Dizhan Xue

CRISPR: Context-Refined Information Spatial Pooling with Region-awareness for Efficient Visual Token Compression in VLMs

This paper proposes CRISPR, a context-refined information spatial pooling framework with region-awareness for visual token compression, and demonstrates that CRISPR consistently outperforms low-resolution baselines at a comparable compressed sequence length.

Curriculum-based inter-modality contrastive learning for semi-supervised cross-modal retrieval

HarnessBank: Semantic Gene-Bank Search with Gated Verification for Agent-Harness Self-Evolution

Large Language Models (LLMs) have enabled capable agents across diverse applications. Beyond the foundation model, the performance of an agent is governed by the surrounding agent harness, including prompts, tools, control loops, etc. Automatically evolving this harness offers a promising pathway to agent improvement,...