Projects

In many tasks across the cognitive hierarchy, seemingly irrelevant context biases decision-making. This line of studies asks a simple question: Why can’t people be accurate? Why can’t they ignore information that should not affect their decisions? Using computational modeling, we show that biases can be understood as a consequence of optimal inference under uncertainty. In particular, the “demixing model” suggests that biases arise as we try to disentangle the mixed signals from two (or more) items, and it predicts both attraction and repulsion effects.

One of the driving questions in vision science is how internal models of the world relate to the world itself. Quite often, answers to this question are pursued by studying single stimuli (e.g., a green circle) or sets of similar stimuli (a set of green circles). In contrast, this research line studies how people encode perceptual ensembles (such as fields of grass or differently colored marbles). It uses an experimental paradigm based on priming of pop-out in visual search to study ensembles with different probability density functions and to analyze the internal representations corresponding to these functions.

Most cognitive acts can be treated as prediction or inference processes. This line asks what role affect plays in these processes and why positive or negative experience is needed. The central hypothesis, tested across multiple studies, is that affect provides feedback on prediction accuracy weighted by prior probability. One corollary is that errors should yield negative affect proportional to the evidence in favor of the correct response. Crucially, this is not about complex choices (“Should I buy this car?”), but about errors in simple and largely automatic processing (“Did I see this face before?”).