[Colloquium] 1/27 Talk at TTIC: Subhransu Maji, University of Massachusetts Amherst

[Jerome Allen]

*When:*      Monday, January 27th at 11:00 am

*Where:*     TTIC, 6045 S. Kenwood Avenue, 5th Floor, Room 526



*Who: *        Subhransu Maji, University Of Massachusetts Amherst


Title: Task-specific Recognition by Modeling Visual Tasks and their
Relations

Abstract: The AI revolution powered in part by advances in deep learning
has led to many successes in the last decade. Among others, it has enabled
us to study various ecological phenomena by analyzing data from weather
RADAR networks at an unprecedented scale (e.g., [1, 2]). Yet the vast
majority of important applications remain beyond the scope of current AI
systems. One of the barriers is that existing algorithms lack the ability
to learn from limited training data. This is a fundamental challenge
because most real-world data is heavy-tailed and supervision is hard to
acquire. I argue that a principled framework for reasoning about AI
problems can enable modular and data-efficient solutions. Towards this end
I will describe our framework for modeling computer vision tasks and
relations between them (e.g., [3, 4]). Our approach called "task2vec"
computes a vector representation of a task using the Fisher information of
a generic “probe” network. We show that the distance between these vectors
correlates with natural metrics between the domains and labels of tasks. It
is also predictive of transfer, i.e., how much does training a deep network
on one task benefit another, and can be used for model recommendation. On a
portfolio of hundreds of vision tasks the recommended network outperforms
the current gold standard of fine-tuning an ImageNet pre-trained network.
I’ll conclude with some of the life-cycle challenges that we need to
address to make AI systems widely applicable.

References:

   1.

   Kyle G. Horton, Frank A. La Sorte, Daniel Sheldon, Tsung-Yu Lin, Kevin
   Winner, Garrett Bernstein, Subhransu Maji, Wesley M. Hochachka, and
   Andrew Farnsworth.  Phenology of nocturnal avian migration has shifted
   at the continental scale
   <https://www.nature.com/articles/s41558-019-0648-9>, Nature Climate
   Change, Dec 2019
   2.

   Tsung‐Yu Lin, Kevin Winner, Garrett Bernstein, Abhay Mittal, Adriaan M.
   Dokter, Kyle G. Horton, Cecilia Nilsson, Benjamin M. Van Doren, Andrew
   Farnsworth, Frank A. La Sorte, Subhransu Maji and Daniel Sheldon.
   MistNet: Measuring historical bird migration in the US using archived
   weather radar data and convolutional neural networks
   <https://besjournals.onlinelibrary.wiley.com/doi/abs/10.1111/2041-210X.13280>,
   Methods in Ecology and Evolution 10(11):1908-1922, Aug 2019
   3.

   Alessandro Achille, Michael Lam, Rahul Tewari, Avinash Ravichandran,
   Subhransu Maji, Charless Fowlkes, Stefano Soatto, Pietro Perona, Task2Vec:
   Task Embedding for Meta-Learning <https://arxiv.org/abs/1902.03545>,
   ICCV 2019
   4.

   Jong-Chyi Su, Subhransu Maji, Bharath Hariharan, When Does
   Self-supervision Improve Few-shot Learning?
   <https://arxiv.org/abs/1910.03560> arXiv:1910.03560, Oct 2019


Bio: I am an Assistant Professor in the College of Information and Computer
Sciences at the University of Massachusetts Amherst where I co-direct the
Computer Vision Lab. I am also affiliated with the Center of Data Science
and AWS AI. Prior to this I spent three years as a Research Assistant
Professor at TTI Chicago, a philanthropically endowed academic institute in
the University of Chicago campus. I obtained my Ph.D. in Computer Science
from the University of California at Berkeley in 2011 and B.Tech. in
Computer Science and Engineering from IIT Kanpur in 2006. For my work, I
have received a Google graduate fellowship, NSF CAREER Award (2018), and a
best paper honorable mention at CVPR 2018. I also serve on the editorial
board of the International Journal of Computer Vision (IJCV).


*Jerome Allen*
Executive Assistant
*Toyota Technological Institute*
6045 S. Kenwood Avenue
Room 518
Chicago, IL  60637
p:(773) 702-2311
*jallen at ttic.edu <jallen at ttic.edu>*
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20200124/7cddfbff/attachment-0001.html>