December 5th, Wednesday 14:15, Room 303, Jacobs Building

Title: Learning patterns in Big Data from small data using core-sets

Lecturer: Dan Feldman

Lecturer homepage : http://people.csail.mit.edu/dannyf/

Affiliation : Electrical Engineering and Computer Science Department, MIT

 

When we want to discover patterns in data, we usually use the best available algorithm/software or try to improve it. In recent years we have started exploring a different approach: instead of improving the algorithm, reduce the input data and run the existing algorithm on the reduced data to obtain the desired output much faster.

A core-set for a given problem is a semantic compression of its input, in the sense that a solution for the problem with the (small) core-set as input yields an approximate solution to the problem with the original (Big) data. With tools such as Hadoop, core-sets can be computed on a streaming input, using a manageable amount of memory, and in parallel (e.g. clouds or GPUs).

In this talk I will describe the coreset approach and recent algorithmic achievements for computing coresets with performance guarantees. Some examples will be given on how I applied this magical new paradigm on robots, images, and text mining.

Finally, I will describe in detail iDiary: a system that turns large sensor signals collected from smart-phones or robots into textual descriptions of the trajectories. The system features a user interface similar to Google Search that allows users to type text queries on activities (e.g., .Where did I have dinner last time I visited Paris?.) and receive textual answers based on GPS signals.

Bio:

Dan Feldman is a post-doc at MIT in the Distributed Robotics Lab, where he develops systems for handling streaming Big data from sensors, smartphones, images, and robots. He got his Ph.D. from Tel-Aviv University in 2010, under the supervision of Prof. Micha Sharir and Prof. Amos Fiat. He then was a postdoc at the Center for the Mathematics of Information at Caltech for a year and a half, where he started to reduce the gap between theoretical computational geometry and practical machine learning. He is specialized in developing software for scalable data compression, based on core-set constructions with provable guarantees. His coresets were implemented in several start-ups, banks, super-markets, and internet search companies over the recent years, to name just a few. When he is not working, Dan is building robots with his very own coresets, Ariel and Eleanor.