Tuesday, December 18, 2012

[Comp-neuro] CFP: Journal Special Issue on Never Ending Visual Information Learning

Journal of Electrical and Computer Engineering (JECE)

Special Issue on "Never Ending Visual Information Learning"


Much of the recent history of computer vision and visual data
understanding has focused on very simplified settings. Many approaches make
restrictive assumptions such as stationary environments, fixed and known
number of categories to be recognized or learnt, and enough computational
resources to learn and store the necessary models. Nevertheless, in real
applications, the need for a general framework to learn from, and adapt to,
a nonstationary environment can hardly be overstated. Given new data, such
a framework would allow us to learn any novel knowledge, reinforce existing
knowledge that is still relevant, and forget what may no longer be
relevant. Most existing models, which are variations on the static learning
schemes, cannot cope with many real-life challenges, limiting their
application to constrained environments. The models that try to address
these questions have a complexity that always grows over time, making them
impractical for a perpetual learner.

We propose to collect the initial steps towards a much more challenging
never ending learning system from visual data. One of the focuses is on
adaptive learning algorithms for evolving data applied to the development
of robust and flexible video object tracking systems. For instance, we
expect to receive proposals that tackle the problem of tracking objects in
multiple video streams, where objects may have different appearances when
recorded by different cameras. We also expect proposals of models that
implicitly learn the number and identity of objects depicted in a video
stream and adapt to their evolution in time, both in terms of appearance
and number. These characteristics should be attained with bounded
computational resources.

Considering the strong relationship between the performance of different
learning algorithms and the use of appearance models, we will welcome both
perspectives of the problem in video streams. The computer vision community
has been more focused on the representation problem and on finding the
right representation primitive, while the machine learning community has
been addressing the learning issues. We will welcome integrated views and
solutions for video-based applications. Finally, it is likely that
redundancy and feedback will play a significant role in the development of
robust visual learning systems. Potential topics include, but are not
limited to:

* Visual descriptors
* Video analytics in camera networks
* Online and active learning tested on video data
* Learning from video data streams
* Learning in nonstationary environments
* Benchmark data sets and performance metrics
* Application: visual trackingApplication: robotics
* Application: automation
* Application: autonomous driving
* Application: security

Before submission authors should carefully read over the journal's Author
Guidelines, which are located at

Prospective authors should submit an electronic copy of their complete
manuscript through the journal Manuscript Tracking System at
http://mts.hindawi.com/author/submit/journals/jece/nevil/ according to the
following timetable:

- Manuscript Due: Friday, 29 March 2013
- First Round of Reviews: Friday, 21 June 2013
- Publication Date: Friday, 16 August 2013

Lead Guest Editor:

- Jaime S. Cardoso, Faculdade de Engenharia, University of Porto, Porto,

Guest Editors:

- Gustavo Carneiro, School of Computer Science, University of Adelaide,
Adelaide, SA, Australia
- Guilherme Barreto, Universidade Federal do Ceará, Fortaleza, CE, Brazil

Comp-neuro mailing list

No comments:

Post a Comment