Timezone: »
Generative Adversarial Networks (GANs) struggle to generate structured objects like molecules and game maps. The issue is that structured objects must satisfy hard requirements (e.g., molecules must be chemically valid) that are difficult to acquire from examples alone. As a remedy, we propose Constrained Adversarial Networks (CANs), an extension of GANs in which the constraints are embedded into the model during training. This is achieved by penalizing the generator proportionally to the mass it allocates to invalid structures. In contrast to other generative models, CANs support efficient inference of valid structures (with high probability) and allows to turn on and off the learned constraints at inference time. CANs handle arbitrary logical constraints and leverage knowledge compilation techniques to efficiently evaluate the disagreement between the model and the constraints. Our setup is further extended to hybrid logical-neural constraints for capturing very complex constraints, like graph reachability. An extensive empirical analysis shows that CANs efficiently generate valid structures that are both high-quality and novel.
Author Information
Luca Di Liello (University of Trento)
Pierfrancesco Ardino (University of Trento)
Jacopo Gobbi (University of Trento)
Paolo Morettin (University of Trento)
Stefano Teso (University of Trento)
Andrea Passerini (Università degli Studi di Trento)
More from the Same Authors
-
2021 Spotlight: Interactive Label Cleaning with Example-based Explanations »
Stefano Teso · Andrea Bontempelli · Fausto Giunchiglia · Andrea Passerini -
2022 : GlanceNets: Interpretable, Leak-proof Concept-based Models »
Emanuele Marconato · Andrea Passerini · Stefano Teso -
2022 : GlanceNets: Interpretable, Leak-proof Concept-based Models »
Emanuele Marconato · Andrea Passerini · Stefano Teso -
2022 Spotlight: GlanceNets: Interpretable, Leak-proof Concept-based Models »
Emanuele Marconato · Andrea Passerini · Stefano Teso -
2022 Poster: Semantic Probabilistic Layers for Neuro-Symbolic Learning »
Kareem Ahmed · Stefano Teso · Kai-Wei Chang · Guy Van den Broeck · Antonio Vergari -
2022 Poster: GlanceNets: Interpretable, Leak-proof Concept-based Models »
Emanuele Marconato · Andrea Passerini · Stefano Teso -
2021 Poster: Interactive Label Cleaning with Example-based Explanations »
Stefano Teso · Andrea Bontempelli · Fausto Giunchiglia · Andrea Passerini -
2021 Poster: A Compositional Atlas of Tractable Circuit Operations for Probabilistic Inference »
Antonio Vergari · YooJung Choi · Anji Liu · Stefano Teso · Guy Van den Broeck -
2021 Oral: A Compositional Atlas of Tractable Circuit Operations for Probabilistic Inference »
Antonio Vergari · YooJung Choi · Anji Liu · Stefano Teso · Guy Van den Broeck -
2020 Poster: Probabilistic Inference with Algebraic Constraints: Theoretical Limits and Practical Approximations »
Zhe Zeng · Paolo Morettin · Fanqi Yan · Antonio Vergari · Guy Van den Broeck -
2020 Spotlight: Probabilistic Inference with Algebraic Constraints: Theoretical Limits and Practical Approximations »
Zhe Zeng · Paolo Morettin · Fanqi Yan · Antonio Vergari · Guy Van den Broeck -
2016 : Constructive Layout Synthesis via Coactive Learning »
Paolo Dragone · Andrea Passerini -
2016 : Introduction »
Fabrizio Costa · Andrea Passerini · Thomas Gärtner · Francois Pachet -
2016 Workshop: Constructive Machine Learning »
Fabrizio Costa · Thomas Gärtner · Andrea Passerini · Francois Pachet -
2013 Workshop: Constructive Machine Learning »
Thomas Gaertner · Roman Garnett · Andrea Passerini -
2008 Poster: Predicting the Geometry of Metal Binding Sites from Protein Sequence »
Paolo Frasconi · Andrea Passerini -
2008 Spotlight: Predicting the Geometry of Metal Binding Sites from Protein Sequence »
Paolo Frasconi · Andrea Passerini