Descriptive methods and analyses of neural population activity (Mikio Aoi - Princeton University)

Recent developments in neural recording technologies have dramaticallyincreased the number of neurons that can be simultaneously observedduring behavioral experiments. Concurrently, complexity inexperimental environments, and in the tuning properties of individualneurons, have conspired to make the task of summarizing neuronalpopulation activity both conceptually and computationally oneroususing conventional methods. Despite heroic efforts on the part of boththe systems neuroscience and machine learning communities, littleconsensus has been reached on the best ways to summarize these data.In this talk I will demonstrate new analysis techniques developed tomeet some of these challenges. I will highlight a case of neuralpopulation activity from monkey prefrontal cortex (PFC) during acontext-dependent decision making task. The analysis method decomposespopulation activity into a combination of low-dimensionalrepresentations that carry specific task-related information. Specialcases of this model include standard dimension-reduction techniquessuch as principal components and factor analysis. This decompositionreveals a multi-dimensional, dynamic code for decision, context, andstimulus information. I will describe how these observationssubstantially alter our understanding of decision-related populationactivity in PFC and provide a glimpse of how this basic approach issuited to analysis of neural population activity for other brainareas, tasks, animals, and recording modalities with the goal ofestablishing a suite of methods what will be considered the newstandard for analyzing these data. I will conclude with a few examplesof ongoing work where I am developing a variety of data analysis toolsto better understand brain dynamics.

When 4:00 pm to 5:00 pm on Tuesday, February 4, 2020
Location MCS B39, 111 Cummington Mall