PNI Faculty Search Seminars

Abstract:
From a quick glance, the touch of an object, or a brief sound snippet, our brains construct scene representations composed of rich and detailed shapes and surfaces. These representations are not only the targets of perception, but also support aspects of cognition including reasoning about physics of objects, planning actions, and manipulating objects -- as in the paradigmatic case of using or making tools. A longstanding view in the psychology of perception and cognition holds that in order to compute such rich representations, the brain must draw on internal causal models of the outside physical world. How does the brain build and use such causal models of the world? In this talk, I will begin to answer this question by presenting a novel approach that synthesizes a diverse range of tools including generative models, simulation engines, and deep neural networks. In one key high-level visual capacity, I will show that this approach explains both human behavioral data and multiple stages of neural processing in non-human primates, as well as a classic illusion, the "hollow face" effect. In addition to perception of faces, I will also show that this approach can naturally extend to reverse-engineering neural computations in other domains of perception supporting high-level vision more generally, multisensory perception, and aspects of cognition beyond perception such as intuitive physical reasoning and understanding goal-directed action.

Abstract:
Abstraction can be defined as a cognitive process that finds a common feature - an abstract variable, or concept - shared by a number of examples. Knowledge of an abstract variable enables generalization, which in turn allows one to apply inference to new examples based upon old ones. Neuronal ensembles could represent abstract variables by discarding all information about specific examples, but this allows for representation of only one variable. Here we show how to construct neural representations that encode multiple abstract variables simultaneously, and we characterize their geometry. Representations conforming to this geometry were observed in dorsolateral pre-frontal cortex, anterior cingulate cortex, and the hippocampus in monkeys performing a serial reversal-learning task. These neural representations allow for generalization, a signature of abstraction, and similar representations are observed in a simulated multi-layer neural network trained with back-propagation. These findings provide a novel framework for characterizing how different brain areas represent abstract variables, which is critical for flexible conceptual generalization and deductive reasoning.

Abstract:
In order to successfully survive and reproduce, animals must produce a diverse range of innate and learned behaviors in a flexible and context-dependent manner. The computational task of forming an internal representation of an animal’s environment and translating that to the selection of goal-directed actions is dependent on the coordinated activity of multiple brain areas. In this talk, I will show how analysis and modeling of neural activity from different neural circuits can reveal how an animal’s external and internal environments are represented in diverse brain areas, providing insight into how diverse computations are performed by the brain. First, by modeling neural representations in the mushroom body of the Drosophila olfactory system, I find that similarities between odor representations impose constraints on the fly’s capacity for associative learning. These constraints can be overcome by a certain class of model learning rules, which make concrete predictions of neural dynamics during learning. Second, I will show how microendoscopic imaging of hypothalamic circuits in mice has revealed a neural substrate for a pronounced learned component of mouse social behavior, as well as a new model for the distributed regulation of animal behavior by multiple subcortical nuclei.

---