We are a research lab focused on investigating probabilistic models and programs that are reliable and efficient. We are based at the School of Informatics, University of Edinburgh within the Institute for Adaptive and Neural Computations (ANC).
latest news
- [26th Oct 2023]
- [19th Oct 2023]
AV talks at the Simons Institute (UC Berkeley) about circuits with negative parameters.
- [16th Oct 2023]
- [1st Oct 2023]
New preprint on how to represent and learn mixture models with negative coefficients.
- [22nd Sep 2023]
Two papers accepted at NeurIPS 2023: on how to scale knowledge graph embeddings and treat them as generative models (as
oral !) and on formalizing, understanding and mitigating reasoning shortcuts in neuro-symbolic AI.
selected works
A systematic framework in which tractable inference routines can be broken down into smaller and composable primitives operating on circuit representations. NeurIPS 2021 oral (top 0.6%)
We design a differentiable layer that can be plugged into any neural network to guarantee that predictions are always consistent with a set of predefined symbolic constraints and can be trained end-to-end. NeurIPS 2022
KGE models such as CP, RESCAL, TuckER, ComplEx can be re-interpreted as circuits to unlock their generative capabilities, scaling up learning and guaranteeing the satisfaction of logical constraints by design. NeurIPS 2023 oral (top 0.6%)
We propose to build (deep) subtractive mixture models by squaring circuits. We theoretically prove their expressiveness by deriving an exponential lowerbound on the size of circuits with positive parameters only. arXiv 2023
We highlight a weakness of low-rank linear multi-label classifiers: they can have meaningful outputs that cannot be predicted. We design a classifier which guarantees that sparse outputs can be predicted while using less trainable parameters. arXiv 2023