Theoretical Neuroscience & Artificial Intelligence
My research broadly focuses on understanding the dynamics of neural circuits that subserve cognitive functions in both artificial and biological systems, by adopting problem-solving approaches from physics, mathematics, electrical engineering, and computer science.
Raju P. C. Geometric Stability: The Missing Axis of Representations. [link] [pdf] [code]
Raju P. C. The Geometric Canary: Predicting Steerability and Detecting Drift via Representational Stability. [link] [pdf] [code]
Raju P. C. The Geometric Alignment Tax: Tokenization vs. Continuous Geometry in Scientific Foundation Models. [link] [pdf] [code]
Raju P. C. Geometric coherence of single-cell CRISPR perturbations reveals regulatory architecture and predicts cellular stress. [link] [pdf] [code]
Barreiro, A. K., Fontenele, A. J., Cheng, L., Raju P. C., Gautam, S. H., & Shew, W. L. (2024). Sensory input to cortex encoded on low-dimensional periphery-correlated subspaces. PNAS Nexus 3(1), [link] [pdf] [si]
Golan, T., Raju P. C., & Kriegeskorte, N. (2020). Controversial stimuli: Pitting neural networks against each other as models of human cognition. Proceedings of the National Academy of Sciences, 117(47), 29330-29337 [link] [pdf] [si] [code]
Raju P. C. From Syntax to Semantics: Geometric Stability as the Missing Axis of Perturbation Biology. [link] [pdf] [code]
Raju P. C. (2026). Geometric Stability: The Missing Axis of Representations. Spotlight Award. Unpublished conference paper. 7th International Conference on the Mathematics of Neuroscience and AI. Rome, Italy.
Raju P. C. (2026). Geometric Phase Transition Enables Extreme Hippocampal Memory Capacity. Unpublished conference paper. 7th International Conference on the Mathematics of Neuroscience and AI. Rome, Italy.
Golan, T., Raju P. C., & Kriegeskorte, N. (2020). Controversial stimuli: adjudicating between deep neural network models of biological vision with synthetic images. Journal of Vision 20(11), 947 doi: 10.1167/jov.20.11.947 [link]
Golan, T., Raju P. C., & Kriegeskorte, N. (2020). Adjudicating between deep neural network models of biological vision with controversial stimuli. Unpublished conference paper. Computational and Systems Neuroscience (Cosyne). Denver, CO. (Poster III-53) [link] [poster]
Shesha: Self-Consistency Metrics for Representational Stability.
Explore shesha with these interactive notebooks (each takes < 5 minutes to run):
LLM Embeddings — Analyze embedding stability across layers and models using 
feature_split.
Steering Vectors — Compute steering vectors from contrastive pairs and measure their effectiveness and consistency.
Vision Models — Compare geometric stability and class separability across ResNets, ViTs, and other vision architectures.
Representational Drift — Measure drift caused by Gaussian noise injection and LoRA fine-tuning using rdm_drift.
Training Dynamics — Track geometric stability during model training to detect representation collapse or divergence.
CRISPR (Bio) — Use shesha.bio to analyze stability and effect sizes in single-cell CRISPR perturbation experiments.
I am always happy to help anyone looking to learn or looking to discuss anything related to neuroscience or AI.