[Colloquium] Re: REMINDER: 1/6 Research at TTIC: Ayan Chakrabarti, TTIC

Fri Jan 6 11:11:11 CST 2017

When:     Friday, January 6th at noon

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

Who:       Ayan Chakrabarti; TTIC

Title:       Machine Learning for Measurement and Inference from Visual Data

Abstract: Computer vision systems seek to make sense of the physical world
and recover scene attributes---like shape, geometry, motion, material, and
object identity---from visual measurements. As an estimation problem, this
is ill-posed since the light reflected from objects in a scene carries only
indirect information about these attributes, and there are typically
infinitely many physically plausible explanations for any given captured
image or video. Vision systems must rely heavily on leveraging the
statistics of natural scenes using machine learning tools, to design not
just accurate estimation algorithms, but also optimal measurement
strategies.

In the first part of the talk, I will talk about a structured inference
approach to the recovery continuous-valued maps of per-point scene
properties---like depth, surface normals, surface reflectance, motion,
etc.---from images. Such scene-value maps typically exhibit significant
spatial structure (planarity, rigidity, etc.), and leveraging this
structure during inference can lead to more accurate estimates. In this
context, I will describe recently proposed method for estimating scene
depth from a single color image, which trains a neural network to produce
dense probabilistic estimates of different elements of local geometric
structure, and then harmonizes these estimates to produce a consistent
depth map.

While a major focus of computer vision is developing effective inference
methods for images from a traditional camera, in the second half of the
talk I will discuss whether we can redesign the camera itself to make
measurements that are optimal for inference. In particular, I will talk
about a machine-learning approach to automatically "learning" the camera's
optical design jointly with the computational method for inference. Using
color imaging as an example task, I will introduce a framework where the
camera sensor's measurement process is encoded as a neural-network layer,
whose learnable weights parameterize the possible measurement choices for
the sensor. This measurement layer is then trained end-to-end with a neural
network that carries out inference on the corresponding measurements, to
maximize the accuracy of the final output. I'll show that this approach is
able to automatically discover a measurement strategy that, when used with
the jointly learned inference network, significantly outperforms
traditional sensor designs.

************************************************************
****************************************************

*Research at TTIC Seminar Series*

TTIC is hosting a weekly seminar series presenting the research currently
underway at the Institute. Every week a different TTIC faculty member will
present their research.  The lectures are intended both for students
seeking research topics and adviser, and for the general TTIC and
University of Chicago communities interested in hearing what their
colleagues are up to.

To receive announcements about the seminar series, please subscribe to the
mailing list: https://groups.google.com/a/ttic.edu/group/talks/subscribe

Speaker details can be found at: http://www.ttic.edu/tticseminar.php.

For additional questions, please contact Nathan Srebro at nati at ttic.edu
<mcallester at ttic.edu>

Mary C. Marre
Administrative Assistant
*Toyota Technological Institute*
*6045 S. Kenwood Avenue*
*Room 504*
*Chicago, IL  60637*
*p:(773) 834-1757*
*f: (773) 357-6970*
*mmarre at ttic.edu <mmarre at ttic.edu>*

On Thu, Jan 5, 2017 at 10:16 AM, Mary Marre <mmarre at ttic.edu> wrote:

> When:     Friday, January 6th at noon
>
>
>
> Where:    TTIC, 6045 S Kenwood Avenue, 5th Floor, Room 526
>
>
>
> Who:       Ayan Chakrabarti; TTIC
>
>
>
>
> Title:       Machine Learning for Measurement and Inference from Visual
> Data
>
> Abstract: Computer vision systems seek to make sense of the physical world
> and recover scene attributes---like shape, geometry, motion, material, and
> object identity---from visual measurements. As an estimation problem, this
> is ill-posed since the light reflected from objects in a scene carries only
> indirect information about these attributes, and there are typically
> infinitely many physically plausible explanations for any given captured
> image or video. Vision systems must rely heavily on leveraging the
> statistics of natural scenes using machine learning tools, to design not
> just accurate estimation algorithms, but also optimal measurement
> strategies.
>
> In the first part of the talk, I will talk about a structured inference
> approach to the recovery continuous-valued maps of per-point scene
> properties---like depth, surface normals, surface reflectance, motion,
> etc.---from images. Such scene-value maps typically exhibit significant
> spatial structure (planarity, rigidity, etc.), and leveraging this
> structure during inference can lead to more accurate estimates. In this
> context, I will describe recently proposed method for estimating scene
> depth from a single color image, which trains a neural network to produce
> dense probabilistic estimates of different elements of local geometric
> structure, and then harmonizes these estimates to produce a consistent
> depth map.
>
> While a major focus of computer vision is developing effective inference
> methods for images from a traditional camera, in the second half of the
> talk I will discuss whether we can redesign the camera itself to make
> measurements that are optimal for inference. In particular, I will talk
> about a machine-learning approach to automatically "learning" the camera's
> optical design jointly with the computational method for inference. Using
> color imaging as an example task, I will introduce a framework where the
> camera sensor's measurement process is encoded as a neural-network layer,
> whose learnable weights parameterize the possible measurement choices for
> the sensor. This measurement layer is then trained end-to-end with a neural
> network that carries out inference on the corresponding measurements, to
> maximize the accuracy of the final output. I'll show that this approach is
> able to automatically discover a measurement strategy that, when used with
> the jointly learned inference network, significantly outperforms
> traditional sensor designs.
>
>
>
>
> ************************************************************
> ****************************************************
>
> *Research at TTIC Seminar Series*
>
> TTIC is hosting a weekly seminar series presenting the research currently
> underway at the Institute. Every week a different TTIC faculty member will
> present their research.  The lectures are intended both for students
> seeking research topics and adviser, and for the general TTIC and
> University of Chicago communities interested in hearing what their
> colleagues are up to.
>
> To receive announcements about the seminar series, please subscribe to the
> mailing list: https://groups.google.com/a/ttic.edu/group/talks/subscribe
>
> Speaker details can be found at: http://www.ttic.edu/tticseminar.php.
>
> For additional questions, please contact Nathan Srebro at nati at ttic.edu
> <mcallester at ttic.edu>
>
>
>
> Mary C. Marre
> Administrative Assistant
> *Toyota Technological Institute*
> *6045 S. Kenwood Avenue*
> *Room 504*
> *Chicago, IL  60637*
> *p:(773) 834-1757 <(773)%20834-1757>*
> *f: (773) 357-6970 <(773)%20357-6970>*
> *mmarre at ttic.edu <mmarre at ttic.edu>*
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20170106/31efc34b/attachment-0001.html>