Abstract: We start with a simple question: how many bananas are there in the grocery store down the road? It should be a straightforward calculation: just add the shipment and subtract the sales. Easy enough, right? Not so fast! What about bananas that got thrown out but not recorded? What if shipment data is missing? What if the sales of conventional bananas are getting mixed up with organic bananas? Good machine learning requires good data. In this talk, we will cover lessons learned in the trenches of messy data. We discuss systematic approaches to data hygiene. We will also cover the use of bayesian statistics to estimate data fidelity and techniques for generating synthetic ground truths to test data-cleaning methodologies.

Bio:Mark is a Staff Applied Scientist at Afresh, where he develops algorithms for inventory management of produce with the goal of reducing food waste. Previously, he was a Senior Research Scientist at Amazon. He received his Ph.D. from UC Berkeley, where his most notable accomplishment was giving a presentation on convex optimization at which Salar solemnly nodded in tacit approval.

Office hours with the speaker: If you are interested in meeting with the speaker, please send me an email by March 26. If there is enough demand, I will try to schedule an office hour with the speaker.