Maojiang Su

A Theoretical Analysis of Discrete Flow Matching Generative Models

Minimalist Softmax Attention Provably Learns Constrained Boolean Functions

On Differentially Private String Distances

Fast and Low-Cost Genomic Foundation Models via Outlier Removal

High-Order Flow Matching: Unified Framework and Sharp Statistical Rates

In-Context Deep Learning via Transformer Models

Computational Limits of Low-Rank Adaptation (LoRA) Fine-Tuning for Transformer Models

SPARQ: Outlier-free SpeechLM with Fast Adaptation and Robust Quantization

Pretrained Transformers are Deep Optimizers: Provable In-Context Learning for Deep Model Training