AOEB: Benchmarking Agent-Oriented Multimodal Embeddings
ICML 2026
Evaluating multimodal embeddings against the retrieval needs of LLM agents, including code, tools, reasoning, and memory.
Jiaxin Xu许佳鑫
Master’s student · HIT Shenzhen硕士研究生 · 哈尔滨工业大学（深圳）
I study information retrieval for LLM agents, with a focus on efficient reranking, agent-oriented evaluation, and long-horizon memory retrieval.我研究面向大语言模型智能体的信息检索，关注高效重排序、面向智能体的评测，以及长程记忆检索。
I’m pursuing a master’s in Computer Technology at Harbin Institute of Technology, Shenzhen, advised by 目前在哈尔滨工业大学（深圳）攻读计算机技术硕士学位，导师是Prof. Meishan Zhang张梅山教授.。
Research & resources研究与资源
KaLM-Reranker-V1: Fast but Not Late Interaction for Compressed Document Reranking
Preprint · 2026
An efficient reranker that separates offline passage encoding from online query processing while retaining expressive cross-attention.
LMEB: Long-horizon Memory Embedding Benchmark
Preprint · 2026
A benchmark for long-horizon memory retrieval: 22 datasets and 193 retrieval tasks spanning episodic, dialogue, semantic, and procedural memory.