Attention and perception

Nonlinear probability weighting allows cumulative prospect theory (CPT) to account for key phenomena in decision making under risk (e.g., certainty effect, fourfold pattern of risk attitudes). It describes the impact of risky outcomes on preferences in terms of a rank-dependent nonlinear transformation of their objective probabilities. The attentional Drift Diffusion Model (aDDM) formalizes the finding that attentional biases toward an option can shape preferences within a sequential sampling process. Here we link these two influential frameworks. We used the aDDM to simulate choices between two options while systematically varying the strength of attentional biases to either option. The resulting choices were modeled with CPT. Changes in preference due to attentional biases in the aDDM were reflected in highly systematic signatures in the parameters of CPT's weighting function (curvature, elevation). Based on these insights, we predicted and demonstrated—in a re-analysis of a large set of previously published empirical data—that attentional biases are also systematically linked to patterns in probability weighting empirically. These findings highlight that distortions in probability weighting can arise from simple option-specific attentional biases in information search, and suggest an alternative to common interpretations of weighting-function parameters in terms of probability sensitivity and optimism. They also point to novel, attention-based explanations for empirical phenomena associated with characteristic shapes of CPT's probability-weighting function (e.g., certainty effect, description–experience gap). The results advance the integration of two prominent computational frameworks for decision making.

For better or worse, humans live a resource constrained existence; for example, only a fraction of the sensations our body experiences ever reach conscious awareness, and we store a shockingly small subset of these experiences in short-term memory for later use. Despite these observations, most theories of learning assume that, given feedback about a new experience, our representations are updated so as to minimize subsequent errors with minimal consideration of cognitive capacity constraints. Acknowledging that human cognition has clear biological limitations, we explored the degree to which human learning could be better described with sets of biases toward simpler and more parsimonious mental representations (i.e., simplicity biases) relative to an error-driven, accuracy-maximizing normative model. Taking the normative model as a basis, we developed a suite of computational models that use various mechanistic simplicity biases to explain learning. We fit these models to four data sets that varied in the type of learning needed to achieve high accuracy. Across all data sets, we found consistent evidence that the best descriptors of human learning were models with mechanisms that instantiated a constrained optimization process, where errors were minimized subject to constraints on both attention and memory. Importantly, whereas normative models failed to account for patterns of attentional deployment over time, models with simplicity biases accounted well for both choice responses and gaze fixation data as participants learned various tasks.

Here, we take a computational approach to understand the mechanisms underlying face perception biases in depression. Participants diagnosed with Major Depressive Disorder (MDD, N=30) and healthy controls (N=30) took part in a study involving recognition of identity and emotion in faces. We used signal detection theory (SDT) to determine whether any perceptual biases exist in depression aside from decisional biases. We found lower sensitivity to happiness in general, and lower sensitivity to both happiness and sadness with ambiguous stimuli. We found no systematic effect of depression on the perceptual interactions between face expression and identity, suggesting that depression is not associated with difficulty selectively attending to one of these dimensions. Our use of SDT allows us to link these psychophysical results to an neurocomputational model of the encoding of facial expression. We show through simulation that the overall pattern of results, as well as other biases found in the literature, can be explained by selective suppression of neural populations encoding positive expressions in MDD. In a second study, we used reverse correlation to show that one source of this suppression could be a difference between participants diagnosed with MDD and healthy controls in the information sampled in order to detect happiness and sadness in faces. We show that the psychophysical observer models obtained through reverse correlation offer a complementary way to account for the results of our first study. Our model-based approach is a step forward toward understanding the mechanisms underlying face perception biases in psychiatric disorders.

The study of attention dynamics in decision making has been increasingly important in uncovering the cognitive principles of choice behavior, including intertemporal choice. Recently, empirical work on this topic has suggested that the two attributes involved in intertemporal choice (monetary amounts and time delays) have distinct and independent influences on the choice process, and that these attributes are additively aggregated in an evidence accumulation process. In this paper we outline theoretical problems with such an account, and argue that intertemporal choice processes necessarily includes interaction between the two attributes in order to generate reasonable choice behavior. Furthermore, we re-examine existing eye-tracking datasets of intertemporal choice using a Markov model of attentional dynamics. Our model assumes that the transitions between distinct attentional states (e.g. amounts and delays of the two options) depend on a large number of variables, including, crucially, the most recently attended attribute value. We estimate model parameters within a hierarchical Bayesian framework and find that high values of currently sampled information lead to more frequent transitions to the other attribute within the same option. Thus, for example, participants are more likely to sample the time delay of an option when the monetary amount is high, relative to when the amount is low (and vice versa). This corresponds to interdependent and interactive attention dynamics during decision making. We conclude by examining how such an interactive attentional process can combine with an attention-based evidence accumulation process to generate observed patterns in intertemporal choice behavior.

Research on numerical cognition has suggested that there is compression in both, symbolic (e.g., Arabic numerals) and non-symbolic (e.g., dot clouds) number perception. More specifically, symbolic and non-symbolic numbers are supposed to be mapped onto the same compressed mental analogue representation. However, experiments using magnitude estimation tasks show logarithmic compression of symbolic numbers while the compression of non-symbolic numbers has a power-function shape. This warrants closer inspection at what differentiates the two processes. In this study, we hypothesized that estimates of symbolic numbers are systematically shaped by the format in which they are represented, namely the place value system. To investigate this, we tested adults (n = 188) on a repeated magnitude estimation task with unfamiliar base-26 and base-5 scales and fitted a hierarchical logarithmic, a hierarchical power and a hierarchical linear model to the data. Results revealed that adults showed systematic, logarithmic-looking underestimation on both scales, indicating that the place value system itself can cause the pattern. Additionally, the observed shape of participants’ estimates on both scales could be well-explained with a simple model that assumed insufficient understanding of exponential growth (i.e., a characteristic of place value systems). Taken together, our results suggest that the discrepancy between symbolic and non-symbolic number compression can be explained by taking the effect of the place-value system into account.

Our ability to discriminate short durations can be studied through Signal-Detection-Theory based models. They incorporate the sensory, decision, and response mechanisms that govern observers’ responses in duration discrimination tasks, and serve as a guide to test substantive hypotheses about each of these components. The standard version of these models states that the sensory mechanism relies solely on the magnitude of the difference in duration of the stimuli to be compared. This is incompatible with some empirical results, which have shown that psychometric functions change with the duration of the reference stimulus. These results have been attributed to some form of the scalar property of time perception, but they could also be produced by shifts in decisional criteria. Here we present a series of four models that incorporate the scalar property, decisional criteria that vary with stimulus duration, or both, along with the standard model. We show that each model gives rise to psychometric functions with distinct characteristics, which raises the question of whether these models are also distinguishable in practice. We tackle this question through a simulation study whose results show that parameters can be adequately recovered, and that the data-generating model can be correctly selected using goodness-of-fit procedures. This framework provides a solid ground to design experiments that allow testing how sensory and decisional mechanisms contribute to judgements in duration discrimination tasks.