[Colloquium] 5/14 Talks at TTIC: José Oramas, University of Leuven

[Alicia McClarin]

When:     Tuesday, May 14th at *11:00 am*

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

*Who:*       José Oramas, University of Leuven


*Title: *"Visual Explanation by Interpretation: Improving Visual Feedback
Capabilities of Deep Neural Networks"


*Abstract: *Interpretation and explanation of deep models is critical
towards wide adoption of systems that rely on them. In this paper, we
propose a novel scheme for both interpretation as well as explanation in
which, given a pre-trained model, we automatically identify internal
features relevant for the set of classes considered by the model, without
relying on additional annotations. We interpret the model through average
visualizations of this reduced set of features. Then, at test time, we
explain the network prediction by accompanying the predicted class label
with supporting visualizations derived from the identified features. In
addition, we propose a method to address the artifacts introduced by
stridded operations in deconvNet-based visualizations. Moreover, we
introduce an8Flower, a dataset specifically designed for objective
quantitative evaluation of methods for visual explanation. Extensive
experiments on various datasets show that our method produces detailed
explanations with good coverage of relevant features of the classes of
interest.


*Bio:* José Oramas is a postdoctoral researcher at the Center for
Processing Speech and Images at the Department of Electrical Engineering
from the KU Leuven. In 2015, he received the PhD in Engineering Sciences
from the KU Leuven, Belgium. During his Ph.D. he worked in problems related
to object detection, pose estimation, relational learning and collective
classification. In 2008, he received the degree of Computer Engineering
with a major in Multimedia Systems from the Escuela Superior Politécnica
del Litoral (ESPOL), Ecuador.
During the last 10 years he has conducted research on problems at the
intersection of computer vision and artificial intelligence. More
specifically, on understanding how groups of visual elements (objects,
object-parts, image regions, trajectories, etc.) interact and how the
relationships between them can be exploited to improve artificial visual
perception problems. More recently, his interests are focused towards
making AI systems intelligible, i.e. being able to be understood by humans,
by using these representations as means to provide insights on what a model
has actually learned (interpretation), and justify the decisions made by a
model (explanation).


*Host:     *Greg Shakhnarovich <greg at ttic.edu>
-- 
*Alicia McClarin*
*Toyota Technological Institute at Chicago*
*6045 S. Kenwood Ave., **Office 518*
*Chicago, IL 60637*
*www.ttic.edu* <http://www.ttic.edu/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20190508/652602dc/attachment-0001.html>