CfP: MLJ Special Issue "Machine Learning in Adversarial Environments"

[Pavel Laskov]

*** Apologies for multiple posting ***

========================== CALL FOR PAPERS ===========================
Special Issue of the Machine Learning Journal
"Machine Learning in Adversarial Environments"
----------------------------------------------------------------------

Guest Editors:

Pavel Laskov (Fraunhofer FIRST and University of Tuebingen, Germany)
Richard Lippmann (MIT Lincoln Laboratory, USA)


Motivation and Scope:

Machine learning techniques are often used in environments where
adversaries can consciously act to limit or prevent accurate
performance. A classical example is spam filtering where spammers
tailor messages to avoid the most recent spam detection techniques.
Further examples of adversarial learning arise in the field of
computer security where there is an escalating competition between
detection and evasion techniques for various types of malware. In
general, one can expect that whenever machine learning is used to
provide protection from some illegal activity, adversaries will
deliberately attempt to circumvent these approaches.

The behavior of learning systems in adversarial environments is
currently not well understood. Previous work has focused on
identifying application-specific attacks against machine
learning as well as developing algorithms that are robust in
certain scenarios, e.g. feature deletion at test time. A more general
understanding of theoretical foundations and cross-application
commonalities is essential for further progress in this field.

We would like to invite submissions to a Special Issue of the Machine
Learning Journal on "Machine Learning in Adversarial
Environments". Submissions in all relevant fields are encouraged,
especially addressing the following issues:

- Analysis of potential threats to machine learning algorithms.
How can an adversary influence, in general and in specific
applications, various stages of learning algorithms?

- Theoretical foundations of adversarial learning. How robust, in
terms of generalization ability, are current machine learning
algorithms against common attacks? How much effort does an adversary
require to achieve certain evasion or subversion goals? What are the
possible tradeoffs between accuracy and robustness?

- Specialized learning algorithms for adversarial environments.
How can machine learning algorithms deal with particular attacks
against feature extraction, training, and classification? How
effective are these techniques in practice? What is the price to be
paid in terms of performance and accuracy?

- Applications of adversarial learning. How can learning
techniques be applied in fields typical for adversarial scenarios
such as computer security, spam filtering, and fraud detection? What
are other potentially relevant application domains?

An initial discussion of key issues related to adversarial learning
took place at the NIPS 2007 Workshop on Machine Learning in
Adversarial Environments for Computer Security. Workshop participants
are especially encouraged to submit full-scale contributions based on
their work. However, the Special Issue is open to a wide range of
contributions addressing theoretical and practical aspects of machine
learning in adversarial environment.

Further information on the NIPS 2007 workshop can be found at:
http://mls-nips07.first.fraunhofer.de


Important Dates:

Papers due: 31 March 2008
Author notification: 15 June 2008
Revisions due: 15 July 2008
Final decisions: 22 August 2008


Submission Guidelines:

Submissions must follow the publication guidelines set forth by the
Machine Learning Journal. Templates and style files are available at:
ftp://ftp.springer.de/pub/tex/latex/svjour3/global.zip

Further information for authors can be found at:
http://pages.stern.nyu.edu/$\sim$fprovost/MLJ/info-for-authors.html

Manuscripts should be submitted via Springer's submission management
system located at:
http://mach.edmgr.com

The article type "ML in Adversarial Environment" should be chosen
when submitting a manuscript to this special issue. Note that
submissions in PDF format will only be accepted for the review stage;
editable source files will be required for the final versions of
accepted manuscripts. Please do not send submissions per email
directly to the guest editors.

Manuscripts submitted to the special issue must contain unpublished
original research. If related work has been previously published, the
submitted manuscript must involve significant revision or
extension. Manuscripts submitted to Machine Learning must not be
concurrently under review at any other journal.