Reproduce “HiVe-Mem: Interleaving Horizontal Memory Expansion and Vertical Evidence Composition for Multimodal Agents” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haozhen Zhang, Bohan Liu, Jianzhu Bao, Qingyuan Liu, Haodong Yue, Xiao Li, Zi Pong Lim, Wenya Wang
HiVe-Mem retrieves episodes by expanding semantic search through shared textual and visual anchors, then selects evidence separately for each episode. A PPO-trained selector balances answer quality against token use. Across three benchmarks with Qwen3-VL-4B-Instruct, it averages 32.32 F1 and 57.96 LJ using 0.658 million tokens, leading multimodal baselines on average; evidence covers three benchmarks and two base models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.