Understanding Black-box Predictions via Influence Functions · HackerPapers discussions

Understanding Black-box Predictions via Influence Functions — mentioned by gradys

Tue, 21 Mar 2023 20:26:07 GMT

https://arxiv.org/abs/1703.04730 > How can we explain the predictions of a black-box model? In this paper, we use influence functions -- a classic technique from robust statistics -- to trace a model's prediction through the learning algorithm and back to its training data, thereby identifying training points most responsible for a given prediction.

Paper and discussion history. A mention is not an endorsement.

Understanding Black-box Predictions via Influence Functions — mentioned by version_five

Sun, 05 Feb 2023 23:31:47 GMT

Your idea is the gold standard in explaining the influence of training data. People may be interested in this paper and more modern variations: https://arxiv.org/abs/1703.04730 It attempts to do as you suggest in a tractable way, to understand which training data is most influential.

Paper and discussion history. A mention is not an endorsement.

Understanding Black-box Predictions via Influence Functions — mentioned by andy99

Sat, 17 Jul 2021 00:17:46 GMT

Related to understanding how ML models are making their decisions is the concept of influence [1] (or broadly, instance based explanations) [2] where the training data that is most influential in a prediction (basically the analogy the model used) is highlighted. See also [3] where I provide a take on using these methods for understanding if you can trust a prediction or not, based on whether the model has an appropriate analogy to work from [1] Understanding black box predictions through influence functions, https://arxiv.org/abs/1703.04730 [2] Evaluation of Similarity-based Explanations…

Paper and discussion history. A mention is not an endorsement.

Understanding Black-box Predictions via Influence Functions — HN submission

Mon, 20 Mar 2017 15:24:46 GMT

Understanding Black-Box Machine Learning Predictions via Influence Functions

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.