I am a postdoctoral fellow in the Jazayeri lab at MIT, where I study how the brain supports flexible generalization abilities, specifically related to physical inference abilities using neurophysiology in monkeys, behavior, and computational (neural network) models. Prior to this, I received my Ph.D. from MIT; my Ph.D. work focused on the neural basis of visual object recognition. I used pharmacological and optogenetic perturbation methods to test the causal role of the primate ventral stream in object recognition abilities, as well behavioral benchmarks to compare object recognition behavior between deep neural networks and primates. My undergraduate training is in electrical engineering and I am excited by the confluence of quantitative models and benchmarks with rich neurophysiological measurements.
Principal Investigator: Mehrdad Jazayeri
Fellow: Clara Melhem
From a few glances of a visual scene, primates can make many rich physical inferences, such as how heavy an object is, if/where it might fall, how to pick it up, etc.. Such inferences are thought to be supported by “mental simulations” of internal models, but it remains unclear if/how the brain might implement such computations. To study this, we trained macaque monkeys on a simple video game where the goal is to intercept a moving ball in the face of occlusion. We then recorded from thousands of neurons in a candidate area: the dorsomedial frontal cortex (DMFC). Our results suggest that monkeys “mentally simulate” the ball, and that DMFC activity reflects this simulation. The proposed project would use sophisticated population analyses, including recurrent neural network models, to investigate how this neural representation is distributed over DMFC, along both spatial (topographic and columnar organization) and temporal (neural dynamics) dimensions.