|
Defeating Cognitive Sensors
Navy SBIR 2019.3 - Topic N193-143 NAVAIR - Ms. Donna Attick - [email protected] Opens: September 24, 2019 - Closes: October 23, 2019 (8:00 PM ET)
TECHNOLOGY
AREA(S): Air Platform, Battlespace ACQUISITION
PROGRAM: CTO - AI Transformational Thrust Areas The
technology within this topic is restricted under the International Traffic in
Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and
import of defense-related material and services, including export of sensitive
technical data, or the Export Administration Regulation (EAR), 15 CFR Parts
730-774, which controls dual use items. Offerors must disclose any proposed use
of foreign nationals (FNs), their country(ies) of origin, the type of visa or
work permit possessed, and the statement of work (SOW) tasks intended for
accomplishment by the FN(s) in accordance with section 3.5 of the Announcement.
Offerors are advised foreign nationals proposed to perform on this topic may be
restricted due to the technical data under US Export Control Laws. OBJECTIVE:
Develop innovative and operationally efficient approaches to exploit weaknesses
in an adversary�s neural network-based cognitive sensing systems, and by
association, techniques to protect our own systems from deception. DESCRIPTION:
The 2018 National Defense Strategy notes the challenge presented by new
technologies such as big data analytics, artificial intelligence, and autonomy.
Because of the lower barriers of entry, the utilization of these approaches are
moving at accelerating speed. [Ref 1] These technologies are enabling the
development and fielding of a class of cognitive sensing systems. A variety of
neural networking approaches are being employed as the basis for the underlying
machine learning. In many instantiations, these sensing systems train
continuously while operational in an unsupervised fashion in an effort to gain
maximum additivity to a dynamic threat environment. For example, concepts for
true cognitive electronic warfare systems envision a neural network-driven
sensor that �should be able enter into an environment not knowing anything
about adversarial systems, understand them and even devise countermeasures
rapidly�. [Ref 2] Obviously as our adversaries field these systems, we will
seek methods to counter them and in the same vein as we develop the very
adaptive systems, we must understand their vulnerabilities and take steps to
mitigate threats. It has been shown that neural network-based classifiers can
be fooled by subtle undetected adversarial training leading to sensor responses
that are inappropriate or incorrect. These vulnerabilities are widely
recognized and the research community has proposed many defenses that attempt
to detect and defend the network from adversarial training. �Unfortunately, most
of these defenses are not effective at classifying adversarial examples
correctly.� [Ref 3] We must better understand how to exploit these fundamental
blind spots in the training algorithms which adversary might utilize and how to
protect our own system from such deception. Consider undetectable adversarial
training techniques as well as other approaches when designing a solution. PHASE
I: Conceptually develop robust and operationally feasible approaches to defeat
emerging cognitive sensor systems by exploiting weaknesses of these high
data-driven neural network approaches. Perform an unclassified proof of concept
demonstration to show the scientific and technical merit of candidate
approaches. Consider undetectable adversarial training techniques as well as
other approaches in the design. The Phase I effort will include prototype plans
to be developed under Phase II. PHASE
II: Perform detailed development and demonstrate algorithm performance in terms
of ease of operational implementation, effectiveness in degrading system
performance, and adaptability. Consider candidate cognitive sensor systems in
electronic warfare and radar. Consider how own systems might be protected from
such deception while maintaining advantages of cognitive system adaptability.
Demonstrate the algorithms in high-fidelity, operationally representative
scenarios. Prepare a detailed concept of operations describing the
implementation of the approach in the field and potential challenges in its
implementation. PHASE
III DUAL USE APPLICATIONS: Implement algorithmic approaches and concepts to
defeat adversarial cognitive-based systems into Navy operation systems and
concepts of operations. Incorporate methods to protect our own cognitive based
sensors from exploitation. The same general techniques are applicable to a wide
range of data-driven cognitive systems including commercial applications
utilizing internet-based data mining. REFERENCES: 1.
Summary of the 2018 National Defense Strategy of the United States of America. https://dod.defense.gov/Portals/1/Documents/pubs/2018-National-Defense-Strategy-Summary.pdf 2.
Pomerleau, M. �What is the Difference Between Adaptive and Cognitive Electronic
Warfare?� C2/Comms, December 16. 2016. https://www.c4isrnet.com/c2-comms/2016/12/16/what-is-the-difference-between-adaptive-and-cognitive-electronic-warfare/ 3.
Carlini, N. & Wagner, D. �Adversarial Examples Are Not Easily Detected:
Bypassing Ten Detection Methods.� University of California, Berkeley, 1
November 2017.� https://arxiv.org/pdf/1705.07263.pdf KEYWORDS:
Cognitive Sensors; Radar; Electronic Warfare; Electronic Support Measures;
Deception; Behavior Manipulation
|