Captum: A unified and generic model interpretability library for PyTorch

TL;DR

Captum is an open-source, versatile library for interpreting PyTorch models, supporting various attribution algorithms, modalities, and evaluation metrics, with an interactive visualization tool for model debugging.

Contribution

It introduces a unified, extensible interpretability library for PyTorch models, including a visualization tool, supporting multiple data modalities and scalable computations.

Findings

Supports multiple data modalities including images, text, audio, video

Provides scalable, memory-efficient attribution algorithms

Includes an interactive visualization tool for model debugging

Abstract

In this paper we introduce a novel, unified, open-source model interpretability library for PyTorch [12]. The library contains generic implementations of a number of gradient and perturbation-based attribution algorithms, also known as feature, neuron and layer importance algorithms, as well as a set of evaluation metrics for these algorithms. It can be used for both classification and non-classification models including graph-structured models built on Neural Networks (NN). In this paper we give a high-level overview of supported attribution algorithms and show how to perform memory-efficient and scalable computations. We emphasize that the three main characteristics of the library are multimodality, extensibility and ease of use. Multimodality supports different modality of inputs such as image, text, audio or video. Extensibility allows adding new algorithms and features. The library is also designed for easy understanding and use. Besides, we also introduce an interactive visualization tool called Captum Insights that is built on top of Captum library and allows sample-based model debugging and visualization using feature importance metrics.