Abstract: Machine learning is increasingly being used for automated decisions in applications such as health care, finance, and cyber security. In these critical environments, attackers have strong incentives to manipulate the results and models generated by machine learning algorithms. The area of adversarial machine learning studies the effect of adversarial attacks against machine learning models and aims to design robust defense algorithms. In this talk I will describe our work on applying machine learning to detect advanced adversaries in cyber networks. I will introduce several types of attacks against machine learning and discuss their impact on real-world applications. I will also mention a number of challenges and open problems in securing machine learning in critical adversarial environments.
Bio: Alina Oprea is an Associate Professor at Northeastern University in the Khoury College of Computer Sciences. She joined Northeastern University in Fall 2016 after spending 9 years as a research scientist at RSA Laboratories. Her research interests are broadly in cyber security, with a focus on adversarial machine learning, threat detection, cloud security, and applied cryptography. She is the recipient of the Technology Review TR35 award for research in cloud security in 2011 and the recipient of the Google Security and Privacy Award in 2019. Alina serves currently as Program Committee co-chair of the IEEE Security and Privacy Symposium, 2021, and as Associate Editor of the ACM Transactions of Privacy and Security (TOPS) journal.
Program Schedule (All times in Eastern Standard Time (GMT-5))
Date: Monday, 7 December 2020
|Start||End||Event||Authors or Speaker||Affiliation|
|09:00 am||10:00 am||Keynote: Towards Resilient Machine Learning in Adversarial Environments||Alina Oprea||Northeastern University|
|10:00 am||10:10 am||Break + Networking|
|10:10 am||10:35 am||Paper 1 - Optimizing Information Loss Towards Robust Neural Networks||Philip Sperl, Konstantin Böttinger||Fraunhofer AISEC|
|10:35 am||11:00 am||Paper 2 - Efficient Black-Box Search for Adversarial Examples using Relevance Masks||Katja Auernhammer, Ramin Tavakoli Kolagari, Felix Freiling||Nuremberg Institute of Technology|
|Lunch + Networking|
|11:50 am||12:15 pm||Paper 3 - Defending Against Adversarial Denial-of-Service Data Poisoning Attacks||Simon Roschmann, Nicolas Müller, Konstantin Böttinger||Fraunhofer AISEC|
|12:15 pm||12:40 pm||Paper 4 - WikipediaBot: Machine Learning Assisted Adversarial Manipulation of Wikipedia Articles||Filipo Sharevski||DePaul|
|12:40 pm||1:00 pm||Break + Networking|
|1:00 pm||1:25 pm||Paper 5 - Program Behavior Analysis and Clustering using Performance Counters||Sai Praveen Kadiyala, Truong Huu Tram, Sukruth Kartheek||Agency for Science, Technology, and Research (A*STAR), Singapore|
|1:25 pm||1:50 pm||Paper 6 - A Statistical Approach to Detecting Low-Throughput Exfiltration through the Domain Name System Protocol||Emily Joback, Kenneth Alperin||MIT Lincoln lab|
|1:50 pm||2:00 pm||Break + Networking|
|2:00 pm||2:25 pm||Paper 7 - The Semantic Processing Pipeline: Quantifying the Network-Wide Impact of Security Tools||Katarzyna Olejnik||Raytheon|
|2:25 pm||2:50 pm||Paper 8 - Why Deep Learning Makes it Difficult to Keep Secrets in the FPGA-as-a-Service Setting||Yang Yu||KTH - Royal Institute of Technology, Sweden|
|2:50||3:15 pm||Wrap up, Best Paper presentation, networking|
2020 Call for Submissions
About the DYNAMICS Workshop
The 2020 DYnamic and Novel Advances in Machine Learning and Intelligent Cyber Security (DYNAMICS) Workshop will be held on Monday, December 7 th. The workshop will be co-located with the 2020 Annual Computer Security Applications Conference (ACSAC), held at the AT&T Hotel and Conference Center in Austin, Texas. Due to the evolving COVID situation, DYNAMICS will be held virtually.
Machine learning has become critical to the evolution and sustainability of cybersecurity. While the theoretical objectives and principles behind cybersecurity are still valid, traditional technologies that require humans in the loop to read log files, triage alerts, and harden devices are neither fast enough nor scalable enough to meet the demands of modern networks and attacks. While network traffic and devices have grown by orders of magnitude, the ability for operators to triage the resulting alerts has not.
The sophistication of threats has also increased substantially. Sophisticated zero-day attacks go undetected for months at a time. Attacks take place over extended periods, effectively outwaiting traditional intrusion detection technology. Worse, new attack tools and strategies can now be developed using adver ial machine learning techniques, requiring rapid co-evolution of defenses that match the speed and sophistication of machine learning-based offensive techniques.
