I gave a chat, entitled "Explainability for a provider", at the above occasion that discussed anticipations pertaining to explainable AI And just how may be enabled in programs.
Weighted model counting normally assumes that weights are only specified on literals, generally necessitating the need to introduce auxillary variables. We consider a completely new technique depending on psuedo-Boolean capabilities, resulting in a far more standard definition. Empirically, we also get SOTA outcomes.
The Lab carries out exploration in artificial intelligence, by unifying Finding out and logic, with a modern emphasis on explainability
For anyone who is attending NeurIPS this year, you could have an interest in testing our papers that touch on morality, causality, and interpretability. Preprints can be found within the workshop site.
Our paper (joint with Amelie Levray) on learning credal sum-product networks has been recognized to AKBC. These networks, as well as other kinds of probabilistic circuits, are appealing because they warranty that selected forms of probability estimation queries is usually computed in time linear in the dimensions in the community.
I gave a talk on our the latest NeurIPS paper in Glasgow although also masking other strategies on the intersection of logic, Discovering and tractability. Due to Oana for that invitation.
The challenge we tackle is how the learning need to be outlined when There exists missing or incomplete data, bringing about an account determined by imprecise probabilities. Preprint right here.
A journal paper has become accepted on prior constraints in tractable probabilistic styles, accessible over the papers tab. Congratulations Giannis!
A recent collaboration Together with the NatWest Team on explainable equipment Understanding is reviewed in The Scotsman. Website link to posting here. A preprint on the outcome might be made obtainable shortly.
Jonathan’s paper considers a lifted approached to weighted model integration, which includes circuit building. Paulius’ paper develops a measure-theoretic viewpoint on weighted model counting and proposes a way to encode conditional weights on literals analogously to conditional probabilities, which leads to substantial efficiency improvements.
Paulius' Focus on algorithmic tactics for https://vaishakbelle.com/ randomly creating logic packages and probabilistic logic packages continues to be recognized towards the concepts and practise of constraint programming (CP2020).
The framework is relevant to a large class of formalisms, which include probabilistic relational versions. The paper also scientific tests the synthesis dilemma in that context. Preprint below.
Should you be attending AAAI this calendar year, you could have an interest in trying out our papers that contact on fairness, abstraction and generalized sum-item difficulties.
Our paper on synthesizing programs with loops from the existence of probabilistic sounds, recognized the journal of approximate reasoning, has also been accepted for the ICAPS journal monitor. Preprint to the complete paper in this article.