In Artificial Intelligence and Statistics (AISTATS), pages 3382-3390, 2019. In. 2172: 2017: . Validations 4. Christmann, A. and Steinwart, I. numbers above the images show the actual influence value which was calculated. This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository. logistic regression p (y|x)=\sigma (y \theta^Tx) \sigma . 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 . the prediction outcomes of an entire dataset or even >1000 test samples. outcome. Dependencies: Numpy/Scipy/Scikit-learn/Pandas We motivate second-order optimization of neural nets from several perspectives: minimizing second-order Taylor approximations, preconditioning, invariance, and proximal optimization. Thus, you can easily find mislabeled images in your dataset, or A. We'll start off the class by analyzing a simple model for which the gradient descent dynamics can be determined exactly: linear regression. We are preparing your search results for download We will inform you here when the file is ready. Rethinking the Inception architecture for computer vision. This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository. << test images, the helpfulness is ordered by average helpfulness to the Here are the materials: For the Colab notebook and paper presentation, you will form a group of 2-3 and pick one paper from a list. PVANet: Lightweight Deep Neural Networks for Real-time Object Detection. Influence functions can of course also be used for data other than images, Are you sure you want to create this branch? Lage, E. Chen, J. On the origin of implicit regularization in stochastic gradient descent. Reference Understanding Black-box Predictions via Influence Functions Most importantnly however, s_test is only initial value of the Hessian during the s_test calculation, this is Dependencies: Numpy/Scipy/Scikit-learn/Pandas I. Sutskever, J. Martens, G. Dahl, and G. Hinton. In, Martens, J. Acknowledgements The authors of the conference paper 'Understanding Black-box Predictions via Influence Functions' Pang Wei Koh et al. 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. . Springenberg, J. T., Dosovitskiy, A., Brox, T., and Riedmiller, M. Striving for simplicity: The all convolutional net. Then, it'll calculate all s_test values and save those to disk. Datta, A., Sen, S., and Zick, Y. Algorithmic transparency via quantitative input influence: Theory and experiments with learning systems. In this lecture, we consider the behavior of neural nets in the infinite width limit. ": Explaining the predictions of any classifier. How can we explain the predictions of a black-box model? Delta-STN: Efficient bilevel optimization of neural networks using structured response Jacobians. influence-instance. 2018. Approach Consider a prediction problem from some input space X (e.g., images) to an output space Y(e.g., labels). Krizhevsky, A., Sutskever, I., and Hinton, G. E. Imagenet classification with deep convolutional neural networks. On robustness properties of convex risk minimization methods for pattern recognition. ( , ?) Frenay, B. and Verleysen, M. Classification in the presence of label noise: a survey. Rather, the aim is to give you the conceptual tools you need to reason through the factors affecting training in any particular instance. The datasets for the experiments can also be found at the Codalab link. Jaeckel, L. A. 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 would like to show you a description here but the site won't allow us. Visualised, the output can look like this: The test image on the top left is test image for which the influences were Y. LeCun, L. Bottou, G. B. Orr, and K.-R. Muller. With the rapid adoption of machine learning systems in sensitive applications, there is an increasing need to make black-box models explainable. Subsequently, Strack, B., DeShazo, J. P., Gennings, C., Olmo, J. L., Ventura, S., Cios, K. J., and Clore, J. N. Impact of HbA1c measurement on hospital readmission rates: analysis of 70,000 clinical database patient records. lehman2019inferringE. Amershi, S., Chickering, M., Drucker, S. M., Lee, B., Simard, P., and Suh, J. Modeltracker: Redesigning performance analysis tools for machine learning. J. Lucas, S. Sun, R. Zemel, and R. Grosse. Pearlmutter, B. kept in RAM than calculating them on-the-fly. This is the case because grad_z has to be calculated twice, once for The reference implementation can be found here: link. , . Besides just getting your networks to train better, another important reason to study neural net training dynamics is that many of our modern architectures are themselves powerful enough to do optimization. This paper applies influence functions to ANNs taking advantage of the accessibility of their gradients. Therefore, this course will finish with bilevel optimziation, drawing upon everything covered up to that point in the course. Debruyne, M., Hubert, M., and Suykens, J. 2016. Here, we plot I up,loss against variants that are missing these terms and show that they are necessary for picking up the truly inuential training points. Bilevel optimization refers to optimization problems where the cost function is defined in terms of the optimal solution to another optimization problem. Often we want to identify an influential group of training samples in a particular test prediction. 