation from latent representations; second, Auditor-Guided RLHF (AG-RLHF) gates reward signals to penalize detected hacking, transforming reward hacking from an unobservable failure into a measurable, controllable signal. Experiments across three hacking scenarios demonstrate that ARA achieves the best alignment-utility tradeoff among all baselines: reducing sycophancy to near-SFT levels while improving helpfulness, decreasing verbosity while achieving the highest ROUGE-L, and suppressing code gaming while improving Pass@1. Beyond single-domain evaluation, we show that reward hacking, detection, and mitigation all generalize across domains -- a Hacker trained on code gaming exhibits increased sycophancy despite no reward for this behavior, and an Auditor trained on one domain effectively suppresses exploitation in others, enabling efficient multi-domain defense with a single model.
Open AccessAdversarial Reward Auditing for Active Detection and Mitigation of Reward Hacking
Reinforcement Learning from Human Feedback (RLHF) remains vulnerable to reward hacking, where models exploit spurious correlations in learned reward models to achieve high scores while violating human intent. Existing mitigations rely on static defenses that cannot adapt to novel exploitation strategies. We propose Adversarial Reward Auditing (ARA), a framework that reconceptualizes reward hacking as a dynamic, competitive game. ARA operates in two stages: first, a Hacker policy discovers reward model vulnerabilities while an Auditor learns to detect exploitation from latent representations; second, Auditor-Guided RLHF (AG-RLHF) gates reward signals to penalize detected hacking, transforming reward hacking from an unobservable failure into a measurable, controllable signal. Experiments across three hacking scenarios demonstrate that ARA achieves the best alignment-utility tradeoff among all baselines: reducing sycophancy to near-SFT levels while improving helpfulness, decreasing verbosity while achieving the highest ROUGE-L, and suppressing code gaming while improving Pass@1. Beyond single-domain evaluation, we show that reward hacking, detection, and mitigation all generalize across domains -- a Hacker trained on code gaming exhibits increased sycophancy despite no reward for this behavior, and an Auditor trained on one domain effectively suppresses exploitation in others, enabling efficient multi-domain defense with a single model.
- Open Access
Sycophancy Mitigation Through Reinforcement Learning with Uncertainty-Aware Adaptive Reasoning Trajectories
Despite the remarkable capabilities of large language models, current training paradigms inadvertently foster \textit{sycophancy}, i.e., the tendency of a model to agree with or reinforce user-provided information even when it's factually incorrect. To address this challenge, we introduce \textbf{SMART} (Sycophancy Mitigation through Adaptive Reasoning Trajectories), which reframes sycophancy as a \textit{reasoning optimization problem} rather than an output alignment issue. SMART is a two-stage framework comprising: (1) Uncertainty-Aware Adaptive Monte Carlo Tree Search (UA-MCTS), which dynamically adjusts model exploration based on state-level uncertainty to collect high-quality, diverse reasoning trajectories alongside both stepwise progress and final outcome rewards; and (2) progress-based reinforcement learning, which fine-tunes the model using the collected trajectories and reward signals to reinforce effective reasoning patterns. Through extensive experiments, we show that SMART significantly reduces sycophantic behavior while preserving strong performance on out-of-distribution inputs and maintaining general capabilities. These results underscore the importance of optimizing internal reasoning mechanisms to build more truthful and aligned AI assistants.
Open AccessSycophancy Mitigation Through Reinforcement Learning with Uncertainty-Aware Adaptive Reasoning Trajectories
Despite the remarkable capabilities of large language models, current training paradigms inadvertently foster \textit{sycophancy}, i.e., the tendency of a model to agree with or reinforce user-provided information even when it's factually incorrect. To address this challenge, we introduce \textbf{SMART} (Sycophancy Mitigation through Adaptive Reasoning Trajectories), which reframes sycophancy as a \textit{reasoning optimization problem} rather than an output alignment issue. SMART is a two-stage framework comprising: (1) Uncertainty-Aware Adaptive Monte Carlo Tree Search (UA-MCTS), which dynamically adjusts model exploration based on state-level uncertainty to collect high-quality, diverse reasoning trajectories alongside both stepwise progress and final outcome rewards; and (2) progress-based reinforcement learning, which fine-tunes the model using the collected trajectories and reward signals to reinforce effective reasoning patterns. Through extensive experiments, we show that SMART significantly reduces sycophantic behavior while preserving strong performance on out-of-distribution inputs and maintaining general capabilities. These results underscore the importance of optimizing internal reasoning mechanisms to build more truthful and aligned AI assistants.
- Open Access
Survey of Uncertainty Estimation in Large Language Models -Sources, Methods, Applications, and Challenge
Large Language Models (LLMs) have demonstrated exceptional performance across a wide range of domains. However, inaccuracies in their outputs can lead to severe consequences in high-stakes areas such as finance and healthcare, where errors may result in the loss of money, time, or even lives. As a result, recent research has increasingly focused on uncertainty estimation in LLMs, aiming to quantify the trustworthiness of model-generated content given specific inputs. Despite this growing interest, the sources of uncertainty in LLMs remain insufficiently understood. As a result, this survey provides a comprehensive overview of uncertainty estimation for LLMs from the perspective of uncertainty sources, serving as a foundational resource for researchers entering the field. We begin by reviewing essential background on LLMs, followed by a detailed clarification of uncertainty sources relevant to them. We then introduce various uncertainty estimation methods, including both commonly used and LLM-specific approaches. Metrics for evaluating uncertainty are discussed, along with key application areas. Finally, we highlight major challenges and outline future research directions aimed at improving the trustworthiness and reliability of LLMs.
Open AccessSurvey of Uncertainty Estimation in Large Language Models -Sources, Methods, Applications, and Challenge
Large Language Models (LLMs) have demonstrated exceptional performance across a wide range of domains. However, inaccuracies in their outputs can lead to severe consequences in high-stakes areas such as finance and healthcare, where errors may result in the loss of money, time, or even lives. As a result, recent research has increasingly focused on uncertainty estimation in LLMs, aiming to quantify the trustworthiness of model-generated content given specific inputs. Despite this growing interest, the sources of uncertainty in LLMs remain insufficiently understood. As a result, this survey provides a comprehensive overview of uncertainty estimation for LLMs from the perspective of uncertainty sources, serving as a foundational resource for researchers entering the field. We begin by reviewing essential background on LLMs, followed by a detailed clarification of uncertainty sources relevant to them. We then introduce various uncertainty estimation methods, including both commonly used and LLM-specific approaches. Metrics for evaluating uncertainty are discussed, along with key application areas. Finally, we highlight major challenges and outline future research directions aimed at improving the trustworthiness and reliability of LLMs.
- Open Access
A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models
The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This survey explores two key aspects: (1) the adaptation of LLM interpretability methods to multimodal models and (2) understanding the mechanistic differences between unimodal language models and crossmodal systems. By systematically reviewing current MMFM analysis techniques, we propose a structured taxonomy of interpretability methods, compare insights across unimodal and multimodal architectures, and highlight critical research gaps.
Open AccessA Survey on Mechanistic Interpretability for Multi-Modal Foundation Models
The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This survey explores two key aspects: (1) the adaptation of LLM interpretability methods to multimodal models and (2) understanding the mechanistic differences between unimodal language models and crossmodal systems. By systematically reviewing current MMFM analysis techniques, we propose a structured taxonomy of interpretability methods, compare insights across unimodal and multimodal architectures, and highlight critical research gaps.