This paper introduces a principled approach for the design of a scalable general reinforcement learning agent. This approach is based on a direct approximation of AIXI, a Bayesian optimality notion for general reinforcement learning agents. Previously, it has been unclear whether the theory of AIXI could motivate the design of practical algorithms. We answer this hitherto open question in the affirmative, by providing the first computationally feasible approximation to the AIXI agent. To develop our approximation, we introduce a Monte Carlo Tree Search algorithm along with an agent-specific extension of the Context Tree Weighting algorithm. Empirically, we present a set of encouraging results on a number of stochastic, unknown, and partially observable domains.

In an interlinked corpus of documents, the context in which a citation appears provides extra information about the cited document. However, associating terms in the context to the cited document remains an open problem. We propose a novel document generation approach that statistically incorporates the context in which a document links to another document. We quantitatively show that the proposed generation scheme explains the linking phenomenon better than previous approaches. The context information along with the actual content of the document provides significant improvements over the previous approaches for various real world evaluation tasks such as link prediction and log-likelihood estimation on unseen content. The proposed method is more scalable to large collection of documents compared to the previous approaches.

Many problems require repeated inference on probabilistic graphical models, with different values for evidence variables or other changes. Examples of such problems include utility maximization, MAP inference, online and interactive inference, parameter and structure learning, and dynamic inference. Since small changes to the evidence typically only affect a small region of the network, repeatedly performing inference from scratch can be massively redundant. In this paper, we propose expanding frontier belief propagation (EFBP), an efficient approximate algorithm for probabilistic inference with incremental changes to the evidence (or model). EFBP is an extension of loopy belief propagation (BP) where each run of inference reuses results from the previous ones, instead of starting from scratch with the new evidence; messages are only propagated in regions of the network affected by the changes. We provide theoretical guarantees bounding the difference in beliefs generated by EFBP and standard BP, and apply EFBP to the problem of expected utility maximization in influence diagrams. Experiments on viral marketing and combinatorial auction problems show that EFBP can converge much faster than BP without significantly affecting the quality of the solutions.

We propose Dirichlet Process mixtures of Generalized Linear Models (DP-GLM), a new method of nonparametric regression that accommodates continuous and categorical inputs, and responses that can be modeled by a generalized linear model. We prove conditions for the asymptotic unbiasedness of the DP-GLM regression mean function estimate. We also give examples for when those conditions hold, including models for compactly supported continuous distributions and a model with continuous covariates and categorical response. We empirically analyze the properties of the DP-GLM and why it provides better results than existing Dirichlet process mixture regression models. We evaluate DP-GLM on several data sets, comparing it to modern methods of nonparametric regression like CART, Bayesian trees and Gaussian processes. Compared to existing techniques, the DP-GLM provides a single model (and corresponding inference algorithms) that performs well in many regression settings.

There is a growing need for scalable semantic web repositories which support inference and provide efficient queries. There is also a growing interest in representing uncertain knowledge in semantic web datasets and ontologies. In this paper, I present a bit vector schema specifically designed for RDF (Resource Description Framework) datasets. I propose a system for materializing and storing inferred knowledge using this schema. I show experimental results that demonstrate that this solution simplifies inference queries and drastically improves results. I also propose and describe a solution for materializing and persisting uncertain information and probabilities. Thresholds and bit vectors are used to provide efficient query access to this uncertain knowledge. My goal is to provide a semantic web repository that supports knowledge inference, uncertainty reasoning, and Bayesian networks, without sacrificing performance or scalability.

bnlearn is an R package which includes several algorithms for learning the structure of Bayesian networks with either discrete or continuous variables. Both constraint-based and score-based algorithms are implemented, and can use the functionality provided by the snow package to improve their performance via parallel computing. Several network scores and conditional independence algorithms are available for both the learning algorithms and independent use. Advanced plotting options are provided by the Rgraphviz package.

Unifying logical and probabilistic reasoning is a longstanding goal of AI. While recent work in lifted belief propagation, handling whole sets of indistinguishable objects together, are promising steps towards achieving this goal that even scale to realistic domains, they are not tailored towards solving combinatorial problems such as determining the satisfiability of Boolean formulas. Recent results, however, show that certain other message passing algorithms, namely, survey propagation, are remarkably successful at solving such problems. In this paper, we propose the first lifted variants of survey propagation and its simpler version warning propagation. Our initial experimental results indicate that they are faster than using lifted belief propagation to determine the satisfiability of Boolean formulas.

Representations that combine first-order logic and probability have been the focus of much recent research. Lifted inference algorithms for them avoid grounding out the domain, bringing benefits analogous to those of resolution theorem proving in first-order logic. However, all lifted probabilistic inference algorithms to date treat potentials as black boxes, and do not take advantage of their logical structure. As a result, inference with them is needlessly inefficient compared to the logical case. We overcome this by proposing the first lifted probabilistic inference algorithm that exploits determinism and context specific independence. In particular, we show that AND/OR search can be lifted by introducing POWER nodes in addition to the standard AND and OR nodes. Experimental tests show the benefits of our approach.

Activity recognition plays an important role in many areas such as smart environments by offering unprecedented opportunities for assisted living, automation, security and energy efficiency. It’s also an essential component for planning and plan recognition in smart environments. One challenge of activity recognition is the need for collecting and annotating huge amounts of data for each new physical setting in order to be able to carry out the conventional activity discovery and recognition algorithms. This extensive initial phase of data collection and annotation results in a prolonged installation process and excessive time investment for each new space. In this paper we propose a new method of transferring learned knowledge of activities to a new physical space in order to leverage the learning process in the new environment. Our method called ”Home to Home Transfer Learning” (HHTL) is based on using a semi EM framework and modeling activities using structural, temporal and spatial features. This method allows us to avoid the tedious task of collecting and labeling huge amounts of data in the target space, and allows for a more accelerated and more scalable deployment cycle in the real world. It also allows us to exploit the insights learned in previous spaces. To validate our algorithms, we use the data collected in several smart apartments with different physical layouts.

Lifting can greatly reduce the cost of inference on first-order probabilistic models, but constructing the lifted network can itself be quite costly. In addition, the minimal lifted network is often very close in size to the fully propositionalized model; lifted inference yields little or no speedup in these situations. In this paper, we address both these problems. We propose a compact hypercube-based representation for the lifted network, which can greatly reduce the cost of lifted network construction. We also present two methods for approximate lifted network construction, which groups together similar but distinguishable objects and treats them as if they were identical. This can greatly reduce the size of the lifted network as well as the time required for lifted network construction, but potentially at some cost to accuracy. The coarseness of the approximation can be adjusted depending on the accuracy required, and we can bound the resulting error. Experiments on six domains show great efficiency gains with only minor loss in accuracy.