Applications of deep learning for primate neuroethology spatial, behavioral, and social coding in macaque mid-superior temporal sulcus Felipe Parodi

Format:

Book

Thesis/Dissertation

Author/Creator:

Parodi, Felipe, author.

Contributor:

University of Pennsylvania. Neuroscience., degree granting institution.

Language:

English

Subjects (All):

Neurosciences.

Computer science.

Animal sciences.

0317.

0984.

0475.

Local Subjects:

Neurosciences.

Computer science.

Animal sciences.

0317.

0984.

0475.

Genre:

Academic theses

Physical Description:

1 online resource (232 pages)

Contained In:

Dissertations Abstracts International 87-12B

Place of Publication:

Ann Arbor : ProQuest Dissertations and Theses, 2026

Language Note:

English

Summary:

How primate brains generate behavior in the real world remains largely unknown, in part because the tools to measure both behavior and neural activity during unconstrained action have not existed. This dissertation develops deep-learning-based computational infrastructure for naturalistic primate neuroscience and applies it to the mid-superior temporal sulcus (mSTS), a region defined by decades of restrained recording as a hub for social perception but never studied during free behavior. Emerging work across species has shown that neural activity in freely moving animals is intricately bound to movement, behavioral context, and social dynamics, yet mSTS has eluded this approach. Recording from sulcal cortex is technically demanding, but the deeper reason mSTS has not been studied in freely behaving primates is the field's conviction that experimental control is the only rigorous way to study the primate brain. To address this, we built an integrated experimental platform comprising a large open-field arena with 30-camera coverage, a deep-learning computer vision pipeline for markerless 3D pose reconstruction at sub-pixel precision, wireless depth-electrode arrays targeting both banks of mSTS, and unsupervised behavioral segmentation that extracts discrete behavioral syllables from continuous pose dynamics. In parallel, we developed PrimateFace, a cross-species facial analysis resource and dataset spanning 60+ primate genera, enabling automated face detection and landmark estimation from tarsiers to gorillas. We applied these tools to record from mSTS in macaques freely exploring a three-dimensional arena. During solitary behavior, mSTS firing rates were jointly modulated by spatial position, body kinematics, and behavioral state, with three-dimensional position explaining more unique variance than any measured visual feature group. The same behavior occupied distinct regions of population neural space at different spatial locations, revealing context-dependent encoding invisible to restrained paradigms. Simultaneous recordings from two macaques during social interaction revealed that mSTS encoded partner kinematics and relational variables carrying unique explanatory power beyond all self-related predictors, even when the recorded animal was stationary. Neural prediction error-the deviation of population activity from its own recent trajectory-was elevated when either animal changed behavioral state but not at dyadic interaction transitions, consistent with predictive monitoring at the individual level for both self and partner. Social encoding in mSTS emerged from behavioral context, not from dedicated circuitry. Together, these findings reinterpret mSTS as a region whose population activity represented self-related, other-related, and interaction-related variables beyond the observer-centric social perception role defined by restrained recording. More broadly, these discoveries demonstrate that computational approaches to natural behavior can reveal neural principles invisible to constrained paradigms, opening new directions for understanding how primate brains navigate the physical and social world

Notes:

Source: Dissertations Abstracts International, Volume: 87-12, Section: B.

Advisors: Platt, Michael L.; Kording, Konrad P. Committee members: Schmidt, Marc; Beauchamp, Michael; Daniilidis, Kostas; Seyfarth, Robert

Ph.D. University of Pennsylvania 2026

Vendor supplied data

Local Notes:

School code: 0175

ISBN:

9798247983491

Access Restriction:

Restricted for use by site license