New research explores advanced memory systems for LLM agents · 3 sources tracked
ByPulseAugur Editorial·[18 sources]·
Three new research papers explore advancements in memory systems for large language model (LLM) agents. The first paper, 'HasMem,' introduces a method for adaptive memory softening to improve long-term recall and context compression, achieving high F1 scores on reconstruction probes and LongMemEval-S. The second paper, 'MemHarm,' focuses on the severity of memory attacks, formalizing 'counterfactual memory regret' and demonstrating that optimizing for severity, rather than just success, leads to greater downstream loss. The third paper, 'Retrieved but Not Delivered,' investigates the 'delivery' stage of multimodal memory, proposing 'DeliverMem' to optimize how retrieved information reaches the model, significantly improving accuracy on benchmarks like MemLens and DMV-Bench.
AI
IMPACT
These advancements in memory management and delivery could lead to more capable and persistent LLM agents for long-horizon tasks and personalized interactions.
RANK_REASON
Three distinct academic papers detailing novel methods for LLM agent memory systems.
arXiv:2609.36595v1 Announce Type: cross Abstract: Visual-memory systems commonly retain or compress past observations. Robot control additionally requires interaction-derived state that no individual frame may explicitly represent, such as persistent identity relations, accumulat…
arXiv:2609.37930v1 Announce Type: new Abstract: Persistent textual memory allows language models to carry information across long interactions, but learning what to remember is fundamentally a credit-assignment problem. A memory rewrite may only become useful many steps later, wh…
arXiv:2609.36130v1 Announce Type: new Abstract: Long-running LLM agents compress past interactions into persistent memories that may be reused as premises for later tasks. This creates a distinct derivation problem: whether the memory actually follows from what the interaction hi…
arXiv:2609.36805v1 Announce Type: new Abstract: Large language model (LLM) agents reuse external memory to guide new tasks, but effective retrieval requires learning which memory sets improve execution. Such learning relies on costly outcome feedback: ordinary retrieval observes …
arXiv cs.AI
TIER_1English(EN)·Haohao Qu, Yongcheng Jing, Chun Hin Chan, Shanru Lin, Wenqi Fan, Dacheng Tao·
arXiv:2609.37311v1 Announce Type: new Abstract: Recent Recommendation Agents (RecAgents) offer a promising alternative by shifting recommendation to an active, user-side paradigm, where generative agents autonomously perceive external platforms, reason over user preferences, and …
arXiv:2609.37544v1 Announce Type: new Abstract: Content-generation agents continuously receive impressions, clicks, conversions, and negative feedback from recommendation systems, providing real-world outcome signals for memory evolution. However, these signals are delayed and no…
arXiv:2609.37590v1 Announce Type: new Abstract: LLM agents accumulate interaction histories that grow linearly with task length, causing quadratic inference cost scaling and performance degradation from attention dilution. Existing context-compression methods learn what to discar…
arXiv:2609.35808v1 Announce Type: cross Abstract: Experience reuse can reduce repeated exploration in embodied agents, but a trajectory that succeeded previously may be unsuitable for the current execution context. Existing memory systems pri marily optimize construction and retr…
Recent Recommendation Agents (RecAgents) offer a promising alternative by shifting recommendation to an active, user-side paradigm, where generative agents autonomously perceive external platforms, reason over user preferences, and execute decisions. However, existing RecAgents s…
arXiv:2609.30797v1 Announce Type: new Abstract: Text-based memory and context compression support reuse of past interactions. Resizing continuous memory changes the input to a frozen LLM, coupling capacity allocation with readout. We propose Hard-Origin Adaptively Softened Memory…
Agents can learn from past executions, but enabling different agents to reuse and build on one another's experience remains challenging. We introduce EpiCon, a shared multimodal memory framework for agent collective learning without updating host model parameters. EpiCon links qu…
Long-form video understanding requires multimodal agents to iteratively gather evidence over many reasoning steps. However, most existing agentic methods suffer from semantic thrashing: as append-only working memory grows, attention to key evidence collapses, and the agent loses …
As LLM agents increasingly rely on persistent memory for long-horizon and personalized behavior, they can retain and reuse information across interactions, but this also creates a lasting channel through which malicious memory writes can influence future behavior. Persistent-memo…
Work on memory for multimodal agents optimizes what is written, updated and retrieved. Between retrieval and the answer, however, is a stage that multimodal memory evaluations do not isolate: what of the retrieved memory reaches the model, and in what form. We call it delivery, a…
arXiv:2609.38119v1 Announce Type: new Abstract: Long-form video understanding requires multimodal agents to iteratively gather evidence over many reasoning steps. However, most existing agentic methods suffer from semantic thrashing: as append-only working memory grows, attention…
arXiv:2609.36906v1 Announce Type: new Abstract: Reliable embodied decisions under partial observability require informative observations and sufficient supporting evidence. However, semantic scores alone do not reveal which viewpoints justify a claim or where additional evidence …
arXiv cs.CV
TIER_1English(EN)·Ziyun Zeng, Hang Hua, Shaden Alshammari, Rogerio Feris, William T. Freeman, Jiebo Luo·
arXiv:2609.37923v1 Announce Type: new Abstract: Agents can learn from past executions, but enabling different agents to reuse and build on one another's experience remains challenging. We introduce EpiCon, a shared multimodal memory framework for agent collective learning without…
