Tianyi Zhou

Assistant Professor@University of Maryland, College Park·美国·OpenReview

研究方向

LLMs · multi-modal agent · personalization and controllability of generative AI · multi-modality models · vision-language models · multi-objective optimization · continual learning · interpretable AI · self-supervised learning · reinforcement learning · federated learning · meta-learning · curriculum learning · mixture-of-experts · NLP · attention mechanism · submodular optimization · combinatorial optimization · spectral method · dimension reduction · matrix factorization · sparse learning · compressed sensing · multi-label learning · multi-task learning

Tianyi Zhou

Your Mixture-of-Experts LLM Is Secretly an Embedding Model for Free

Preference Controllable Reinforcement Learning with Advanced Multi-Objective Optimization

Reweighting Local Mimina with Tilted SAM

The Crystal Ball Hypothesis in diffusion models: Anticipating object positions from initial noise

Many-Objective Multi-Solution Transport

BenTo: Benchmark Reduction with In-Context Transferability

Is Your Multimodal Language Model Oversensitive to Safe Queries?

DataGen: Unified Synthetic Dataset Generation via Large Language Models

OmnixR: Evaluating Omni-modality Language Models on Reasoning across Modalities

Customizing Reinforcement Learning Agent with Multi-Objective Preference Control

One Model for All: Multi-Objective Controllable Language Models

R2-T2: Re-Routing in Test-Time for Multimodal Mixture-of-Experts

Mosaic-IT: Free Compositional Data Augmentation Improves Instruction Tuning

Tilted Sharpness-Aware Minimization

Federated Adapter on Foundation Models: An Out-Of-Distribution Approach

OPTune: Efficient Online Preference Tuning

Diffusion Curriculum: Synthetic-to-Real Data Curriculum via Image-Guided Diffusion

MuLan: Multimodal-LLM Agent for Progressive and Interactive Multi-Object Diffusion

Task-Adaptation Curriculum Learning

GenQA: An Instruction Dataset of LLM Generated Questions and Answers

Hybrid Memory Replay: Blending Real and Distilled Data for Class Incremental Learning

WALL-E: World Alignment by Rule Learning Improves World Model-based LLM Agents

Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning

Federated Recommendation with Additive Personalization

BumbleBee: Dynamic KV-Cache Streaming Submodular Summarization for Infinite-Context Transformers

Adaptive Regularization of Representation Rank as an Implicit Constraint of Bellman Equation

AlpaGasus: Training a Better Alpaca with Fewer Data

InstructZero: Efficient Instruction Optimization for Black-Box Large Language Models

Curriculum Reinforcement Learning via Morphology-Environment Co-Evolution

NLPBench: Evaluating Large Language Models on Solving NLP Problems

YoooP: You Only Optimize One Prototype per Class for Non-Exemplar Incremental Learning

Learning UI-to-Code Reverse Generator Using Visual Critic Without Rendering

Mixture-of-Experts in Prompt Optimization

Multi-Objective Multi-Solution Transport

Curiosity Driven Protein Sequence Generation via Reinforcement Learning

Controllable Pareto Trade-off between Fairness and Accuracy