Products of Hidden Markov Models(PoHMMs) are an interesting class of generative models which have received little attention since their introduction. This maybe in part due to their more computationally expensive gradient-based learning algorithm,and the intractability of computing the log likelihood of sequences under the model. In this paper, we demonstrate how the partition function can be estimated reliably via Annealed Importance Sampling. We perform experiments using contrastive divergence learning on rainfall data and data captured from pairs of people dancing. Our results suggest that advances in learning and evaluation for undirected graphical models and recent increases in available computing power make PoHMMs worth considering for complex time-series modeling tasks.

We study a subclass of POMDPs, called Deterministic POMDPs, that is characterized by deterministic actions and observations. These models do not provide the same generality of POMDPs yet they capture a number of interesting and challenging problems, and permit more efficient algorithms. Indeed, some of the recent work in planning is built around such assumptions mainly by the quest of amenable models more expressive than the classical deterministic models. We provide results about the fundamental properties of Deterministic POMDPs, their relation with AND/OR search problems and algorithms, and their computational complexity.

The specification of aMarkov decision process (MDP) can be difficult. Reward function specification is especially problematic; in practice, it is often cognitively complex and time-consuming for users to precisely specify rewards. This work casts the problem of specifying rewards as one of preference elicitation and aims to minimize the degree of precision with which a reward function must be specified while still allowing optimal or near-optimal policies to be produced. We first discuss how robust policies can be computed for MDPs given only partial reward information using the minimax regret criterion. We then demonstrate how regret can be reduced by efficiently eliciting reward information using bound queries, using regret-reduction as a means for choosing suitable queries. Empirical results demonstrate that regret-based reward elicitation offers an effective way to produce near-optimal policies without resorting to the precise specification of the entire reward function.

Imaging spectrometers measure electromagnetic energy scattered in their instantaneous field view in hundreds or thousands of spectral channels with higher spectral resolution than multispectral cameras. Imaging spectrometers are therefore often referred to as hyperspectral cameras (HSCs). Higher spectral resolution enables material identification via spectroscopic analysis, which facilitates countless applications that require identifying materials in scenarios unsuitable for classical spectroscopic analysis. Due to low spatial resolution of HSCs, microscopic material mixing, and multiple scattering, spectra measured by HSCs are mixtures of spectra of materials in a scene. Thus, accurate estimation requires unmixing. Pixels are assumed to be mixtures of a few materials, called endmembers. Unmixing involves estimating all or some of: the number of endmembers, their spectral signatures, and their abundances at each pixel. Unmixing is a challenging, ill-posed inverse problem because of model inaccuracies, observation noise, environmental conditions, endmember variability, and data set size. Researchers have devised and investigated many models searching for robust, stable, tractable, and accurate unmixing algorithms. This paper presents an overview of unmixing methods from the time of Keshava and Mustard's unmixing tutorial [1] to the present. Mixing models are first discussed. Signal-subspace, geometrical, statistical, sparsity-based, and spatial-contextual unmixing algorithms are described. Mathematical problems and potential solutions are described. Algorithm characteristics are illustrated experimentally.

We present the discrete infinite logistic normal distribution (DILN), a Bayesian nonparametric prior for mixed membership models. DILN is a generalization of the hierarchical Dirichlet process (HDP) that models correlation structure between the weights of the atoms at the group level. We derive a representation of DILN as a normalized collection of gamma-distributed random variables, and study its statistical properties. We consider applications to topic modeling and derive a variational inference algorithm for approximate posterior inference. We study the empirical performance of the DILN topic model on four corpora, comparing performance with the HDP and the correlated topic model (CTM). To deal with large-scale data sets, we also develop an online inference algorithm for DILN and compare with online HDP and online LDA on the Nature magazine, which contains approximately 350,000 articles.

The common internal structure and algorithmic organization of object detection, detection-based tracking, and event recognition facilitates a general approach to integrating these three components. This supports multidirectional information flow between these components allowing object detection to influence tracking and event recognition and event recognition to influence tracking and object detection. The performance of the combination can exceed the performance of the components in isolation. This can be done with linear asymptotic complexity.

This document is my summary explanation of the algorithm in "A Spectral Algorithm for Learning Hidden Markov Models" (COLT 2009), though there may be some slight notational inconsistencies with the original paper. The exposition and the math here are quite different, so if you don't like this explanation, try the original paper! The idea is to maintain output predictions in a recursive inference algorithm, instead of the usual method of maintaining hidden state predictions, and to represent the HMM only in terms of the maps necessary to update output predictions given new data. This approach limits the inference computations the algorithm can perform (it can't answer any queries about the hidden states since it doesn't explicitly deal with them at all), but it also reduces the complexity of the model parameters that are learned and thus makes learning easier. The learning algorithm uses an SVD and matrix operations, so it avoids the local-optima problems of EM or any other algorithms based on maximizing data likelihood over the usual HMM parameterization.

The interactive partially observable Markov decision process (I-POMDP) is a recently developed framework which extends the POMDP to the multi-agent setting by including agent models in the state space. This paper argues for formulating the problem of an agent learning interactively from a human teacher as an I-POMDP, where the agent \emph{programming} to be learned is captured by random variables in the agent's state space, all \emph{signals} from the human teacher are treated as observed random variables, and the human teacher, modeled as a distinct agent, is explicitly represented in the agent's state space. The main benefits of this approach are: i. a principled action selection mechanism, ii. a principled belief update mechanism, iii. support for the most common teacher \emph{signals}, and iv. the anticipated production of complex beneficial interactions. The proposed formulation, its benefits, and several open questions are presented.

In a deterministic world, a planning agent can be certain of the consequences of its planned sequence of actions. Not so, however, in dynamic, stochastic domains where Markov decision processes are commonly used. Unfortunately these suffer from the `curse of dimensionality': if the state space is a Cartesian product of many small sets (`dimensions'), planning is exponential in the number of those dimensions. Our new technique exploits the intuitive strategy of selectively ignoring various dimensions in different parts of the state space. The resulting non-uniformity has strong implications, since the approximation is no longer Markovian, requiring the use of a modified planner. We also use a spatial and temporal proximity measure, which responds to continued planning as well as movement of the agent through the state space, to dynamically adapt the abstraction as planning progresses. We present qualitative and quantitative results across a range of experimental domains showing that an agent exploiting this novel approximation method successfully finds solutions to the planning problem using much less than the full state space. We assess and analyse the features of domains which our method can exploit.

Relational data representations have become an increasingly important topic due to the recent proliferation of network datasets (e.g., social, biological, information networks) and a corresponding increase in the application of statistical relational learning (SRL) algorithms to these domains. In this article, we examine a range of representation issues for graph-based relational data. Since the choice of relational data representation--for the nodes, links, and features--can dramatically affect the capabilities of SRL algorithms, we survey approaches and opportunities for relational representation transformation designed to improve the performance of these algorithms. This leads us to introduce an intuitive taxonomy for data representation transformations in relational domains that incorporates link transformation and node transformation as symmetric representation tasks. In particular, the transformation tasks for both nodes and links include (i) predicting their existence, (ii) predicting their label or type, (iii) estimating their weight or importance, and (iv) systematically constructing their relevant features. We motivate our taxonomy through detailed examples and use it to survey and compare competing approaches for each of these tasks. We also discuss general conditions for transforming links, nodes, and features. Finally, we highlight challenges that remain to be addressed.