[Theory] REMINDER: 12/9 TTIC Colloquium: Adam Kalai, Microsoft Research

[Mary Marre]

*When:*      Monday, December 9th at 11:00 am



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



*Who: *       Adam Kalai, Microsoft Research




*Title:        *Expect the Unexpected: Learning to Generalize across
Similarly Different Datasets



*Abstract: *People learn to drive in one city and are easily able to drive
in others, exhibiting a robustness that Machine Learning algorithms often
lack. We consider a model of learning where test data may be very different
from training data, e.g., training data is collected in two cities but test
data is from a third city (or training data is from a single city but can
be appropriately subsampled). We analyze and experiment with algorithms for
learning from the differences across training data splits to predict well
on the unseen distribution. Our model differs from most work on domain
adaptation in that: the test distribution may not even overlap with the
training distribution, we have no unlabeled samples from the test
distribution, and yet the training and test distributions are assumed to be
drawn from the same meta-distribution over distributions. For these
reasons, importance-weighting approaches are not applicable and different
algorithms are necessary.



Joint work with Vikas Garg (MIT), Katrina Ligett (Hebrew University), and
Steven Wu (University of Minnesota).



*Host:* Avrim Blum <avrim at ttic.edu>
<madhurt at ttic.edu>


For more information on the colloquium series or to subscribe to the
mailing list, please see http://www.ttic.edu/colloquium.php





Mary C. Marre
Administrative Assistant
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 517*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*


On Sun, Dec 8, 2019 at 8:05 PM Mary Marre <mmarre at ttic.edu> wrote:

> *When:*      Monday, December 9th at 11:00 am
>
>
>
> *Where:*     TTIC, 6045 S. Kenwood Avenue, 5th Floor, Room 526
>
>
>
> *Who: *       Adam Kalai, Microsoft Research
>
>
>
>
> *Title:        *Expect the Unexpected: Learning to Generalize across
> Similarly Different Datasets
>
>
>
> *Abstract: *People learn to drive in one city and are easily able to
> drive in others, exhibiting a robustness that Machine Learning algorithms
> often lack. We consider a model of learning where test data may be very
> different from training data, e.g., training data is collected in two
> cities but test data is from a third city (or training data is from a
> single city but can be appropriately subsampled). We analyze and experiment
> with algorithms for learning from the differences across training data
> splits to predict well on the unseen distribution. Our model differs from
> most work on domain adaptation in that: the test distribution may not even
> overlap with the training distribution, we have no unlabeled samples from
> the test distribution, and yet the training and test distributions are
> assumed to be drawn from the same meta-distribution over distributions. For
> these reasons, importance-weighting approaches are not applicable and
> different algorithms are necessary.
>
>
>
> Joint work with Vikas Garg (MIT), Katrina Ligett (Hebrew University), and
> Steven Wu (University of Minnesota).
>
>
>
> *Host:* Avrim Blum <avrim at ttic.edu>
> <madhurt at ttic.edu>
>
>
> For more information on the colloquium series or to subscribe to the
> mailing list, please see http://www.ttic.edu/colloquium.php
>
>
> Mary C. Marre
> Administrative Assistant
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Room 517*
> *Chicago, IL  60637*
> *p:(773) 834-1757*
> *f: (773) 357-6970*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
>
> On Mon, Dec 2, 2019 at 1:36 PM Mary Marre <mmarre at ttic.edu> wrote:
>
>> *When:*      Monday, December 9th at 11:00 am
>>
>>
>>
>> *Where:*     TTIC, 6045 S. Kenwood Avenue, 5th Floor, Room 526
>>
>>
>>
>> *Who: *       Adam Kalai, Microsoft Research
>>
>>
>>
>>
>> *Title:        *Expect the Unexpected: Learning to Generalize across
>> Similarly Different Datasets
>>
>>
>>
>> *Abstract: *People learn to drive in one city and are easily able to
>> drive in others, exhibiting a robustness that Machine Learning algorithms
>> often lack. We consider a model of learning where test data may be very
>> different from training data, e.g., training data is collected in two
>> cities but test data is from a third city (or training data is from a
>> single city but can be appropriately subsampled). We analyze and experiment
>> with algorithms for learning from the differences across training data
>> splits to predict well on the unseen distribution. Our model differs from
>> most work on domain adaptation in that: the test distribution may not even
>> overlap with the training distribution, we have no unlabeled samples from
>> the test distribution, and yet the training and test distributions are
>> assumed to be drawn from the same meta-distribution over distributions. For
>> these reasons, importance-weighting approaches are not applicable and
>> different algorithms are necessary.
>>
>>
>>
>> Joint work with Vikas Garg (MIT), Katrina Ligett (Hebrew University), and
>> Steven Wu (University of Minnesota).
>>
>>
>>
>> *Host:* Avrim Blum <avrim at ttic.edu>
>> <madhurt at ttic.edu>
>>
>>
>> For more information on the colloquium series or to subscribe to the
>> mailing list, please see http://www.ttic.edu/colloquium.php
>>
>>
>>
>>
>> Mary C. Marre
>> Administrative Assistant
>> *Toyota Technological Institute*
>> *6045 S. Kenwood Avenue*
>> *Room 517*
>> *Chicago, IL  60637*
>> *p:(773) 834-1757*
>> *f: (773) 357-6970*
>> *mmarre at ttic.edu <mmarre at ttic.edu>*
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20191209/e913a4fc/attachment-0001.html>