Skip to yearly menu bar Skip to main content


DL2: Training and Querying Neural Networks with Logic

Marc Fischer · Mislav Balunovic · Dana Drachsler-Cohen · Timon Gehr · Ce Zhang · Martin Vechev

Pacific Ballroom #90

Keywords: [ Systems and Software ] [ Others ] [ Adversarial Examples ]


We present DL2, a system for training and querying neural networks with logical constraints. Using DL2, one can declaratively specify domain knowledge constraints to be enforced during training, as well as pose queries on the model to find inputs that satisfy a set of constraints. DL2 works by translating logical constraints into a loss function with desirable mathematical properties. The loss is then minimized with standard gradient-based methods. We evaluate DL2 by training networks with interesting constraints in unsupervised, semi-supervised and supervised settings. Our experimental evaluation demonstrates that DL2 is more expressive than prior approaches combining logic and neural networks, and its loss functions are better suited for optimization. Further, we show that for a number of queries, DL2 can find the desired inputs in seconds (even for large models such as ResNet-50 on ImageNet).

Live content is unavailable. Log in and register to view live content