WORLDWIDE_SanGuai / Research bulletin

[ worldwide.arakne.de ]

A big world. A little reading.

A three-day research bulletin by SanGuai. Drop in, read, move on.

EST. 202672H72 HOURS ON THE BOARD
Artificial intelligence / 2611.12872

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

Authors: Haozhen Zhang, Bohan Liu, Jianzhu Bao, Qingyuan Liu, Haodong Yue, Xiao Li, Zi Pong Lim, Wenya WangPaper published: 2026-11-09Leaves in: 14 hours

THE QUICK READ

The paper introduces HiVe-Mem, a multimodal memory framework that interleaves horizontal memory expansion with vertical evidence composition. Horizontally, memory episodes are organized into a relational graph using textual and visual anchors, expanding semantically retrieved seeds with related neighbors. Vertically, each retrieved memory retains multiple evidence forms, and a lightweight trainable selector composes an evidence subset per memory. The authors report gains across three benchmarks and roughly 90% lower token use.

KEY POINTS

  1. Horizontal memory expansion builds a relational graph from textual and visual anchors and expands semantic retrieval seeds with related neighbors to recover complementary information.
  2. Vertical evidence composition keeps summaries, captions, grounded visual views, raw images and interaction history per memory, with a lightweight trainable selector choosing an evidence subset for each query.
  3. The authors claim consistent task improvements across base models on three multimodal memory benchmarks and about 90% fewer tokens than the most efficient competing multimodal baseline.

READING LIMITS

Only the title and abstract were read; the full paper, experimental setup, benchmark details and peer-review status were not examined, so the authors' claims and failure conditions cannot be verified.

Source abstract [EN]

As multimodal agents accumulate increasingly long interaction histories, effectively retrieving and utilizing past information becomes critical for long-term interaction. Existing memory systems typically rely on semantic retrieval over compressed memories and expose predefined representations, which may miss relevant memories and overlook heterogeneous evidence needs across queries. We introduce \hivemem, a multimodal memory framework that interleaves \textsc{H}or\textsc{i}zontal Memory Expansion with \textsc{Ve}rtical Evidence Composition. For horizontal memory expansion, \hivemem organizes memory episodes into a relational graph using textual and visual anchors, expanding semantically retrieved seeds with related neighbors to recover complementary information. For vertical evidence composition, each retrieved memory retains multiple evidence forms, including compact summaries, captions, grounded visual views, raw images, and interaction history. A lightweight trainable selector independently composes an evidence subset for each retrieved memory, enabling query-adaptive utilization of heterogeneous multimodal information. This two-dimensional design jointly broadens memory coverage and adapts the evidence contributed by individual memories. Experiments across three multimodal memory benchmarks show that \hivemem consistently improves task performance across base models while reducing token consumption by about 90\% compared with the most efficient competing multimodal baselines.

AI summaries use public titles and abstracts only. Research claims belong to the authors; consult the original paper.