Security and Privacy in the (golden) Age of AI
Any computer scientist, software engineer or information technology specialist will become a user and practitioner of artificial intelligence (AI) technology. Quite some of the longstanding promises of AI, are nowadays being harvested in multiple application domains. This is not different for the broad area of Security & Privacy, obviously of critical importance in a world that has to be concerned about cybersecurity – more than ever. Even though great progress has been achieved, much more is required. This obviously is an exciting setting for PhD candidates who target advanced research goals in Security & Privacy and in Machine Learning.
Praktische info:
Inschrijven?
- Prijs: 300€ : early bird fee tot 16 augustus | 400€ vanaf 17 Augustus
georganiseerd door:
Leertraject
In short, there is a strong and exciting interplay between Security & Privacy and AI. On the one hand, specific S&P challenges, ranging from various types of anomaly detection (e.g. network intrusion, malware, data exfiltration attacks), spam and social engineering detection, to biometric authentication and user behaviour analytics, will benefit from novel research in AI, Machine Learning (ML) and Deep Learning (DL). On the other hand, ML solutions encounter challenges by adversaries: ML applications need strengthening for real world (malicious) operating environments. Last but not least, adversaries are exploiting the latest advances in ML and DL themselves to tweak their attacks (e.g. phishing campaigns, misinformation with deep fakes) and increase the likelihood that users become a victim of spoofing, fraud or identity theft.
KU Leuven & Cybersecurity Flanders have invited and gathered a group of experts that will teach, reflect and challenge us all, by addressing major challenges in this versatile and rapidly evolving domain.
Themes
- Case studies in enhancing Cybersecurity through ML.
- Grand challenges for S&P and Cybersecurity, and the role of AI in the endeavour.
- Inroads in discovering new attacks and countermeasures for ML.
- Basic building blocks in protecting data, highlighting data sharing for federated ML.
- Lessons learned and insights in the collection, management and processing of critical (real-world) data.
- Going forward: reflections and lessons learned, best practices.
- Poster presentations by PhD students, feedback sessions in small workshops.
Day by Day
Day 1: Frontiers in AI (and get together)
The technological revolution of AI and recent development in deep learning have impacted many aspects of modern life, including finance, medical diagnosis, fully automated driving, and personal assistance. This module will start with a state-of-the-art overview of essential ML concepts to accommodate a broad audience with diverse backgrounds. The opening lectures will mainly highlight notable AI advancements and salient developments in light of an application domain other than S&P: sports.
The first day will also offer the opportunity for attendees to briefly introduce their own research interests to peer researchers and experts in the field. This will occur in an informal atmosphere with a poster session.
Day 2: Trustworthy Data in ML – and Critical Applications Related to S&P
Vast amounts of data have enabled researchers and companies to harness the power of ML and deep learning. The ability to learn from data and uncover hidden relationships is limited by the quantity of available data, but obviously even more by the quality of these data. Impactful applied research demands for real world, reliable data that must be treated carefully – to state the least.
In the opening session we will zoom into a couple of case studies that have built on real world data, e.g. from government and law enforcement parties. Next, we will address the challenge of ensuring robustness in a context of adversaries, while gradually turning from classical ML research to applications on the domain of security and privacy.
This day will thus cover data related aspects to effectively build and deploy machine learning systems. Starting from data quality and trustworthiness concerns of real world data, the school further explores the risks and challenges of malicious data perturbations and other threats that emerged after the rise of adversarial machine learning about two decades ago.
The programme will then gradually build up into assessing the implications for real world applications, for example 5G infrastructures, and cases in S&P.
Day 3: Towards Best Practices – AI for cybersecurity
Machine learning applications are expanding in many areas at a rapid pace and scale. Nonetheless, the AI that is powering these complex applications can make mistakes too. Furthermore, the implications in the context of cybersecurity applications definitely are not entirely understood.
On day 3, we further explore the field, first by addressing about pitfalls and lessons learned on data pre-processing. Furthermore, we will cover techniques and best practices to design, build and validate Machine Learning pipelines for cybersecurity systems and applications. In addition, lessons learned will be consolidated in a keynote. In addition to the more traditional lectures, participants will translate insights into their own agenda during a workshop-style session. The last session of the day will be a panel where we address both lessons learned, new challenges and advice for researchers and PhD Students.
Day 4: Data Protection
… and the research road ahead for Security and Privacy in the age of AI
Machine learning and deep learning applications are data hogs. This data is an important asset that often needs to be protected, not only because companies may lose their competitive edge if rivals can otherwise build similar machine learning pipelines from that same data, but also because of privacy concerns when the data contains sensitive or personal information. Data not only needs to be protected at rest, but also when in transit and processed in possibly untrusted environments, such as MLaaS frameworks in the cloud. The first lectures will explore the benefits and challenges of different cryptographic techniques to protect data while training and/or evaluating ML models, the threats against which these techniques do and do not offer protection, etc.
Participants will share lessons learned before concluding the summer school with a session on the road ahead for security and privacy in the golden age of AI.
Gerelateerde opleidingen
