Dongming Jiang

V-Mem achieves a groundbreaking LLM-judge score of 0.82 in multimodal memory retrieval, outperforming existing systems by leveraging modality-specific routing and LLM-generated anchors.

Multi-agent LLMs can achieve up to 30% higher accuracy and consume 6.5x fewer tokens by minimizing direct agent communication and instead coordinating via a structured belief state.

Suppressing background noise with heatmap-guided positional embeddings slashes transformer detector parameters by 59% without sacrificing accuracy in small object detection.

Key contribution not extracted.