Captum: A unified and generic model interpretability library for PyTorch

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.

References in corpus (2)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• SmoothGrad: removing noise by adding noise

Cited by in corpus (12)

• Gifsplanation via Latent Shift: A Simple Autoencoder Approach to Counterfactual Generation for Chest X-rays
• Speaker-Conditioned Hierarchical Modeling for Automated Speech Scoring
• Challenges for cognitive decoding using deep learning methods
• Visualizing Automatic Speech Recognition -- Means for a Better Understanding?
• AutoNLU: Detecting, root-causing, and fixing NLU model errors
• Bangla Text Classification using Transformers
• Quantifying Explainability in NLP and Analyzing Algorithms for Performance-Explainability Tradeoff
• A Review of Bangla Natural Language Processing Tasks and the Utility of Transformer Models
• Explaining Classes through Word Attribution
• Multi-concept adversarial attacks
• Explainability Requires Interactivity
• Attribution Mask: Filtering Out Irrelevant Features By Recursively Focusing Attention on Inputs of DNNs