I gave a chat on the workshop on how the synthesis of logic and machine learning, Primarily places such as statistical relational Mastering, can enable interpretability.
Weighted product counting normally assumes that weights are only specified on literals, normally necessitating the necessity to introduce auxillary variables. We look at a different tactic determined by psuedo-Boolean features, leading to a far more standard definition. Empirically, we also get SOTA success.
I gave a chat entitled "Perspectives on Explainable AI," at an interdisciplinary workshop focusing on setting up belief in AI.
I attended the SML workshop while in the Black Forest, and talked about the connections amongst explainable AI and statistical relational Studying.
Gave a talk this Monday in Edinburgh within the ideas & apply of machine learning, masking motivations & insights from our survey paper. Crucial concerns raised included, how you can: extract intelligible explanations + modify the model to suit shifting wants.
I’ll be supplying a talk with the conference on reasonable and liable AI inside the cyber Actual physical methods session. Owing to Ram & Christian for your invitation. Backlink to event.
Serious about schooling neural networks with sensible constraints? We have a brand new paper that aims in the direction of total fulfillment of Boolean and linear arithmetic constraints on coaching at AAAI-2022. Congrats to Nick and Rafael!
A journal paper has long been recognized on prior constraints in tractable probabilistic styles, offered within the papers tab. Congratulations Giannis!
Just lately, he has consulted with major banking institutions on explainable AI and its impression in monetary institutions.
Along with colleagues from Edinburgh and Herriot Watt, Now we have set out https://vaishakbelle.com/ the demand a brand new analysis agenda.
Extended abstracts of our NeurIPS paper (on PAC-Discovering in initially-purchase logic) and also the journal paper on abstracting probabilistic models was approved to KR's just lately printed investigate observe.
The framework is applicable to a large class of formalisms, including probabilistic relational styles. The paper also research the synthesis trouble in that context. Preprint right here.
Our work on synthesizing ideas with loops inside the presence of noise will appear within the Worldwide journal of approximate reasoning.
Meeting website link Our work on symbolically interpreting variational autoencoders, in addition to a new learnability for SMT (satisfiability modulo concept) formulas bought approved at ECAI.