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

Recent breakthroughs in AI have shown the remarkable power of deep learningand deep reinforcement learning. These developments, however, have been tied tospecific tasks, and progress in out-of-distribution generalization has beenlimited. While it is assumed that these limitations can be overcome byincorporating suitable inductive biases, the notion of inductive biases itselfis often left vague and does not provide meaningful guidance. In the paper, Iarticulate a different learning approach where representations do not emergefrom biases in a neural architecture but are learned over a given targetlanguage with a known semantics. The basic ideas are implicit in mainstream AIwhere representations have been encoded in languages ranging from fragments offirst-order logic to probabilistic structural causal models. The challenge isto learn from data the representations that have traditionally been crafted byhand. Generalization is then a result of the semantics of the language. Thegoals of this paper are to make these ideas explicit, to place them in abroader context where the design of the target language is crucial, and toillustrate them in the context of learning to act and plan. For this, after ageneral discussion, I consider learning representations of actions, generalpolicies, and subgoals ("intrinsic rewards"). In these cases, learning isformulated as a combinatorial problem but nothing prevents the use of deeplearning techniques instead. Indeed, learning representations over languageswith a known semantics provides an account of what is to be learned, whilelearning representations with neural nets provides a complementary account ofhow representations can be learned. The challenge and the opportunity is tobring the two together.

Target Languages (vs. Inductive Biases) for Learning to Act and Plan