Tianyi Zhou

MADAug improves data augmentation by adapting techniques to model needs during training.

A new method allows RL agents to change body shapes for better task adaptation.

A new dataset for evaluating LLMs on natural language processing questions.

Exploring the challenges and solutions of reward hacking in AI model training.

MuLan enhances image generation from text by breaking tasks into simpler steps.

New method improves sentence embeddings using multiple prompts for better accuracy.

Examining LLMs' capability to address mathematical problems, especially modular arithmetic.

Exploring how transformers favor low sensitivity functions for improved performance.

A new method to improve federated learning through Multi-Level Additive Modeling.

Research reveals how trigger patches influence image generation in diffusion models.

Explore how LLMs perform addition using unique mathematical techniques.

A new method enhances language models by integrating knowledge across languages.

This method enhances prompt creation for large language models, improving overall performance.

A new framework allows AI agents to create actions dynamically for better problem-solving.

SpecHub speeds up language models' text generation with a fresh approach.

Exploring how audio tricks confuse language models.

Florence-2 and DBFusion redefine how machines interpret images and text.