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

Since reward functions are hard to specify, recent work has focused onlearning policies from human feedback. However, such approaches are impeded bythe expense of acquiring such feedback. Recent work proposed that agents haveaccess to a source of information that is effectively free: in any environmentthat humans have acted in, the state will already be optimized for humanpreferences, and thus an agent can extract information about what humans wantfrom the state. Such learning is possible in principle, but requires simulatingall possible past trajectories that could have led to the observed state. Thisis feasible in gridworlds, but how do we scale it to complex tasks? In thiswork, we show that by combining a learned feature encoder with learned inversemodels, we can enable agents to simulate human actions backwards in time toinfer what they must have done. The resulting algorithm is able to reproduce aspecific skill in MuJoCo environments given a single state sampled from theoptimal policy for that skill.

Learning What To Do by Simulating the Past