Felipe Parodi

Research

I study intelligence by combining neuroscience, machine learning, and ethology. My work follows a deliberate arc: understand natural social cognition with quantitative rigor, build measurement tools that bridge biological and artificial systems, then use those tools to explain how brains compute social meaning – and what that might teach us about building AI that understands the social world.


Natural social cognition
Grounding

The Primate Brain During Natural Social Behavior

Testard & Tremblay et al., 2024 · Nature

Quantitative analysis of neural population activity during naturalistic primate social behavior.

This work established a foundation for studying primate social cognition under naturalistic conditions, combining ethological annotation, computer vision, and wireless neural recording in unrestrained dyads.

I led the behavioral quantification pipeline, producing standardized behavioral event streams (labeling/QC and time alignment) used for downstream single-neuron and population analyses.

PrimateFace
Tool-Building

PrimateFace: Resource for Cross-Species Face Analysis

Parodi et al., 2025 · bioRxiv

A cross-species primate face dataset and benchmark for measuring generalization in facial analysis models.

To connect animal behavior to modern CV, I built PrimateFace as an evaluation substrate: a curated dataset spanning multiple primate species, with standardized training/evaluation infrastructure and systematic comparisons across model families (CNNs, transformers, vision-language models).

PrimateFace is genus-balanced to make cross-species generalization measurable rather than incidental. In practice it supports (i) individual identification/verification for longitudinal behavioral studies, (ii) robust face embeddings for tracking and re-identification, and (iii) pretraining/evaluation for downstream models used in primate video analysis.

mid-STS neural computation
Synthesis

Neural Basis of Social Intelligence

PhD Thesis · mid-STS

Integrating deep behavioral modeling with wireless neural recordings to study social intelligence.

My thesis brings these threads together: I use deep learning to extract behavioral representations (pose, kinematics, interaction structure, behavioral syllables) to bridge complex social behavior and neural activity in macaque mid-superior temporal sulcus (STS).

These representations let me test mechanistic hypotheses with single-unit and population methods, including dimensionality reduction, time-resolved analyses, and RL-based modeling.


All Publications