Programme

• Sunday, Sep. 12, 18:00-21:00: Welcoming reception at the Scandic Marina Hotel.
• Monday, Sep. 13, 08:45-20:00: Technical sessions
• Monday, Sep. 13, 20:00-22:00: Reception offered by the Rector of University of Helsinki (joint event with the JELIA conference)
• Tuesday, Sep. 14, 09:00-19:30: Technical sessions.
• Tuesday, Sep. 14, 20:00-23:00: Banquet Dinner at Restaurant Katsomo (directions)
• Wednesday, Sep. 15, 09:00-19:00: Technical sessions.

Monday, September 13

08:45-09:00 Opening of the Workshop

09:00-10:20 Session 1: Inference I
Chair: Jirka Vomlel

• Babak Ahmadi, Kristian Kersting, Fabian Hadiji: Lifted belief propagation: Pairwise marginals and beyond
• Cory Butz, Wen Yan: The semantics of intermediate CPTs in variable elimination
• Morgan Chopin, Pierre-Henri Wuillemin: Optimizing the triangulation of dynamic Bayesian networks
• Sander Evers, Peter Lucas: Variable elimination by factor indexing

10:20-10:50 Coffee

10:50-12:10 Session 2: Inference II
Chair: Silja Renooij

• Heejin Lim, Changhe Yuan, Eric Hansen: Scaling up MAP search in Bayesian networks using external memory
• Wannes Meert, Jan Struyf, Hendrik Blockeel: Contextual variable elimination with overlapping contexts
• Thorsten Ottosen, Jirka Vomlel: All roads lead To Rome—New search methods for optimal triangulations
• Thorsten Ottosen, Jirka Vomlel: Honour thy neighbour—Clique maintenance in dynamic graphs

12:10-14:00 Buffet Lunch

14:00-15:20 Session 3: Learning
Chair: Patrik Hoyer

• Sourour Ammar, Philippe Leray, François Schnitzler, Louis Wehenkel: Sub-quadratic Markov tree mixture learning based on randomizations of the Chow-Liu algorithm
• Hanen Borchani, Concha Bielza, Pedro Larrañaga: Learning CB-decomposable multi-dimensional Bayesian network classifiers
• Antonio Fernández, Helge Langseth, Thomas Nielsen, Antonio Salmerón: Parameter learning in MTE networks using incomplete data
• Alberto Roverato, Robert Castelo: Learning undirected graphical models from multiple datasets with the generalized non-rejection rate

15:20-15:45 Coffee

15:45-17:45 Poster Session I

18:45-19:45 Invited Talk (at the Second festive hall ("Pieni juhlasali") of the University Main Building, joint event with JELIA)

• Adnan Darwiche: Relax, compensate and then recover: A theory of anytime, approximate inference

20:00-22:00 Reception offered by the Rector of University of Helsinki ("Lehtisali" in the Main Building)

Tuesday, September 14

09:00-10:20 Session 4: Models and Representation I
Chair: Antonio Salmeron

• Wray Buntine, Lan Du, Petteri Nurmi: Bayesian networks on Dirichlet distributed vectors
• Arthur Choi, Adnan Darwiche: On a discrete Dirichlet model
• Daniele Codetta-Raiteri, Luigi Portinale: Generalized continuous time Bayesian networks and their GSPN semantics
• Peter Lucas, Arjen Hommersom: Modelling the interactions between discrete and continuous causal factors in Bayesian networks

10:20-10:50 Coffee

10:50-12:10 Session 5: Models and Representation II
Chair: Peter Lucas

• Jose Peña: Reading dependencies from polytree-like Bayesian networks revisited
• Milan Studený, Raymond Hemmecke, Silvia Lindner: Characteristic imset: a simple algebraic representative of a Bayesian network structure
• Lionel Torti, Pierre-Henri Wuillemin, Christophe Gonzales: Reinforcing the object-oriented aspect of probabilistic relational models
• Marco Cattaneo: Likelihood-based inference for probabilistic graphical models: Some preliminary results

12:10-14:00 Buffet Lunch

14:00-15:00 Invited Talk

• Chris Howe: What do genes, medieval manuscripts, languages, musical instruments, factories and Persian carpets have in common?

15:00-15:40 Session 6: Models and Representation III
Chair: Milan Studeny

• Andrès Cano, Manuel Gómez-Olmedo, Serafin Moral, Antonio Salmerón, Cora Pèrez Ariza: Learning recursive probability trees from probabilistic potentials
• Yang Xiang: Acquisition and computation issues with NIN-AND tree models

15:40-16:10 Coffee

16:10-16:50 Session 7: Influence Diagrams
Chair: Concha Bielza

• Barry Cobb: Continuous decision variables with multiple continuous parents
• Finn Jensen, Elena Gatti: Information enhancement for approximate representation of optimal strategies from influence diagrams

16:50-17:30 Business Meeting

17:45-19:30 Poster Session II

20:00-23:00 Banquet Dinner at Restaurant Katsomo (directions)

