2021/07/16 by Yves-Laurent Kom Samo, Samo, Yves-Laurent Kom · 1 voice
Business, Management and Accounting · Computer Science · Decision Sciences · Engineering · Mathematics · #Advanced Statistical Process Monitoring #Big Data and Business Intelligence #FOS: Computer and information sciences #Forecasting Techniques and Applications #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Manufacturing Process and Optimization #Software Engineering (cs.SE) #cs.LG #cs.SE #stat.ML
paper · pdf · doi:10.48550/arxiv.2107.08066
openalex publication_date 2021/07/16 · arxiv published 2021/07/16 · arxiv updated 2021/08/12 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
We introduce the first application of the lean methodology to machine\nlearning projects. Similar to lean startups and lean manufacturing, we argue\nthat lean machine learning (LeanML) can drastically slash avoidable wastes in\ncommercial machine learning projects, reduce the business risk in investing in\nmachine learning capabilities and, in so doing, further democratize access to\nmachine learning. The lean design pattern we propose in this paper is based on\ntwo realizations. First, it is possible to estimate the best performance one\nmay achieve when predicting an outcome y \∈ \Y using a given set of\nexplanatory variables x \∈ \X, for a wide range of performance\nmetrics, and without training any predictive model. Second, doing so is\nconsiderably easier, faster, and cheaper than learning the best predictive\nmodel. We derive formulae expressing the best R2, MSE, classification\naccuracy, and log-likelihood per observation achievable when using x to\npredict y as a function of the mutual information I\(y; x\), and\npossibly a measure of the variability of y (e.g. its Shannon entropy in the\ncase of classification accuracy, and its variance in the case regression MSE).\nWe illustrate the efficacy of the LeanML design pattern on a wide range of\nregression and classification problems, synthetic and real-life.\n