A good way to end a blog drought is with a post for Ada Lovelace Day.
I was going to write about Daphne Koller (who is such an easy pick! the MacArthur fellowship alone…), but there was a post last year about her. Now, I’m not adverse to repeating, but breath is good!
There are, to my eye, comparatively a lot of women in probabilistic reasoning and machine learning (see a prior Lovelace day post (which strangely omits Lise Getoor, there’s a mailing list, a workshop series!, lots of great stuff, indeed, it seems to be a great model for how a subdiscipline can be supportive).
Someone I’ve been reading recently (as part of my research in probabilistic logic) in addition to Koller is Kathryn B. Laskey (see her DBLP page, her publication page, and her courses page).
She is a pioneer in Knowledge Base Model Construction (KBMC) formalisms and techniques for modeling probabilistic knowledge. KBMC languages typically are generalizations of Bayesian Networks (BNs) which attempt to reduce the complexity of specifying large BNs by, roughly, specifying how the larger BN is to be assembled out of a collection of templates or fragments. They are a way of lifting the highly propositional nature of BNs by identifying and abstracting out repeated substructure.
Now, I’m still learning this literature, so I’m not committing to this scholarly assessment yet, but I identify Mahoney’s and Laskey’s “Network Engineering for Complex Belief Networks” as coming quite early in the history of KBMCs. Indeed, Koller and Pfeffer cite it in their “Object-Oriented Bayesian Networks“:
Independently of our work, Laskey and Mahoney [Laskey and Mahoney, 1997] have developed a framework for representing probabilistic knowledge that shares some features with OOBNs. In their framework, based on network fragments, complex fragments are built out of simpler ones, in much the same way as complex OONFs are built out of simpler ones. Their framework currently supports the representation of certain features such as hypotheses that we are in the process of incorporating into OOBNs. However, their approach to building complex models is procedural in nature, whereas ours uses a declarative object-oriented representation language. As a result, our approach allows the organizational structure of a model, in particular the encapsulation of objects and the reuse of OONFs within a model, to be expressed explicitly and utilized by the inference algorithm.
Laskey’s more recent work on Multi-Entity Bayesian Networks (MBENs) certain more than addresses this “procedural” point, being based on an extension of First Order Logic (as the “template” language). Even more recently, she has a paper axiomitizing a theory of probability in first order logic in such a way that we end up with an interesting probabilisitic logic. (I confess to not having fully assimilated it yet, esp. as related to standard Halpren and Bacchus style first order probabilistic logics…but that’s part of the fun!)
Besides her research (which I’ve only briefly touched upon), Laskey also was co-chair of the W3C’s Uncertainty Reasoning for the Semantic Web Incubator Group (thus, engaged in significant outreach). I’m also finding her slides for her course “Computational Models for Probabilistic Reasoning” quite interesting and helpful (and dense!). Looking at her lecture on Knowledge Engineering makes me itchy about several of my lectures and is inspiring me to rework them.
In research, service, and teaching she seems to be a model and inspiring academic.
Mitigated Frenzy 