FoMo: A unifying theory of visual foraging

Visual foraging lies at the intersection of visual perception, decision-making and action planning. An attractive feature of this paradigm is that it generates a rich stream of sequential decision data. However, this presents a number of challenges for analysis. To this end, we have developed FoMo, a robust and flexible generative model for spatial-sequential data that allows prediction of participants’ selection behaviour on a target-by-target basis. Building upon our initial work, we present an updated version of FoMo (Clarke AD, Hunt AR, Hughes AE. Foraging as sampling without replacement: A Bayesian statistical model for estimating biases in target selection. PLOS Computational Biology. 2022;18(1):e1009813.)‌‌, which incorporates spatial structure allowing us to model organised spatial behaviours. FoMo provides estimates of a range of interpretable parameters, meaning we can use it to understand the causes of behavioural differences: for example, incorporating spatial-structure parameters improves model prediction accuracy for a number of visual foraging datasets, predominantly due to improvements for a subset of participants who use grid-following strategies. Our approach can also account for individual differences across the wide range of descriptive statistics that have previously been used to explore human and non-human animal behaviour, providing a unified framework for analysing these data.Author summary: Foraging is an important behaviour for humans and other animals, helping them to find food, habitats, and mates. It can involve searching both through visual space (such as searching for berries on a bush) and abstract space (such as when retrieving a word from memory). Here, we update our computational model of visual foraging behaviour (FoMo) which allows us to predict the behaviour of foragers in tasks where they must sequentially collect multiple targets. In particular, we allow the model to predict organised spatial behaviours, such as using a ‘reading-like’ strategy where participants search systematically in horizontal directions. Importantly, this means that we can study individual differences in foraging strategies in a quantitative, principled manner. We also show that our model can act as a unifying framework for visual foraging research, as it is able to predict a wide range of different descriptive statistics that have been used in previous work.