AI TrackerFrontier desk · AEST
Back to news
normalaitech

HiVe-Mem: Interleaving Horizontal Memory Expansion and Vertical Evidence Composition for Multimodal Agents

arXiv cs.AI · 10 Nov 2026, 04:57 AEST

The paper introduces HiVe-Mem, a multimodal memory framework designed to enhance information retrieval for agents with long interaction histories. It combines horizontal memory expansion and vertical evidence composition to improve task performance and reduce token consumption.

Key facts

Read aloud

The paper introduces HiVe-Mem, a multimodal memory framework designed to enhance information retrieval for agents with long interaction histories. It combines horizontal memory expansion and vertical evidence composition to improve task performance and reduce token consumption. As multimodal agents accumulate increasingly long interaction histories, effectively retrieving and utilizing past information becomes critical for long-term interaction. For horizontal memory expansion, HiVe-Mem organizes memory episodes into a relational graph using textual and visual anchors, expanding semantically retrieved seeds with related neighbors to recover complementary information. Experiments across three multimodal memory benchmarks show that HiVe-Mem consistently improves task performance across base models while reducing token consumption by about ninety percent compared with the most efficient competing multimodal baselines.

Sources in this cluster