representation-similarity

A Python Toolbox for Multimode Neural Data Representation Analysis - A Representational Analysis Toolbox for Neuroscience, including Representational Similarity Analysis (RSA), & Inter-Subject Correlation (ISC)

CKA (Centered Kernel Alignment) implemented in PyTorch

Tutorial on representational similarity analysis for MIND 2018

Analysis and investigating the confounding effect of accents in end-to-end Automatic Speech Recognition models.

Predict how well a transfer-learning dataset will work before fine-tuning on it. TLDChoiceNet cuts prediction MSE 5x, and an unsupervised class-correlation metric explains fine-tune accuracy with R2=0.97. Stanford CS 330.

Predict how well a transfer-learning dataset will work before spending compute on fine-tuning.

A theoretical and empirical framework for measuring when two self-supervised learning views can effectively share a common linear encoder.

Representation Similarity Analysis (RSA) to measure the similarity of the feature maps extracted by two models.

From Representation Change to Repair: does layer-wise representation change predict where a budget-constrained intervention actually restores accuracy in vision models (DeiT-S, ImageNet-C)?