It is now common practice to try and solve machine learning problems by starting with a complex existing model or architecture, and fine-tuning/adapting it to the task at hand. However, outliers, errors or even just sloppiness in training data often lead to drastic drops in performance.
We investigate a simple generic approach to correct for this, motivated by a classic statistical idea: trimmed loss. This advocates jointly (a) selecting which training samples to ignore, and (b) fitting a model on the remaining samples. As such this is computationally infeasible even for linear regression. We propose and study the natural iterative variant that alternates between these two steps (a) and (b) - each of which individually can be easily accomplished in pretty much any statistical setting. We also study the batch-SGD variant of this idea. We demonstrate both theoretically (for generalized linear models) and empirically (for moderate-sized neural network models) that this effectively recovers accuracy in the presence of bad training data.
This work is joint with Yanyao Shen and Vatsal Shah and appears in NeurIPS 2019, ICML 2019 and AISTATS 2020.
Sujay Sanghavi is an Associate Professor at the University of Texas, Austin. He received a PhD in ECE and an MS in Math and ECE from the University of Illinois, and was a postdoc at LIDS in MIT. Sujay's research focus is rigorous methodological innovation in machine learning, using ideas from optimization, statistics and graph theory. He has received early career awards from the NSF and DoD, and currently leads the NSF TRIPODS Institute on the Foundations of Data Science at UT.
Sujay is also interested in learning from and applying his ideas in industry. He has been a Visiting Scientist at Google Research, a senior quant at Engineers Gate and is currently a Principal Researcg Scientist and Amazon Scholar at Amazon.