English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Reinforcement learning agents often forget details of the past, especiallyafter delays or distractor tasks. Agents with common memory architecturesstruggle to recall and integrate across multiple timesteps of a past event, oreven to recall the details of a single timestep that is followed by distractortasks. To address these limitations, we propose a Hierarchical Chunk AttentionMemory (HCAM), which helps agents to remember the past in detail. HCAM storesmemories by dividing the past into chunks, and recalls by first performinghigh-level attention over coarse summaries of the chunks, and then performingdetailed attention within only the most relevant chunks. An agent with HCAM cantherefore "mentally time-travel" -- remember past events in detail withoutattending to all intervening events. We show that agents with HCAMsubstantially outperform agents with other memory architectures at tasksrequiring long-term recall, retention, or reasoning over memory. These includerecalling where an object is hidden in a 3D environment, rapidly learning tonavigate efficiently in a new neighborhood, and rapidly learning and retainingnew object names. Agents with HCAM can extrapolate to task sequences muchlonger than they were trained on, and can even generalize zero-shot from ameta-learning setting to maintaining knowledge across episodes. HCAM improvesagent sample efficiency, generalization, and generality (by solving tasks thatpreviously required specialized architectures). Our work is a step towardsagents that can learn, interact, and adapt in complex and temporally-extendedenvironments.

Towards mental time travel: a hierarchical memory for reinforcement learning agents