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

Imitation learning practitioners have often noted that conditioning policieson previous actions leads to a dramatic divergence between "held out" error andperformance of the learner in situ. Interactive approaches can provably addressthis divergence but require repeated querying of a demonstrator. Recent workidentifies this divergence as stemming from a "causal confound" in predictingthe current action, and seek to ablate causal aspects of current state usingtools from causal inference. In this work, we argue instead that thisdivergence is simply another manifestation of covariate shift, exacerbatedparticularly by settings of feedback between decisions and input features. Thelearner often comes to rely on features that are strongly predictive ofdecisions, but are subject to strong covariate shift.  Our work demonstrates a broad class of problems where this shift can bemitigated, both theoretically and practically, by taking advantage of asimulator but without any further querying of expert demonstration. We analyzeexisting benchmarks used to test imitation learning approaches and find thatthese benchmarks are realizable and simple and thus insufficient for capturingthe harder regimes of error compounding seen in real-world decision makingproblems. We find, in a surprising contrast with previous literature, butconsistent with our theory, that naive behavioral cloning provides excellentresults. We detail the need for new standardized benchmarks that capture thephenomena seen in robotics problems.

Feedback in Imitation Learning: The Three Regimes of Covariate Shift