The Verifying Learning AI Systems (VeriLearn) Workshop

Schedule

09:15	Opening
09:30	Policy-Specific Abstraction Predicate Selection in Neural Policy Safety Verification Marcel Vinzent (Saarland University); MIn Wu (Stanford University); Haoze Wu (Stanford University); Joerg Hoffmann (Saarland University)
09:50	Multi-class Robustness Verification for Tree Ensembles Lorenzo Cascioli (KU Leuven); Laurens Devos (KU Leuven); Jesse Davis (KU Leuven)
10:10	Safe Reinforcement Learning via Probabilistic Logic Shields Wen-Chi Yang (KU Leuven); Giuseppe Marra (KU Leuven); Gavin Brian Rens (Stellenbosch University); Luc De Raedt (KU Leuven)
10:20	VAEL: Bridging Variational Autoencoders and Probabilistic Logic Programming Eleonora Misino (University of Bologna); Giuseppe Marra (KU Leuven); Emanuele Sansone (KU Leuven)
10:30	Coffee Break
11:00	Towards Framed Autonomy Invited talk by Giuseppe De Giacomo Abstract: Autonomy is one of the grand objectives of AI. It aims at empowering AI systems with the ability to deliberate how to act in the world autonomously without being preprogrammed to do so. Obviously, empowering an AI system with the ability to self-deliberate its own behavior carries significant risks, and regulating the autonomy of AI systems is a crucial challenge for future AI. One interesting concept in this context is that of "framed autonomy", a term recently introduced in an AI-Augmented Business Process Management Manifesto, that indicates that the system is allowed to behave with maximally permissive freedom within the boundaries of its current "frame". This concept is related to reactive synthesis in Formal Methods, supervisory control in Discrete-Event Systems, and the characterization of "capabilities" (knowing-how) in Reasoning about Action, Goal Reasoning, and Theory of Mind. In this talk, we will concretely look at framed autonomy through the lens of Linear Temporal Logics on finite traces and DFA-based synthesis, linking it to winning regions in games on graphs, nondeterministic strategies, and characterization of the set of winning strategies.
12:00	Bi-Objective Lexicographic Optimization in Markov Decision Processes with Related Objectives Damien Busatto-Gaston (Université Paris Est Créteil); Debraj Chakraborty (Masaryk University); Anirban Majumdar (Université Libre de Bruxelles); Sayan Mukherjee (Université Libre de Bruxelles); Guillermo Perez (UAntwerpen); Jean-François Raskin (Université Libre de Bruxelles)
12:10	Deep Neural Network Benchmarks for Selective Classification Andrea Pugnana (Scuola Normale Superiore); Lorenzo Perini (KU Leuven); Jesse Davis (KU Leuven); Salvatore Ruggieri (University of Pisa - IT)
12:30	Lunch
13:40	Panel or 2nd Invited Talk
14:40	DeepSaDe: Learning Neural Networks that Guarantee Domain Constraint Satisfaction Kshitij Goyal (KU Leuven); Sebastijan Dumancic (TU Delft); Hendrik Blockeel (KU Leuven)
15:00	Coffee Break
15:30	Assessing and improving the robustness of Binarized Neural Networks Benoît Ronval (Université Catholique de Louvain, BE); Siegfried Nijssen (Université Catholique de Louvain, BE)
15:50	Towards a Unified Framework for Probabilistic Verification of AI Systems Paolo Morettin (University of Trento); Andrea Passerini (University of Trento); Roberto Sebastiani (University of Trento)
16:10	Formally-Sharp DAgger for MCTS: Lower-Latency Monte Carlo Tree Search using Data Aggregation with Formal Methods Damien Busatto-Gaston (Université Paris Est Créteil); Debraj Chakraborty (Masaryk University); Jean-François Raskin (Université Libre de Bruxelles); Guillermo Perez (UAntwerpen)
16:20	Faster Robustness Verification by Exploiting Repeated Structure in Adversarial Examples Lorenzo Cascioli (KU Leuven); Laurens Devos (KU Leuven); Jesse Davis (KU Leuven)
16:40	Visualizing Deep Neural Networks with Topographic Activation Maps Valerie Krug (Otto-von-Guericke-University Madgeburg); Raihan Kabir Ratul (MOTOR Ai); Christopher Olson (OVGU Magdeburg); Sebastian Stober (Otto von Guericke University)
16:50	Closing