Wang Yang
16 indexed papers
Research Timeline
MOSAIC is a multi-objective framework that efficiently allocates a fixed supervised fine-tuning budget by turning failure profiles into actionable data mixtures, significantly improving model alignment across multiple safety and capability metrics.
FedMPT introduces a novel federated learning framework for Multi-Label Recognition (MLR) using Vision-Language Models (VLMs) by leveraging generalizable conditions to mitigate label overfitting and improve robustness.
FedMPT introduces a novel federated learning framework for Multi-Label Recognition (MLR) using Vision-Language Models (VLMs) by leveraging generalizable conditions to mitigate label overfitting and improve robustness.
The paper introduces PTCG-Bench, a new benchmark using the Pokémon TCG to evaluate LLM agents' strategic decision-making and ability to self-evolve, finding that sustained self-evolution remains challenging.
The paper introduces PTCG-Bench, a new benchmark using the Pokémon TCG to evaluate LLM agents' strategic decision-making and ability to self-evolve, finding that sustained self-evolution remains challenging.
The eMoT framework enhances multi-step reasoning in LLMs by treating reasoning as an evolving memory, stabilizing performance through symbolic computation and structured refinement.
MOSS-Audio is a unified audio-language model designed for comprehensive understanding of speech, environmental sounds, and music, achieving strong performance across various audio-grounded tasks.
The paper identifies a failure mode called spatial lexical bias in MLLMs, where adding a spatial word to options biases the model's choice, and demonstrates that this failure originates primarily from the language processing side rather than poor visual attention.
The eMoT framework enhances multi-step reasoning in LLMs by treating reasoning as an evolving memory, stabilizing performance through symbolic computation and structured refinement.
MOSS-Audio is a unified audio-language model designed for comprehensive understanding of speech, environmental sounds, and music, achieving strong performance across various audio-grounded tasks.
The paper identifies a failure mode called spatial lexical bias in MLLMs, where adding a spatial word to options biases the model's choice, and demonstrates that this failure originates primarily from the language processing side rather than poor visual attention.
One clear sentence describing what this paper does
This paper proposes a new framework, Aligned Refinement Policy (ARP), to learn visuomotor policies for long-horizon manipulation tasks.
This paper introduces an agentic framework for industrial video anomaly detection, capable of tracking spatial-temporal dynamics and underlying transformations of objects to identify abnormalities.
The paper presents Foil, a method for looping sparse mixture-of-experts (MoE) models to improve expert usage and performance.
The paper proposes a search-to-list item distillation framework called SLIDE to improve generative retrieval by decoupling offline search from online list generation.
Papers
SLIDE: Search-to-List Item Distillation for Single-Rollout Generative Retrieval
Tianyu Zhan, Gui Ling, Ran Tao, Kunhai Lin +8 more
The paper proposes a search-to-list item distillation framework called SLIDE to improve generative retrieval by decoupling offline search from online list generation.