Understanding how people generalize and extrapolate from limited amounts of data remains an outstanding challenge. We study this question in the domain of scalar function learning, and propose a simple model based on the Principle of Maxi- mum Entropy (Jaynes, 1957). Through computational model- ing, we demonstrate that the theory makes two specific predic- tions about peoples’ extrapolation judgments, that we validate through experiments. Moreover, we show that existing Gaus- sian Process models of function learning cannot account for these effects.

The ability to learn and predict simple functions is a key aspect of human intelligence. Recent works have started to explore this ability using transformer architectures, however it remains unclear whether this is sufficient to recapitulate the extrapolation abilities of people in this domain. Here, we propose to ad- dress this gap by augmenting the transformer architecture with two simple inductive learning biases, that are directly adapted from recent models of abstract reasoning in cognitive science. The results we report demonstrate that these biases are helpful in the context of large neural network models, as well as shed light on the types of inductive learning biases that may contribute to human abilities in extrapolation.

In a classic paper, Miller (1956) summarized findings showing that people can only identify a limited number of distinctstimuli at a time. One puzzling aspect of this capacity limitation is that it is approximately invariant to range. Thatis, the number of accurately identifiable stimuli is approximately the same regardless of how far apart the stimuli arespaced. Models of this phenomenon have suggested that people operate in a context-coding mode when performing thesetasks, effectively carrying out a form of contextual normalization, but why such normalization might take place is unclear.Here, we propose an explanation by appealing to a tradeoff with generalization. Specifically, we implement contextualnormalization in a recurrent neural network and show that this normalization enables stronger generalization in a relationalreasoning task, but also results in a perceptual capacity limitation which captures many of these classic phenomena.