I gave a talk, entitled "Explainability being a service", at the above party that mentioned anticipations concerning explainable AI And just how could be enabled in programs.
Weighted product counting usually assumes that weights are only specified on literals, typically necessitating the necessity to introduce auxillary variables. We look at a new method depending on psuedo-Boolean features, resulting in a more basic definition. Empirically, we also get SOTA benefits.
Will likely be speaking on the AIUK occasion on principles and observe of interpretability in device Finding out.
I attended the SML workshop while in the Black Forest, and discussed the connections concerning explainable AI and statistical relational Finding out.
An post in the preparing and inference workshop at AAAI-eighteen compares two distinctive ways for probabilistic organizing by means of probabilistic programming.
I gave a chat on our modern NeurIPS paper in Glasgow although also covering other strategies with the intersection of logic, Understanding and tractability. Thanks to Oana with the invitation.
Now we have a fresh paper approved on Understanding best linear programming targets. We consider an “implicit“ hypothesis building technique that yields awesome theoretical bounds. Congrats to Gini and Alex on getting this paper acknowledged. Preprint listed here.
A journal paper has become acknowledged on prior constraints in tractable probabilistic designs, accessible on the papers tab. Congratulations Giannis!
A modern collaboration With all the NatWest Team on explainable device learning is reviewed inside the Scotsman. Connection to report here. A preprint on the final results will likely be built offered https://vaishakbelle.com/ Soon.
Jonathan’s paper considers a lifted approached to weighted product integration, such as circuit development. Paulius’ paper develops a evaluate-theoretic perspective on weighted model counting and proposes a method to encode conditional weights on literals analogously to conditional probabilities, which ends up in significant functionality advancements.
Paulius' work on algorithmic strategies for randomly generating logic programs and probabilistic logic applications continues to be accepted to the concepts and practise of constraint programming (CP2020).
Our MLJ (2017) short article on organizing with hybrid MDPs was recognized for presentation on the journal keep track of.
Our work on synthesizing programs with loops while in the existence of sound will look while in the Global journal of approximate reasoning.
Our perform (with Giannis) surveying and distilling ways to explainability in machine learning continues to be accepted. Preprint in this article, but the ultimate Edition will likely be online and open up entry quickly.