I gave a talk for the workshop on how the synthesis of logic and equipment Understanding, Specifically places like statistical relational learning, can empower interpretability.
Weighted model counting usually assumes that weights are only specified on literals, typically necessitating the need to introduce auxillary variables. We look at a different solution dependant on psuedo-Boolean capabilities, bringing about a far more standard definition. Empirically, we also get SOTA benefits.
The Lab carries out research in synthetic intelligence, by unifying Understanding and logic, having a the latest emphasis on explainability
When you are attending NeurIPS this yr, you may be interested in testing our papers that touch on morality, causality, and interpretability. Preprints are available on the workshop web site.
We evaluate the question of how generalized ideas (programs with loops) might be deemed proper in unbounded and ongoing domains.
A consortia challenge on trustworthy systems and goverance was recognized late past calendar year. Information backlink below.
The work is motivated by the necessity to check and Appraise inference algorithms. A combinatorial argument for your correctness of your Concepts is also regarded. Preprint here.
Bjorn and I are promoting a 2 year postdoc on integrating causality, reasoning and information graphs for misinformation detection. See listed here.
Not long ago, he has consulted with important banking companies on explainable AI and its affect in monetary establishments.
, to allow techniques to know faster and much more correct versions of the globe. We are interested in establishing computational frameworks that can easily reveal their selections, modular, re-usable
Extended abstracts of https://vaishakbelle.com/ our NeurIPS paper (on PAC-Discovering in first-get logic) as well as journal paper on abstracting probabilistic styles was approved to KR's not too long ago printed research observe.
A journal paper on abstracting probabilistic products has become recognized. The paper reports the semantic constraints which allows one particular to summary a complex, minimal-amount model with a simpler, substantial-amount 1.
The 1st introduces a first-purchase language for reasoning about probabilities in dynamical domains, and the next considers the automatic resolving of probability challenges laid out in all-natural language.
Meeting backlink Our work on symbolically interpreting variational autoencoders, as well as a new learnability for SMT (satisfiability modulo idea) formulas got recognized at ECAI.