Evolving Culture vs Local Minima

TL;DR

This paper presents a theory linking the difficulty of learning in deep neural architectures to human culture and language, suggesting that social interactions and language facilitate the learning of complex high-level abstractions.

Contribution

It introduces a novel theory connecting deep learning challenges with cultural evolution and language, providing a new perspective on human cognitive development.

Findings

Learning in humans is hindered by local minima in neural representations.

Language and social cues help humans overcome learning difficulties in high-level abstractions.

Cultural evolution accelerates the development of complex mental representations.

Abstract

We propose a theory that relates difficulty of learning in deep architectures to culture and language. It is articulated around the following hypotheses: (1) learning in an individual human brain is hampered by the presence of effective local minima; (2) this optimization difficulty is particularly important when it comes to learning higher-level abstractions, i.e., concepts that cover a vast and highly-nonlinear span of sensory configurations; (3) such high-level abstractions are best represented in brains by the composition of many levels of representation, i.e., by deep architectures; (4) a human brain can learn such high-level abstractions if guided by the signals produced by other humans, which act as hints or indirect supervision for these high-level abstractions; and (5), language and the recombination and optimization of mental concepts provide an efficient evolutionary recombination operator, and this gives rise to rapid search in the space of communicable ideas that help humans build up better high-level internal representations of their world. These hypotheses put together imply that human culture and the evolution of ideas have been crucial to counter an optimization difficulty: this optimization difficulty would otherwise make it very difficult for human brains to capture high-level knowledge of the world. The theory is grounded in experimental observations of the difficulties of training deep artificial neural networks. Plausible consequences of this theory for the efficiency of cultural evolutions are sketched.