AI
Beyond Dyadic Memory: Interaction-Aware Multimodal Memory with Adaptive Agentic Retrieval for Multi-Party Spoken Conversations
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 25/09/2026, 20:00:00 UTC
Briefing created: 30/09/2026, 07:26:48 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 07:18:24 UTC.
Briefings
This article introduces VoxPolyMem, an interaction-aware multimodal memory framework designed for multi-party spoken conversations. The system combines incremental speaker identification with a memory hierarchy including interaction memory, fact memory, and participant profiles. Retrieval is formulated as sequential decision-making where agents rewrite queries and select tools based on accumulated evidence. The framework includes Evidence-Gain GRPO for credit assignment and VoxPolyBench for evaluation. VoxPolyMem achieved an overall score of 85.0 on VoxPolyBench, surpassing the strongest baseline by 23.6 points, while scoring 89.6 on Mem-Gallery and 74.4 on H2HMem-Multi.
Key points
- VoxPolyMem introduces a memory hierarchy consisting of interaction memory, fact memory, and participant profiles to track multi-party spoken conversations.
- The framework utilizes Evidence-Gain GRPO to encourage complementary evidence acquisition through round-wise credit assignment.
Context and limitations
Evaluation is limited to VoxPolyBench, Mem-Gallery, and H2HMem-Multi benchmarks; results are from a single snapshot on September 25, 2026, and do not account for long-term memory evolution or personalized answering performance.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 07:26:48 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeie5pnthip36xqstb4trd4kyfmluqnrzhqctduoumluvlc25oso2li