(1) Data Science in Business (2) Model Evaluation with Skater

Please RSVP for free tickets on Eventbrite: https://www.eventbrite.com/e/pydata-la-meet-up-tickets-46535281279

6:30 - 7:00pm, Networking + Food/Drink
7:00 - 7:45pm, Nathan Janos
7:45 - 8:30pm, Pramit Choudhary
8:30 - 9:00pm, Networking

Presentations:

Nathan Janos

Bio: Nathan Janos is Chief Data Officer at System1. Previously, at Business.com he built SEM optimizations for $100M+ of search engine spend (Business.com sold for $350M). At MarketShare he worked with Fortune 50 companies creating highly customized econometric models. As Convertro's Chief Data Officer he developed their statistical model solutions for cross-channel/cross-device attribution and patented methods in TV attribution (Convertro sold for $100M to AOL). He has a B.S. in C.S. and Engineering from MIT with an emphasis on A.I. and spent three years at the MIT Media Lab. He enjoys sailing around the Channel Islands, fly fishing and swimming.

Title: How to Get Data Science Buy-In from Your Business Organization

Abstract: Whenever we apply data science and optimization to a new business area at System1, there is a recurring pattern of challenges that we’ve learned to anticipate and overcome. This talk describes how we get people interested in data science solutions, the importance of well-defined goals and metrics, how to overcome negative perceptions, and tempering expectations. In addition to this, technical solutions to some of these problems will be highlighted and elaborated upon further in the subsequent talk by Pramit Choudhary.

Pramit Choudhary

Bio: Pramit Choudhary is a lead data scientist at DataScience.com(R&D Labs), where he focuses on optimizing and applying Machine Learning to solve real-world problems. Currently, he is leading initiatives on figuring out better ways to explain a model’s learned decision policies to reduce the chaos in building effective models and close the gap between a prototype and operationalized models.

Title: Model evaluation in the land of deep learning with Skater

Abstract: Pramit Choudhary shares tricks to enable class-discriminative visualizations for computer vision problems when using convolutional neural networks (CNN's) and approach to help enable transparency of CNN's by capturing metrics during the validation step and highlighting salient features in the image which are driving prediction.
Skater: https://github.com/datascienceinc/Skater

Evaluating model decisions might still be easy for linear models but gets difficult in the world of deep neural networks (DNNs). This complexity might increase multifold for use cases related to computer vision (image classification, image captioning or visual QnA(VQA), text classification), sentiment analysis, or topic modeling. ResNets, a recently published state-of-the-art DNN, has over 200 layers. Interpreting input features and output categorization over multiple layers is challenging. The lack of decomposability and intuitiveness associated with DNNs prevents widespread adoption even with their superior performance compared to more classical machine learning approaches. The faithful interpretation of DNNs will help not only provide insight into the failure modes (false positives and false negatives) but also enable the humans in the loop to evaluate the robustness of the model against noise. This brings in trust and transparency to the predictive algorithm.