Anyone can fit an elephant

Thanks to Zalando (https://www.meetup.com/Zalando-Tech-Events-Dublin/)who kindly offer to be our Host for this meeting.

http://photos4.meetupstatic.com/photos/event/2/0/d/b/600_446288411.jpeg

“With four parameters I can fit an elephant, and with five I can make him wiggle his trunk.” - John von Neumann

Abstract

Predicting the past is easy. With enough model complexity we can create elaborate functions that perfectly explain our training data, even the noise. The learning part of ML is all about generalizing to new situations and new data. That means diagnosing new patients, recommending products to new customers and predicting next year's tornadoes. To perform well at these tasks we often need to cut back on complexity. But how do we even measure it? And how do we know when we're in elephant-trunk-wiggling territory?

http://photos2.meetupstatic.com/photos/event/e/2/e/6/600_448918086.jpeg

This talk is about regularization, a fundamental machine learning technique, and its use in conjunction with cross-validation for addressing what's formally referred to as the bias-variance dilemma. We’ll discuss the importance of regularization in many sorts of machine learning problems, from linear regression to deep neural networks. It'll be a theoretical talk with plenty of practical examples demonstrated with R packages like GLMNET (https://cran.r-project.org/web/packages/glmnet/index.html) and mxnet (https://github.com/dmlc/mxnet).

http://photos4.meetupstatic.com/photos/event/c/3/4/4/600_449149988.jpeg

Speaker

Pat O’Reilly is a senior data scientist at Allstate Insurance and is based in Belfast. The predictive models developed in the Belfast office directly impact pricing for millions of customers in the United States. Other areas of research include churn prediction, deep learning, geospatial data, fraud detection and graph mining.

http://photos1.meetupstatic.com/photos/event/c/2/f/e/600_449149918.jpeg

http://photos2.meetupstatic.com/photos/event/e/2/e/f/600_448918095.jpeg