18 papers
How RL Reshapes LLM Reasoning: Transferability, Coverage, and Scaling Laws
Ziheng Cheng, Yixiao Huang, Hanlin Zhu +1
Recent studies on reinforcement learning (RL) report seemingly conflicting evidence about large language model (LLM) reasoning. Training on mathematics can improve performance in o…
DiscoLoop: Looping Discrete Embeddings and Continuous Hidden States for Multi-hop Reasoning
Hengyu Fu, Tianyu Guo, Zixuan Wang +5
Large language models achieve strong performance on many reasoning tasks when allowed to externalize intermediate steps as Chain-of-Thought (CoT). However, many questions require t…
Transformers Provably Learn to Internalize Chain-of-Thought
Yixiao Huang, Hanlin Zhu, Zixuan Wang +4
Chain-of-Thought (CoT) prompting substantially improves the sample efficiency of transformers, reducing the complexity of tasks like parity learning from exponential to polynomial…
Multi-Objective Learning for Diffusion Models: A Statistical Theory under Semi-Supervised Learning
Ziheng Cheng, Yixiao Huang, Hanlin Zhu +5
Diffusion models are increasingly used as powerful conditional generators, yet real deployments often involve multiple target distributions arising from different tasks, e.g., dive…
CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning
Dachuan Shi, Hanlin Zhu, Xiangchi Yuan +4
Chain-of-thought (CoT) is a standard approach for eliciting reasoning capabilities from large language models (LLMs). However, the common CoT paradigm treats thinking as a prerequi…
Breaking the Reversal Curse in Autoregressive Language Models via Identity Bridge
Xutao Ma, Yixiao Huang, Hanlin Zhu +1
Autoregressive large language models (LLMs) have achieved remarkable success in many complex tasks, yet they can still fail in very simple logical reasoning such as the "reversal c…
GSM-Agent: Understanding Agentic Reasoning Using Controllable Environments
Hanlin Zhu, Tianyu Guo, Song Mei +4
As LLMs are increasingly deployed as agents, agentic reasoning - the ability to combine tool use, especially search, and reasoning - becomes a critical skill. However, it is hard t…
Emergence of Superposition: Unveiling the Training Dynamics of Chain of Continuous Thought
Hanlin Zhu, Shibo Hao, Zhiting Hu +3
Previous work shows that the chain of continuous thought (continuous CoT) improves the reasoning capability of large language models (LLMs) by enabling implicit parallel thinking,…
Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers
Yixiao Huang, Hanlin Zhu, Tianyu Guo +5
Large language models (LLMs) can acquire new knowledge through fine-tuning, but this process exhibits a puzzling duality: models can generalize remarkably from new facts, yet are a…
Auditing Black-Box LLM APIs with a Rank-Based Uniformity Test
Xiaoyuan Zhu, Yaowen Ye, Tianyi Qiu +6
As API access becomes a primary interface to large language models (LLMs), users often interact with black-box systems that offer little transparency into the deployed model. To re…
Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thought
Hanlin Zhu, Shibo Hao, Zhiting Hu +3
Large Language Models (LLMs) have demonstrated remarkable performance in many applications, including challenging reasoning problems via chain-of-thoughts (CoTs) techniques that ge…
Safe Learning Under Irreversible Dynamics via Asking for Help
Benjamin Plaut, Juan Liévano-Karim, Hanlin Zhu +1
Most learning algorithms with formal regret guarantees essentially rely on trying all possible behaviors, which is problematic when some errors cannot be recovered from. Instead, w…
How Do LLMs Perform Two-Hop Reasoning in Context?
Tianyu Guo, Hanlin Zhu, Ruiqi Zhang +4
``Socrates is human. All humans are mortal. Therefore, Socrates is mortal.'' This form of argument illustrates a typical pattern of two-hop reasoning. Formally, two-hop reasoning r…
Towards a Theoretical Understanding of the 'Reversal Curse' via Training Dynamics
Hanlin Zhu, Baihe Huang, Shaolun Zhang +4
Auto-regressive large language models (LLMs) show impressive capacities to solve many complex reasoning tasks while struggling with some simple logical reasoning tasks such as inve…
Efficient Prompt Caching via Embedding Similarity
Hanlin Zhu, Banghua Zhu, Jiantao Jiao
Large language models (LLMs) have achieved huge success in numerous natural language process (NLP) tasks. However, it faces the challenge of significant resource consumption during…
Avoiding Catastrophe in Online Learning by Asking for Help
Benjamin Plaut, Hanlin Zhu, Stuart Russell
Most learning algorithms with formal regret guarantees assume that all mistakes are recoverable and essentially rely on trying all possible behaviors. This approach is problematic…
Towards Optimal Statistical Watermarking
Baihe Huang, Hanlin Zhu, Banghua Zhu +4
We study statistical watermarking by formulating it as a hypothesis testing problem, a general framework which subsumes all previous statistical watermarking methods. Key to our fo…
On Representation Complexity of Model-based and Model-free Reinforcement Learning
Hanlin Zhu, Baihe Huang, Stuart Russell
We study the representation complexity of model-based and model-free reinforcement learning (RL) in the context of circuit complexity. We prove theoretically that there exists a br…