Workshop on Safe and Robust Machine Learning

Event Timing: 27 October, 2023
Event Address: Craig Room in the James Watt Centre, Heriot-Watt University, Edinburgh, Scotland

The workshop will explore the state-of-the-art methods in ensuring safety and robustness of machine learning systems. The workshop is funded by the EPSRC grant AISEC: AI Secure and Explainable by Construction.

The workshop will be live-streamed via Zoom:

We ask all face-2-face participants to register, but online attendance requires no registration.

9.00 — 10.00 Session 1. Robustness from ML perspective. Chair: Katya Komendantskaya
Daniel Kienitz, Heriot-Watt University: Neural network robustness from a geometric point of view
Antonia Marcu, Southampton University: On Pitfalls of Measuring Occlusion Robustness through Data Distortion
10.00 — 10.30 Coffee and Social interaction
10.30 — 12.30 Session 2. Neuro-Symbolic Architectures. Chair: Antonia Marcu
Jonathon Hare, Southampton University: Incorporating constraints in learning machines through model architecture
Antonio Vergari, Edinburgh University: Provably reliable deep learning by design
Vaishak Belle, Edinburgh University: A semantic objective function: unifying some types of loss functions
Natalia Slusarz, Heriot-Watt University: The Logic of Differentiable Logics: how to implement property-driven training.
12.30 — 14.00 Lunch
14.00 — 15. 00 Session 3. Large Language Models and Verification Benchmarks. Chair: Jonathon Hare
Edoardo Manino, Manchester University. Checking the checkers: neural network verification benchmarks with known ground truth
Marco Casadio and Tanvi Dinkar, Heriot-Watt University. Verification for NLP applications
15.00 — 15.30 Coffee and Social interaction
15.30 — 16.30 Session 4. Programming Language Support for Safe AI. Chair: Kathrin Stark
Julien Girard-Satabin, CEA France. Multiple provers, multiple answers: formulating and solving verification problems for diverse artificial intelligence components using CAISAR
Ekaterina Komendantskaya, Heriot-Watt and Southampton Universities. Whats and Whys of Neural Network Verification.
16.30 — 17.30 Session 5. Panel Discussion: Application areas for Neural Network Assurance. Chair: Ekaterina Komendantskaya
Alessandro Suglia, Assistant Professor, Head of Visual Dialogue at Alana AI: Machine Learning in Natural Language Processing, ChatBots and LLMs
Oleksandr Letychevsky, Dep. of Computer Science, Heriot-Watt: — Neural Networks in Security Applications
Michael John Williams, SLB Cambridge: — Neural Networks in Engineering
Vasiliy Demyanov, School of Energy, Geoscience, Infrastructure and Society, Heriot-Watt: Neural networks in Geoscience
Yuhui Lin, Dep. Computer Science, Heriot-Watt University: Autonomous Systems and Autonomous Car Legislation
18.00 — Drinks and dinner
(registration required, details sent to registered participants, email organisers if in doubt)

Travel advice: The workshop will be organised at Heriot-Watt campus. It is a 15 mins ride from Edinburgh Airport (Uber taxi price range £15-20), and approx. 40 -50 mins Bus drive from Edinburgh City Center. On campus, we recommend the hotel Mariott: