How to Shift Bias: Lessons from the Baldwin Effect

TL;DR

This paper explores how the Baldwin effect informs the design of bias-shifting algorithms in machine learning, suggesting starting with weak bias and gradually increasing it for better learning outcomes.

Contribution

It demonstrates the relevance of the Baldwin effect to bias-shifting algorithms and proposes a strategy of gradually increasing bias in learning algorithms.

Findings

Starting with weak bias and gradually increasing it improves learning.

The Baldwin effect provides insights into bias adaptation.

Explicit models illustrate bias shift strategies.

Abstract

An inductive learning algorithm takes a set of data as input and generates a hypothesis as output. A set of data is typically consistent with an infinite number of hypotheses; therefore, there must be factors other than the data that determine the output of the learning algorithm. In machine learning, these other factors are called the bias of the learner. Classical learning algorithms have a fixed bias, implicit in their design. Recently developed learning algorithms dynamically adjust their bias as they search for a hypothesis. Algorithms that shift bias in this manner are not as well understood as classical algorithms. In this paper, we show that the Baldwin effect has implications for the design and analysis of bias shifting algorithms. The Baldwin effect was proposed in 1896, to explain how phenomena that might appear to require Lamarckian evolution (inheritance of acquired characteristics) can arise from purely Darwinian evolution. Hinton and Nowlan presented a computational model of the Baldwin effect in 1987. We explore a variation on their model, which we constructed explicitly to illustrate the lessons that the Baldwin effect has for research in bias shifting algorithms. The main lesson is that it appears that a good strategy for shift of bias in a learning algorithm is to begin with a weak bias and gradually shift to a strong bias.