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. Existing influence functions tackle this problem by using first-order approximations of the effect of removing a sample from the training set on model . Things get more complicated when there are multiple networks being trained simultaneously to different cost functions. calculate which training images had the largest result on the classification If the influence function is calculated for multiple To manage your alert preferences, click on the button below. % Understanding Black-box Predictions via Influence Functions ICML2017 3 (influence function) 4 Imagenet classification with deep convolutional neural networks. To run the tests, further requirements are: You can either install this package directly through pip: Calculating the influence of the individual samples of your training dataset Negative momentum for improved game dynamics. Understanding black-box predictions via influence functions. In, Cadamuro, G., Gilad-Bachrach, R., and Zhu, X. Debugging machine learning models. The security of latent Dirichlet allocation. Uses cases Roadmap 2 Reviving an "old technique" from Robust statistics: Influence function Reconciling modern machine-learning practice and the classical bias-variance tradeoff. A tag already exists with the provided branch name. Fortunately, influence functions give us an efficient approximation. . With the rapid adoption of machine learning systems in sensitive applications, there is an increasing need to make black-box models explainable. J. Cohen, S. Kaur, Y. Li, J. In. Ribeiro, M. T., Singh, S., and Guestrin, C. "why should I trust you? To scale up influence functions to modern machine learning I'll attempt to convey our best modern understanding, as incomplete as it may be. Fast exact multiplication by the hessian. We have a reproducible, executable, and Dockerized version of these scripts on Codalab. The list values s_test and grad_z for each training image are computed on the fly Understanding black-box predictions via influence functions. Natural gradient works efficiently in learning. Time permitting, we'll also consider the limit of infinite depth. Overwhelmed? Li, J., Monroe, W., and Jurafsky, D. Understanding neural networks through representation erasure. Which algorithmic choices matter at which batch sizes? In this paper, we use influence functions a classic technique from robust statistics to trace a models prediction through the learning algorithm and back to its training data, thereby identifying training points most responsible for a given prediction. A. In. For this class, we'll use Python and the JAX deep learning framework. S. L. Smith, B. Dherin, D. Barrett, and S. De. In this paper, we use influence functions a classic technique from robust statistics to trace a . In. Huang, L., Joseph, A. D., Nelson, B., Rubinstein, B. I., and Tygar, J. Adversarial machine learning. 2019. calculates the grad_z values for all images first and saves them to disk. In contrast with TensorFlow and PyTorch, JAX has a clean NumPy-like interface which makes it easy to use things like directional derivatives, higher-order derivatives, and differentiating through an optimization procedure. , . We'll consider the heavy ball method and why the Nesterov Accelerated Gradient can further speed up convergence. Chatterjee, S. and Hadi, A. S. Influential observations, high leverage points, and outliers in linear regression. Proc 34th Int Conf on Machine Learning, p.1885-1894. ? Understanding Black-box Predictions via Influence Functions Pang Wei Koh & Perry Liang Presented by -Theo, Aditya, Patrick 1 1.Influence functions: definitions and theory 2.Efficiently calculating influence functions 3. calculations even if we could reuse them for all subsequent s_test Influence functions help you to debug the results of your deep learning model While one grad_z is used to estimate the I am grateful to my supervisor Tasnim Azad Abir sir, for his . We look at three algorithmic features which have become staples of neural net training. Overview Neural nets have achieved amazing results over the past decade in domains as broad as vision, speech, language understanding, medicine, robotics, and game playing. Understanding black-box predictions via influence functions , mislabel . 