Imprint: Online Memory Compression for Long-Horizon Egocentric QA

2026-07-01

AI Summary

This work addresses the challenge of balancing retrieval efficiency and scalability in memory representation for long-horizon first-person visual question answering. The authors propose an interaction-centric online memory compression framework that eschews conventional textual summarization in favor of structured logging of explicit interaction units. Inspired by human memory consolidation mechanisms, the approach dynamically organizes and compresses memory based on repetition, recency, and uniqueness to construct a compact, retrieval-oriented memory bank. Leveraging large language models for reasoning, the method achieves 35.8% question-answering accuracy on the EgoLifeQA benchmark, yielding a sixfold increase in evidence-supported answers while reducing memory consumption by 2.3× and decreasing retrieval latency by 11.8×.

Abstract

Long-horizon egocentric question answering involves answering about events that have occurred hours or days in the past. This requires memory representations that remain both retrieval-effective and scalable over days or weeks of recording. Existing long-horizon egocentric QA methods construct memory as hierarchical textual summaries of observations. While effective for reducing memory size, summarization optimizes for descriptive compression rather than retrieval: repeated interactions are absorbed into coarse textual descriptions instead of being preserved as explicit, recurring memory units, making long-horizon evidence aggregation difficult. We propose Imprint, an interaction-centric memory framework that formulates long-horizon egocentric memory as an online memory compression problem rather than summarization. Incoming observations are first represented as structured Interaction Records and continuously organized into recurring interaction patterns. Using human memory consolidation signals of recurrence, recency, and distinctiveness, Imprint selectively retains and compresses interactions into a compact retrieval-oriented memory. We evaluate Imprint on EgoLifeQA, a seven-day egocentric benchmark containing questions that require reasoning over interactions occurring hours to days before the query. With the same LLM, Imprint improves QA accuracy from 31.0% to 35.8%, increases evidence-grounded answers by $6\times$ compared with EgoRAG, reduces memory footprint by $2.3\times$, and decreases retrieval latency by $11.8\times$. These results demonstrate that memory compression provides a scalable and retrieval-effective foundation for long-horizon egocentric question answering.

Problem

• long-horizon egocentric QA
• memory compression
• retrieval effectiveness
• interaction-centric memory
• evidence aggregation

Innovation

• online memory compression
• interaction-centric memory
• long-horizon egocentric QA
• memory consolidation
• structured interaction records