10 papers
Query-Efficient Rejection Sampling for Parallel Masked Diffusion
Shiyuan Zhang, Xuheng Li, Qiwei Di +1
Masked diffusion models (MDMs) accelerate generation by unmasking multiple tokens in parallel, but sampling these tokens independently ignores their correlations and biases the out…
One Block Ahead: Promoting Sampler State in Diffusion Language Models
Yangyi Shen, Pin-Jui Ku, Nicolai Oswald +6
Block-diffusion language models denoise a whole block of token positions in parallel so each forward pass can emit many tokens. However, blocks are still denoised in strict sequenc…
Muon Learns Facts Better: Understanding the Role of Spectral Orthogonalization
Xuheng Li, Qiwei Di, Yuan Cao +1
The Muon optimizer applies spectral orthogonalization to matrix-valued updates and has shown strong performance in large-scale neural network training, yet the mechanisms of this t…
Understanding Off- vs On-Policy Distillation: A Tale of Distinct Training Objectives
Qiwei Di, Xuheng Li, Kaixuan Ji +3
On-policy distillation (OPD) learns from teacher feedback on student-generated responses and has shown promise in reducing forgetting relative to supervised fine-tuning (SFT). Howe…
Dimension-Independent Convergence of Underdamped Langevin Monte Carlo in KL Divergence
Shiyuan Zhang, Qiwei Di, Xuheng Li +1
Underdamped Langevin dynamics (ULD) is a widely-used sampler for Gibbs distributions , and is often empirically effective in high dimensions. However, existing non…
Variance-Aware Feel-Good Thompson Sampling for Contextual Bandits
Xuheng Li, Quanquan Gu
Variance-dependent regret bounds have received increasing attention in recent studies on contextual bandits. However, most of these studies are focused on upper confidence bound (U…
Best-of-Majority: Minimax-Optimal Strategy for Pass@ Inference Scaling
Qiwei Di, Kaixuan Ji, Xuheng Li +2
LLM inference often generates a batch of candidates for a prompt and selects one via strategies like majority voting or Best-of- N (BoN). For difficult tasks, this single-shot sele…
Understanding SGD with Exponential Moving Average: A Case Study in Linear Regression
Xuheng Li, Quanquan Gu
Exponential moving average (EMA) has recently gained significant popularity in training modern deep learning models, especially diffusion-based generative models. However, there ha…
Feel-Good Thompson Sampling for Contextual Dueling Bandits
Xuheng Li, Heyang Zhao, Quanquan Gu
Contextual dueling bandits, where a learner compares two options based on context and receives feedback indicating which was preferred, extends classic dueling bandits by incorpora…
Risk Bounds of Accelerated SGD for Overparameterized Linear Regression
Xuheng Li, Yihe Deng, Jingfeng Wu +2
Accelerated stochastic gradient descent (ASGD) is a workhorse in deep learning and often achieves better generalization performance than SGD. However, existing optimization theory…