Activation patching studies the model's computation by altering its latent representations, the token embeddings in transformer-based language models, during the inference process

The discrete analogue of the Gabor transform is obtained by choosing the temporal window function in a time-frequency analysis as the discrete analogue of the Gaussian kernel.

discgabor

A time-causal and time-recursive analogue of the Gabor transform

Patching

Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Models

TextGrad is a powerful framework building automatic ``differentiation'' via text. TextGrad implements backpropagation through text feedback provided by LLMs, strongly building on the gradient metaphor

Source	Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Models
Year
Data Source	Papers with Code archive (CC BY-SA)