This is intended to focus on novel applied and theoretical work that combines machine learning techniques such as reinforcement learning, adversarial machine learning, and deep learning with significant problems in cybersecurity. We consider both offensive and defensive applications of machine learning to security.
Technical Paper Submissions
The DYNAMICS Workshop invites submissions of original, previously unpublished technical papers, posters, and panels on research in machine learning and cybersecurity. Papers should be between 5 and 12 pages, and should use the 2020 ACM Proceedings Template: https://www.acm.org/publications/proceedings-template , using the [sigconf, anonymous] options. Submissions will be evaluated using a standard peer review process. While authors may wish to align their submissions with one of the suggested topics below, submissions on other topics related to the workshop theme are welcome. Papers should emailed at firstname.lastname@example.org .
A DYNAMICS Forum is a 1-2 hour extended discussion on a specific topic of interest to the DYNAMICS machine learning and cyber security community. A forum may focus on a specific technical problem, a policy issue, a social concern, or another DYNAMICS-related topic that you believe should be explored in depth. The intent of the forum format is to bring together a community that will not only explore your topic deeply within the context of the DYNAMICS Workshop itself, but that will have the potential to persist and grow beyond the workshop, in order to develop collaborative solutions over the long term. A DYNAMICS Forum is led by one or more moderators, who facilitate a discussion with a fully engaged audience.
If you would like to submit a DYNAMICS Forum proposal, please send an abstract of no more than 2 pages to email@example.com. Please be sure to include your proposed topic, moderators, a statement of why your idea is relevant and important to the DYNAMICS community, proposed length, and a high-level outline with your major discussion topics. Your submission will be evaluated for inclusion in the workshop based on its relevance to the workshop theme, the quality of the submission, and the availability of space in the workshop schedule. While proposed topics may be aligned with one of the suggested topics below, submissions on other topics related to the workshop theme are welcome.
Panel submissions are invited on topics of interest to the DYNAMICS machine learning and cyber security community. To submit a panel, please send an abstract of up to 1 page to firstname.lastname@example.org. Please be sure to include your proposed topic, panelists, panel chair, affiliations, and position statements for each panelist. DYNAMICS panels follow the same format as those of the main ACSAC conference. Extensive information about ACSAC’s panel requirements can be found at https://www.acsac.org/2020/submissions/panels/ .
Poster submissions are invited on topics of interest to the DYNAMICS machine learning and cyber security community. Posters provide a way for workshop attendees to present early stage, ongoing research that is not yet ready for submission as a peer reviewed paper, Poster presenters also gain feedback from conference and workshop attendees, and spark discussion among conference and workshop participants. Poster dimensions can be up to 36×48 inches (91x122 cm). Poster abstracts are not peer reviewed. Accepted abstracts will be made available on the workshop website prior to the event, but they will not be included in the workshop proceedings. To submit a poster, submit an e-mail with a PDF of your draft poster to email@example.com. For questions on posters, please contact firstname.lastname@example.org.
A Lightning Talk is timed, 5-minute talk on a topic of interest to the DYNAMICS machine learning and cyber security community. While lightning talks may be given on works in progress, or other topics of relevance to the workshop, you can even use a lightning talk to ask a question, find a community of shared interest on a topic, engage people in an issue, ask a question, or solicit feedback! A good lightning talk is fast paced, engaging, and high energy, and can use any desired presentation format.
Note that although Lightning Talk abstracts are reviewed by the DYNAMICS Workshop committee for relevance, they are not peer reviewed, and will not appear in the workshop proceedings. To submit a Lightning Talk, please e-mail your proposed topic to email@example.com
Papers that have been accepted by the DYNAMICS workshop will be published in the workshop proceedings.
By submitting a paper, DYNAMICS Forum, poster, panel, or lightning talk to the DYNAMICS workshop, you agree that if your submission is accepted, one or more of the submission’s authors will present the final version of the submission at the workshop.
Technical paper submission deadline:
Technical paper acceptance notification:
October 31st, 2020
Final technical paper PDF submission deadline:
November 13th, 2020 (11:59 PM Eastern Time)
Forum submission deadline:
November 13th, 2020 (11:59 PM Eastern Time)
Forum acceptance notification:
Rolling acceptances through the deadline
Panels, posters, and lightning talks
Submissions accepted through November 13th, 2020
Attacking and Defending Machine Learning-Based Systems, Models, and Data Sets
Data Generation and Preparation:
Feature Finding and Event Analysis:
Adversarial Machine Learning for Cybersecurity:
Machine Learning-Based Defense and Response:
Machine Learning-Based Offensive Techniques
How to Contact the Workshop Organizers
If you have questions related to the workshop, please e-mail them to firstname.lastname@example.org.
2020 DYNAMICS Workshop Organizers