Adaptive Sampling Policies Imply Biased Beliefs: A Generalization of the Hot Stove Effect

arXiv:2404.02591v1 Announce Type: new
Abstract: The Hot Stove Effect is a negativity bias resulting from the adaptive character of learning. The mechanism is that learning algorithms that pursue alternatives with positive estimated values, but avoid alternatives with negative estimated values, will correct errors of overestimation but fail to correct errors of underestimation. Here, we generalize the theory behind the Hot Stove Effect to settings in which negative estimates do not necessarily lead to avoidance but to a smaller sample size …

abstract algorithms arxiv bias cs.lg errors hot imply negative policies positive sampling stat.ml type values will