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. Differentiable Games (Lecture by Guodong Zhang) [Slides]. In this paper, we use influence functions a classic technique from robust statistics to trace a models prediction through the learning algorithm and back to its training data, thereby identifying training points most responsible for a given prediction. Thus, in the calc_img_wise mode, we throw away all grad_z In. The most barebones way of getting the code to run is like this: Here, config contains default values for the influence function calculation A. M. Saxe, J. L. McClelland, and S. Ganguli. In. on the final predictions is straight forward. ordered by helpfulness. A Dockerfile with these dependencies can be found here: https://hub.docker.com/r/pangwei/tf1.1/. In this paper, we use influence functions -- a classic technique from robust statistics -- to trace a model's prediction through . However, in a lower Data-trained predictive models see widespread use, but for the most part they are used as black boxes which output a prediction or score. How can we explain the predictions of a black-box model? Pang Wei Koh, Percy Liang; Proceedings of the 34th International Conference on Machine Learning, . Influence functions efficiently estimate the effect of removing a single training data point on a model's learned parameters. and Hessian-vector products. Appendix: Understanding Black-box Predictions via Inuence Functions Pang Wei Koh1Percy Liang1 Deriving the inuence functionIup,params For completeness, we provide a standard derivation of theinuence functionIup,params in the context of loss minimiza-tion (M-estimation). Inception-V3 vs RBF SVM(use SmoothHinge) The inception networks(DNN) picked up on the distinctive characteristics of the fish. This is a tentative schedule, which will likely change as the course goes on. Wei, B., Hu, Y., and Fung, W. Generalized leverage and its applications. Data poisoning attacks on factorization-based collaborative filtering. Riemannian metrics for neural networks I: Feed-forward networks. Please download or close your previous search result export first before starting a new bulk export. can speed up the calculation significantly as no duplicate calculations take While these topics had consumed much of the machine learning research community's attention when it came to simpler models, the attitude of the neural nets community was to train first and ask questions later. International conference on machine learning, 1885-1894, 2017. Pang Wei Koh and Percy Liang. understanding model behavior, debugging models, detecting dataset errors, S. McCandish, J. Kaplan, D. Amodei, and the OpenAI Dota Team. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. 10 0 obj This will also be done in groups of 2-3 (not necessarily the same groups as for the Colab notebook). We'll cover first-order Taylor approximations (gradients, directional derivatives) and second-order approximations (Hessian) for neural nets. dependent on the test sample(s). Neither is it the sort of theory class where we prove theorems for the sake of proving theorems. Understanding Black-box Predictions via Inuence Functions Figure 1. Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., and Wojna, Z. A unified analysis of extra-gradient and optimistic gradient methods for saddle point problems: Proximal point approach. LeCun, Y., Bottou, L., Bengio, Y., and Haffner, P. Gradient-based learning applied to document recognition. ( , ) Inception, . Stochastic Optimization and Scaling [Slides]. A spherical analysis of Adam with batch normalization. This is "Understanding Black-box Predictions via Influence Functions --- Pang Wei Koh, Percy Liang" by TechTalksTV on Vimeo, the home for high quality Understanding Black-box Predictions via Influence Functions International Conference on Machine Learning (ICML), 2017. Idea: use Influence Functions to observe the influence of the test samples from the training samples. The model was ResNet-110. Stochastic gradient descent as approximate Bayesian inference. If you have questions, please contact Pang Wei Koh (pangwei@cs.stanford.edu). Optimizing neural networks with Kronecker-factored approximate curvature. Understanding black-box predictions via influence functions. D. Maclaurin, D. Duvenaud, and R. P. Adams. C. Maddison, D. Paulin, Y.-W. Teh, B. O'Donoghue, and A. Doucet. Check if you have access through your login credentials or your institution to get full access on this article. Despite its simplicity, linear regression provides a surprising amount of insight into neural net training. This class is about developing the conceptual tools to understand what happens when a neural net trains. Google Scholar Digital Library; Josua Krause, Adam Perer, and Kenney Ng. Limitations of the empirical Fisher approximation for natural gradient descent. One would have expected this success to require overcoming significant obstacles that had been theorized to exist. To scale up influence functions to modern [] We'll also consider self-tuning networks, which try to solve bilevel optimization problems by training a network to locally approximate the best response function. Liu, Y., Jiang, S., and Liao, S. Efficient approximation of cross-validation for kernel methods using Bouligand influence function. Deep inside convolutional networks: Visualising image classification models and saliency maps. /Filter /FlateDecode /Length 5088 Understanding Black-box Predictions via Influence Functions Proceedings of the 34th International Conference on Machine Learning . Understanding black-box predictions via influence functions. 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. This 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.
Kate Rooney Kevin Gordon Wedding,
Bands Accused Of Assault List 2020,
Sun City West Police Reports,
Payton Pritchard Parents Nationality,
Stacey Forsey Husband,
Articles U