Cross-Category Information (CCI)

The Challenge: Signal vs. Noise

In neuroimaging, “noise” is often defined as trial-to-trial variability that doesn’t match the stimulus. However, much of this variability is shared across neuronal populations and may represent top-down feedback or internal predictions.

The Innovation: The CCI Metric

I invented the Cross-Condition Information (CCI) metric to quantify how much of this “noise” is actually structured and informative. With Manifold Learning (PCA) and Subspace Alignment, CCI measures the overlap between variation in neural representation spaces across objects from the same category (i.e. different animals).

Technical Implementation

• Manifold Alignment: Developed algorithms to align low-dimensional latent representations.
• Noise Decoding: Demonstrated that trial-by-trial variability is not stochastic but follows a specific geometric structure that predicts behavioral outcomes.
• High-Performance Computing: Optimized the pipeline in Python (NumPy/SciPy) to process multi-terabyte datasets across HPC clusters.

Scientific Impact

CCI provides a tool for researchers to interrogate how informative noise in the brain is for object classification in human vision.