SafeAI 2023
Artificial Intelligence Safety 2023

Table of Contents

• Preface
  Summary: There were 48 papers submitted for peer-review to this workshop. Out of these, 28 papers were accepted for this volume, 17 as full papers and 11 as poster presentations.

Session 1: AI/ML Learning, Explainability, Accuracy and Policy Alignment

• Active Reward Learning from Multiple Teachers
  Peter Barnett, Rachel Freedman, Justin Svegliato, Stuart Russell
• REVEALE: Reward Verification and Learning Using Explanations
  Saaduddin Mahmud, Sandhya Saisubramanian, Shlomo Zilberstein
• A Robust Drift Detection Algorithm with High Accuracy and Low False Positives Rate
  Maxime Fuccellaro, Laurent Simon, Akka Zemmari

Session 2: Short Presentations - Safety Assessment of AI-enabled systems

• On Evaluating Adversarial Robustness of Chest X-ray Classification
  Salah Ghamizi, Maxime Cordy, Mike Papadakis, Yves Le Traon
• Towards Multi-timescale Online Monitoring of AI Models
  Fateh Kaakai, Paul-Marie Raffi
• Capabilities for Better ML Engineering
  Chenyang Yang, Rachel Brower-Sinning, Grace A. Lewis, Christian Kästner, Tongshuang Wu

Session 3: AI/ML for Safety Critical Applications: Assurance Cases and Datasets

• Transfer Assurance for Machine Learning in Autonomous Systems
  Chiara Picardi, Richard Hawkins, Colin Paterson, Ibrahim Habli
• Domain-centric ADAS Datasets
  Vaclav Divis, Tobias Schuster, Marek Hruz
• Towards Developing Safety Assurance Cases for Learning-Enabled Medical Cyber-Physical Systems
  Maryam Bagheri, Josephine Lamp, Xugui Zhou, Lu Feng, Homa Alemzadeh

Session 4 - Short Presentations: ML/DL Robustness: GAM and Attack Detection

• Evaluation of GAN Architectures for Adversarial Robustness of Convolution Classifier
  Weimin Zhao, Sanaa Alwidian, Qusay Mahmoud
• Backdoor Attack Detection in Computer Vision by Applying Matrix Factorization on the Weights of Deep Networks
  Khondoker Murad Hossain, Tim Oates

Session 5 - AI Safety Assessment: Failure-Cause Analysis, Assurance, Verification

• A Taxonomic System for Failure Cause Analysis of Open Source AI Incidents
  Nikiforos Pittaras, Sean McGregor
• Towards Safety Assurance of Uncertainty-Aware Reinforcement Learning Agents
  Felippe Schmoeller Roza, Simon Hadwiger, Ingo Thon, Karsten Roscher
• Formal Verification of Tree Ensembles against Real-World Composite Geometric Perturbations
  Valency Oscar Colaco, Simin Nadjm-Tehrani

Session 6: AI Robustness: Adversarial and Attacks Learning

• Critically Assessing the State of the Art in CPU-based Local Robustness Verification
  Matthias König, Annelot Bosman, Holger Hoos, Jan van Rijn
• Towards Understanding How Self-training Tolerates Data Backdoor Poisoning
  Soumyadeep Pal, Ren Wang, Yuguang Yao, Sijia Liu
• Less is More: Data Pruning for Faster Adversarial Training
  Yize Li, Pu Zhao, Xue Lin, Bhavya Kailkhura, Ryan Goldhahn
• Personalized Models Resistant to Malicious Attacks for Human-centered Trusted AI
  Teddy Ferdinan, Jan Kocoń

Session 7: AI Robustness: Deep Reinforcement Learning

• Robustness with Black-Box Adversarial Attack using Reinforcement Learning
  Soumyendu Sarkar, Ashwin Ramesh Babu, Sajad Mousavi, Vineet Gundecha, Sahand Ghorbanpour, Alexander Shmakov, Ricardo Luna Gutierrez, Antonio Guillen, Avisek Naug
• White-Box Adversarial Policies in Deep Reinforcement Learning
  Stephen Casper, Dylan Hadfield-Menell, Gabriel Kreiman
• Bab: A novel algorithm for training clean model based on poisoned data
  Chen Chen, Hong Haibo, Xie Mande, Jun Shao, Tao Xiang
• Safe Reinforcement Learning through Phasic Safety-Oriented Policy Optimization
  Sumanta Dey, Pallb Dasgupta, Soumyajit Dey

Session 8 - Short Presentations: OoD Detection and Uncertainty for ML/DL Safety

• Out-of-Distribution Detection Using Deep Neural Network Latent Space Uncertainty
  Fabio Arnez, Ansgar Radermacher, François Terrier
• Efficient and Effective Uncertainty Quantification in Gradient Boosting via Cyclical Gradient MCMC
  Tian Tan, Carlos Huertas, Qi Zhao
• Safety Assurance with Ensemble-based Uncertainty Estimation and overlapping alternative Predictions in Reinforcement Learning
  Dirk Eilers, Simon Burton, Felippe Schmoeller da Roza, Karsten Roscher

Session 9 - Short Presentations: Methods and Techniques for AI/ML Safety Assessment

• Towards a holistic approach for AI trustworthiness assessment based upon aids for multi-criteria aggregation
  Juliette Mattioli, Henri Sohier, Agnes Delaborde, Gabriel Pedroza, Kahina Amokrane-Ferka, Afef Awadid, Zakaria Chihani, Souhaiel Khalfaoui
• A Framework Quantifying Trustworthiness of Supervised Machine and Deep Learning Models
  Alberto Huertas Celdran, Jan Kreischer, Melike Demirci, Joel Leupp, Pedro Miguel Sanchez, Muriel Franco, Gerome Bovet, Gregorio Martinez Perez, Burkhard Stiller
• Standardizing the Probabilistic Sources of Uncertainty for the sake of Safety Deep Learning
  Axel Brando, Isabel Serra, Enrico Mezzetti, Francisco J. Cazorla, Jaume Abella