Wednesday, September 15

09:00-10:20 Session 8: Sensitivity and Uncertainty
Chair: Thomas Nielsen

• Adnan Darwiche, Arthur Choi: Same-decision probability: A confidence measure for threshold-based decisions under noisy sensors
• Miguel Angel Gómez-Villegas, Paloma Main, Hilario Navarro, Rosario Susi: Dealing with uncertainty in Gaussian Bayesian networks from a regression perspective
• Silja Renooij: Bayesian network sensitivity to arc-removal
• Silja Renooij: Efficient sensitivity analysis in hidden Markov models

10:20-10:50 Coffee

10:50-12:10 Session 9: Causality
Chair: John Agosta

• Tom Claassen, Tom Heskes: Learning causal network structure from multiple (in)dependence models
• Doris Entner, Patrik Hoyer: On causal discovery from time series data using FCI
• Antti Hyttinen, Frederick Eberhardt, Patrik Hoyer: Causal discovery for linear cyclic models with latent variables
• Jan Lemeire, Stijn Meganck, Francesco Cartella: Robust independence-based causal structure learning in absence of adjacency faithfulness

12:10-14:00 Buffet Lunch

14:00-15:20 Session 10: Applications
Chair: Pedro Larranaga

• John Agosta, Omar Zia Khan, Pascal Poupart: Evaluation results for a query-based diagnostics application
• Barry Cobb: An influence diagram model for detecting credit card fraud
• Hongyu Su, Markus Heinonen, Juho Rousu: Multilabel classification of drug-like molecules via max-margin conditional random fields
• Demet Özgür-Ünlüakin, Taner Bilgiç: An aggregation and disaggregation procedure for the maintenance of a dynamic system under partial observations

15:20-15:45 Coffee

15:45-16:45 Invited Talk

• Thore Graepel: Graphical models in Microsoft's online services: TrueSkill, AdPredictor, and Matchbox

16:45-19:00 Cocktails (Offered by the Helsinki Graduate School in Computer Science and Engineering (HeCSE))

17:00-19:00 Poster Session III (jointly with HeCSE poster session)

Adnan Darwiche (UCLA): Relax, Compensate and then Recover: A Theory of Anytime, Approximate Inference → slides

• Abstract: I will present in this talk a theory of anytime, approximate inference, which explains some of the most influential algorithms in probabilistic reasoning, yet is based on one of the older ideas in symbolic reasoning: Relaxations. According to this theory, the fundamental notion of approximation is that of "relaxing" logical constraints (equalities in particular) for the purpose of decomposing a problem into smaller pieces that can be solved independently. A second fundamental notion is that of "compensation", which calls for imposing weaker and, typically, probabilistic notions of equality to compensate for the relaxed equalities. The third fundamental notion of the presented theory is that of "recovery," where some of the relaxed equalities are incrementally recovered, based on an assessment of their impact on improving the quality of approximations. I will discuss how this theory subsumes one of the most influential algorithms in probabilistic inference: loopy belief propagation and some of its generalizations. I will also discuss the relationship between this theory and current developments in symbolic reasoning—in particular, how advances in knowledge compilation can impact the state of the art in approximate (and exact) probabilistic inference.
• Time: 18:45-19:45, Monday, September 13
• Location: The second festive hall ("Pieni juhlasali") of the main building of University of Helsinki (a joint event with the JELIA conference).

Chris Howe (University of Cambridge): What do genes, medieval manuscripts, languages, musical instruments, factories and Persian carpets have in common?

• Abstract: Biologists have long sought to use the information in DNA and protein sequences to infer evolutionary relationships among groups of organisms, and usually depict them either as two-dimensional bifurcating trees or as networks. The concept of showing the relationships among organisms as trees was famously demonstrated by Darwin, who included an evolutionary tree as the only figure in "On the Origin of Species...". Inferring "phylogenetic" trees is made possible by the accumulation of mutations in DNA as it is copied. However, many other things follow a similar process of copying with the incorporation of changes, and the methods used for inferring phylogenetic trees might, in principle, be applied elsewhere. I shall describe studies applying phylogenetic methods to topics as diverse as manuscript versions of texts, languages, musical instruments, commercial organisations and Persian carpets.
• Time: 14:00-15:00, Tuesday, September 14.
• Location: Marina Congress Center

Thore Graepel (Microsoft Research): Graphical Models in Microsoft's Online Services: TrueSkill, AdPredictor, and Matchbox

• Abstract: Probabilistic Graphical Models play a crucial role in Microsoft's online services. In this talk, I will describe three powerful applications of graphical model inference in practice.
  1. TrueSkill is Xbox Live's Ranking and Matchmaking system and ensures that gamers online have balanced and exciting matches with equally skilled opponents.
  2. AdPredictor is the system that estimates click-through rates (CTR) for ad selection and pricing within Microsoft's search engine Bing.
  3. Matchbox is a large scale Bayesian recommender system that combines aspects of collaborative filtering and content-based recommendation. It is currently being used for tweet recommendation within projectemporia.com.
  All three systems have in common that they are based on factor graph models and approximate Bayesian inference. They operate at a very large scale involving millions of gamers, billions of ad impressions, and millions of tweets, respectively. I will discuss the underlying graphical models and inference algorithms as well as application-specific insights and findings. Time permitting, I will show the three systems in action. This is based on joint work with Ralf Herbrich, David Stern, Thomas Borchert, Tom Minka, and Joaquin Quiñonero Candela.
• Time: 15:45-16:45, Wednesday, September 15.
• Location: Marina Congress Center