Pang Wei Koh, Percy Liang
2017 · arXiv
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. To scale up influence functions to modern machine learning settings, we develop a simple, efficient implementation that requires only oracle access to gradients and Hessian-vector products. We show that even on non-convex and non-differentiable models where the theory breaks down, approximations to influence functions can still provide valuable information. On linear models and convolutional neural networks, we demonstrate that influence functions are useful for multiple purposes: understanding model behavior, debugging models, detecting dataset errors, and even creating visually-indistinguishable training-set attacks.
Metadata: arXiv. Abstracts retain their original authors’ rights.
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
3 selected excerpts · 3 total mentions
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.
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.
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, https://arxiv.org/abs/2006.04528 [3] https://www.willows.ai/blog/getting-more
More from the same topics.
Tom B. Brown, Benjamin Mann, Nick Ryder +28
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task.
Anders Sandberg, Eric Drexler, Toby Ord
The Fermi paradox is the conflict between an expectation of a high {\em ex ante} probability of intelligent life elsewhere in the universe and the apparently lifeless universe we in fact observe.
Chiyuan Zhang, Samy Bengio, Moritz Hardt +2
Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance.
Scott Aaronson
In honor of Alan Turing's hundredth birthday, I unwisely set out some thoughts about one of Turing's obsessions throughout his life, the question of physics and free will.