Gradient-Based Attribution Methods

By Marco Ancona, Enea Ceolini, Cengiz Oztireli, Markus H. Gross. 2019

191 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety, Robustness & Interpretability

Citations in the corpus

Quantifying Attention Flow in Transformers (2020, methodology): “Furthermore, we can come up with a new method that adjusts the attention weights using gradient-based attribution methods (Ancona et al., 2019).”
A comprehensive taxonomy for explainable artificial intelligence: a systematic survey of surveys on methods and concepts (2021, methodology): “Task visual tasks [117,118,99,7,2,84,147] reinforcement learning [107,65]”
Towards Explainable Artificial Intelligence (2019, methodology): “Explanation methods relying on the gradient of the function of interest [2] have a long history in machine learning.”
Explainable AI: Interpreting, Explaining and Visualizing Deep Learning (2019, methodology): “Earlier work by [2] explored whether disclosures provided in financial reports are more aptly described as communication or manipulation.”
Multi-Modal Knowledge Graph Construction and Application: A Survey (2022, methodology): “Saliency-based methods treat the marginal effects [63] of pixels to a given phrase by gradient computation as the heatmap value, and attention-based methods treat the cross-modal relevance as the heatmap value.”
Dissecting Recall of Factual Associations in Auto-Regressive Language Models (2023, methodology): “We distinguish between sensitivity and saliency (Ancona et al., 2019; Bastings and Filippova, 2020): methods such as Gradient-L2 (Li et al., 2016) show to which inputs the model is sensitive , i.e., where a small change in the input would make a large change in the output, but a method such as…”