Learn More
Graphical models for structured domains are powerful tools, but the computational complexities of combinatorial prediction spaces can force restrictions on models, or require approximate inference in order to be tractable. Instead of working in a combinatorial space, we use hinge-loss Markov random fields (HL-MRFs), an expressive class of graphical models(More)
Structured predictors enable joint inference over multiple interdependent output variables. These models are often trained on a small number of examples with large internal structure. Existing distribution-free generalization bounds do not guarantee generalization in this setting, though this contradicts a large body of empirical evidence from computer(More)
Updating inference in response to new evidence is a fundamental challenge in artificial intelligence. Many real problems require large probabilistic graphical models, containing millions of interdependent variables. For such large models, jointly updating the most likely (i.e., MAP) configuration of the variables each time new evidence is encountered can be(More)
We propose a modular framework for multirelational learning via tensor decomposition. In our learning setting, the training data contains multiple types of relationships among a set of objects, which we represent by a sparse three-mode tensor. The goal is to predict the values of the missing entries. To do so, we model each relationship as a function of a(More)
In network classification problems such as those found in intelligence gathering, public health, and viral marketing, one is often only interested in inferring the labels of a subset of the nodes. We refer to this subset as the query set, and define the problem as query-driven collective classification. We study this problem in a practical active learning(More)
We propose hinge-loss Markov random fields (HL-MRFs), a powerful class of continuous-valued graphical models, for high-level computer vision tasks. HL-MRFs are characterized by log-concave density functions, and are able to perform efficient, exact inference. Their templated hinge-loss potential functions naturally encode soft-valued logical rules. Using(More)
Recent results have shown that the generalization error of structured predictors decreases with both the number of examples and the size of each example, provided the data distribution has weak dependence and the predictor exhibits a smoothness property called collective stability. These results use an especially strong definition of collective stability(More)
Structured prediction models have been found to learn effectively from a few large examples— sometimes even just one. Despite empirical evidence, canonical learning theory cannot guarantee generalization in this setting because the error bounds decrease as a function of the number of examples. We therefore propose new PAC-Bayesian generalization bounds for(More)
Collective inference has been shown empirically to successfully exploit the natural dependencies in relational and network data [5, 11, 12, 13]. Though many collective techniques are capable of induction, and have been shown to be asymptotically consistent [16], little to no theory exists concerning the generalization of such methods. Collective inference(More)