The human brain excels at recognizing objects across modalities. In my research I seek to harness the neurocomputational principles that underlie these cognitive abilities by building brain-inspired cognitive models of attention and object recognition. My models have given insight into human cognition while also excelling at machine perception. Specifically, in this dissertation I describe four brain-inspired models. I first describe SUN, a saliency-based model of task-driven visual search. SUN uses independent component analysis (ICA) to learn image filters with properties that are qualitatively similar to simple cells in primary visual cortex, and it uses these filters to predict human eye movements. Then I describe NIMBLE, an approach to active object recognition using SUN and simulated eye movements. Subsequently, I describe Gnostic Fields, a brain-inspired model that attacks the universal stimulus recognition problem. Gnostic Fields achieve state-of-the-art performance on benchmark datasets for music, image, and odor classification. Lastly, I combine Gnostic Fields with a space-variant model of visual processing learned using ICA and assess the model's ability to classify objects and faces in images

To recognize objects, the human visual system processes information through a network of hierarchically organized brain regions. Many neurocomputational models have modeled this hierarchical structure, but they have often used hand-crafted features to model early visual areas. According to the linear efficient coding hypothesis, the goal of the early visual pathway is to capture the statistical structure of sensory stimuli, removing redundancy, and factoring the input into independent features. In this work, we use a hierarchical Independent Components Analysis (ICA) algorithm to automatically learn the visual features that account for early visual cortex. We then continue modeling the object recognition pathway using Gnostic Fields, a theory for how the brain does object categorization, in which brain regions devoted to classifying mutually-exclusive categories exist near the top of sensory processing hierarchies. The whole biologically-inspired model not only allows us to develop representations similar to those in primary visual cortex, but also to perform well on standard computer vision object recognition benchmarks.

Since Yarbus's seminal work, vision scientists have argued that our eye movement patterns differ depending upon our task. This has recently motivated the creation of multi-fixation pattern analysis algorithms that try to infer a person's task (or mental state) from their eye movements alone. Here, we introduce new algorithms for multi-fixation pattern analysis, and we use them to argue that people have scanpath routines for judging faces. We tested our methods on the eye movements of subjects as they made six distinct judgments about faces. We found that our algorithms could detect whether a participant is trying to distinguish angriness, happiness, trustworthiness, tiredness, attractiveness, or age. However, our algorithms were more accurate at inferring a subject's task when only trained on data from that subject than when trained on data gathered from other subjects, and we were able to infer the identity of our subjects using the same algorithms. These results suggest that (1) individuals have scanpath routines for judging faces, and that (2) these are diagnostic of that subject, but that (3) at least for the tasks we used, subjects do not converge on the same "ideal" scanpath pattern. Whether universal scanpath patterns exist for a task, we suggest, depends on the task's constraints and the level of expertise of the subject.

Theories for visually guided action account for online con-trol in the presence of reliable sources of visual information,and predictive control to compensate for visuo-motor delayand temporary occlusion. In this study, we characterize thetemporal relationship between information integration windowand prediction distance using computational models. Subjectswere immersed in a simulated environment and attempted tocatch virtual balls that were transiently “blanked” during flight.Recurrent neural networks were trained to reproduce subjectsgaze and hand movements during blank. The models success-fully predict gaze behavior within 3◦, and hand movementswithin 8.5 cm as far as 500 ms in time, with integration windowas short as 27 ms. Furthermore, we quantified the contributionof each input source of information to motor output throughan ablation study. The model is a proof-of-concept for predic-tion as a discrete mapping between information integrated overtime and a temporally distant motor output.