Sokbae "Simon" Lee - Columbia

An Econometric View of Algorithmic Subsampling (joint with Serena Ng)

    Date:  11/29/2018 (Thu)

    Time:  3:30pm- 5:00pm

    Location:  Seminar will be held on-site: Social Sciences room 113

    Organizer:  Adam Rosen


Meeting Schedule: Login or email the organizer to schedule a meeting.

    All meetings will be held in the same location as the seminar unless otherwise noted.

    - AA 3343 LGA - RDU arriving at 9:46 AM

   11:00am - Jia Li (228G)

   11:30am - Francesca Molinari (242)

   12:00pm - Lunch: Simon, Francesca, Andrew Chesher

    1:30pm - Federico Bugni (240)

    2:00pm - Arnaud Maurel (225)

    2:30pm - Matt Masten (202)

    3:00pm - Andrii Babii (220A)

    3:30pm - Seminar Presentation (3:30pm to 5:00pm)

    5:00pm - Adam Rosen (221B)

    6:30pm - Dinner @ Parizade: Simon, Adam, Matt, Fede, Andrew Chesher, Joe Altonji


    Additional Comments:  Abstract: Improved technology has dramatically reduced the cost of collecting data, and datasets that are terabytes in size are not uncommon. Not only can computation be slow with data of this size, memory and storage constraints may render analysis infeasible. This has motivated computer scientists to devise ways to subspace reduction schemes while preserving the structure of the original data. We first review the foundation of these methods and highlight results mostly derived from an algorithmic perspective, void of any probabilistic structure. To gain insights from an econometric point of view, we study the implications of subspace reduction from the viewpoint of inference using the linear regression model as our framework. We also propose divide and conquer methods that can make efficient use of the data while minimizing the computational bottlenecks. These methods can be useful in estimation of simple models with lots of data, or in estimation of complex models when working with good sketches of the data can significantly reduce debugging time.