18 papers
Decoupled Alignment for Robust Plug-and-Play Adaptation
Haozheng Luo, Jiahao Yu, Wenxin Zhang +9
The paper proposes a training-free, plug-and-play method that uses knowledge distillation and model fusion to correct misaligned (shadow-aligned) large language models, improving s…
Transformer Approximations from ReLUs
Jerry Yao-Chieh Hu, Mingcheng Lu, Yi-Chen Lee +1
We provide a systematic recipe for translating ReLU approximation results to softmax attention mechanism. This recipe covers many common approximation targets. Importantly, it yiel…
Discrete Flow Matching Policy Optimization
Maojiang Su, Po-Chung Hsieh, Weimin Wu +4
We introduce Discrete flow Matching policy Optimization (DoMinO), a unified framework for Reinforcement Learning (RL) fine-tuning Discrete Flow Matching (DFM) models under a broad…
Cell-JEPA: Latent Representation Learning for Single-Cell Transcriptomics
Ali ElSheikh, Rui-Xi Wang, Weimin Wu +9
Single-cell foundation models learn by reconstructing masked gene expression, implicitly treating technical noise as signal. With dropout rates exceeding 90%, reconstruction object…
On Structured State-Space Duality
Jerry Yao-Chieh Hu, Xiwen Zhang, Ali ElSheikh +2
Structured State-Space Duality (SSD) [Dao & Gu, ICML 2024] is an equivalence between a simple Structured State-Space Model (SSM) and a masked attention mechanism. In particular, a…
Universal Approximation with Softmax Attention
Jerry Yao-Chieh Hu, Hude Liu, Hong-Yu Chen +2
We prove that with linear transformations, both (i) two-layer self-attention and (ii) one-layer self-attention followed by a softmax function are universal approximators for contin…