17 papers
Score Mismatch and Set-Constrained Sampling in Zero-Shot Conditional Discrete Diffusion
Renxiang Huang, Yuchen Liang, Yingbin Liang +2
Discrete diffusion models have become a powerful framework for modeling complex discrete distributions across a range of applications. While theoretical convergence guarantees depe…
Understanding Unified VLMs: From Attention Probing to Guidance Tuning for Multimodal Generation
Zhengyang Luo, Xiaoyu Qiu, Zhiheng Tan +3
Unified vision-language models (VLMs) perform both text-to-image (T2I) and image-to-text (I2T) generation with one shared Transformer backbone, yet how the backbone processes the c…
PACT: Privileged Trace Co-Training for Multi-Turn Tool-Use Agents
Zhenbang Du, Jun Luo, Zhiwei Zheng +8
Multi-turn tool-use agents must reason, call tools, and adapt to observations across several interaction turns. Post-training such agents is challenging, as reinforcement learning…
From Scores to Gibbs Correctors: Accelerating Uniform-Rate Discrete Diffusion Models
Yuchen Liang, Ness Shroff, Yingbin Liang
Discrete diffusion models have achieved strong empirical performance in text and other symbolic domains, but, especially for uniform-rate models, they often require many steps to g…
Regret Bounds for Reinforcement Learning from Multi-Source Imperfect Preferences
Ming Shi, Yingbin Liang, Ness B. Shroff +1
Reinforcement learning from human feedback (RLHF) replaces hard-to-specify rewards with pairwise trajectory preferences, yet regret-oriented theory often assumes that preference la…
Learnable Chernoff Baselines for Inference-Time Alignment
Sunil Madhow, Yuchen Liang, Ness Shroff +2
We study inference-time reward-guided alignment for generative models. Existing methods often rely on either architecture-specific adaptations or computationally costly inference p…
Discrete Diffusion Models: Novel Analysis and New Sampler Guarantees
Yuchen Liang, Yingbin Liang, Lifeng Lai +1
Discrete diffusion models have recently gained significant prominence in applications involving natural language and graph data. A key factor influencing their effectiveness is the…
Mixture-of-Transformers Learn Faster: A Theoretical Study on Classification Problems
Hongbo Li, Qinhang Wu, Sen Lin +2
Mixture-of-Experts (MoE) models improve transformer efficiency but lack a unified theoretical explanation, especially when both feed-forward and attention layers are allowed to spe…
Monitoring State Transitions in Markovian Systems with Sampling Cost
Kumar Saurav, Ness B. Shroff, Yingbin Liang
We consider a node-monitor pair, where the node's state varies with time. The monitor needs to track the node's state at all times; however, there is a fixed cost for each state qu…
Large Language Models Achieve Gold Medal Performance at the International Olympiad on Astronomy & Astrophysics (IOAA)
Lucas Carrit Delgado Pinheiro, Ziru Chen, Bruno Caixeta Piazza +4
While task-specific demonstrations show early success in applying large language models (LLMs) to automate some astronomical research tasks, they only provide incomplete views of a…
Provable In-Context Learning of Nonlinear Regression with Transformers
Hongbo Li, Lingjie Duan, Yingbin Liang
The transformer architecture, which processes sequences of input tokens to produce outputs for query tokens, has revolutionized numerous areas of machine learning. A defining featu…
Unlocking the Power of Rehearsal in Continual Learning: A Theoretical Perspective
Junze Deng, Qinhang Wu, Peizhong Ju +3
Rehearsal-based methods have shown superior performance in addressing catastrophic forgetting in continual learning (CL) by storing and training on a subset of past data alongside…
Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers
Yuchen Liang, Peizhong Ju, Yingbin Liang +1
The denoising diffusion model has recently emerged as a powerful generative technique, capable of transforming noise into meaningful data. While theoretical convergence guarantees…
Can We Theoretically Quantify the Impacts of Local Updates on the Generalization Performance of Federated Learning?
Peizhong Ju, Haibo Yang, Jia Liu +2
Federated Learning (FL) has gained significant popularity due to its effectiveness in training machine learning models across diverse sites without requiring direct data sharing. W…
Theory on Mixture-of-Experts in Continual Learning
Hongbo Li, Sen Lin, Lingjie Duan +2
Continual learning (CL) has garnered significant attention because of its ability to adapt to new tasks that arrive over time. Catastrophic forgetting (of old tasks) has been ident…
Broadening Target Distributions for Accelerated Diffusion Models via a Novel Analysis Approach
Yuchen Liang, Peizhong Ju, Yingbin Liang +1
Accelerated diffusion models hold the potential to significantly enhance the efficiency of standard diffusion processes. Theoretically, these models have been shown to achieve fast…
Near-Optimal Partially Observable Reinforcement Learning with Partial Online State Information
Ming Shi, Yingbin Liang, Ness B. Shroff
Partially observable Markov decision processes (POMDPs) are a general framework for sequential decision-making under latent state uncertainty, yet learning in POMDPs is intractable…