Showing papers by "Padhraic Smyth published in 2014"

Modeling human location data with mixtures of kernel densities

Abstract: Location-based data is increasingly prevalent with the rapid increase and adoption of mobile devices. In this paper we address the problem of learning spatial density models, focusing specifically on individual-level data. Modeling and predicting a spatial distribution for an individual is a challenging problem given both (a) the typical sparsity of data at the individual level and (b) the heterogeneity of spatial mobility patterns across individuals. We investigate the application of kernel density estimation (KDE) to this problem using a mixture model approach that can interpolate between an individual's data and broader patterns in the population as a whole. The mixture-KDE approach is evaluated on two large geolocation/check-in data sets, from Twitter and Gowalla, with comparisons to non-KDE baselines, using both log-likelihood and detection of simulated identity theft as evaluation metrics. Our experimental results indicate that the mixture-KDE method provides a useful and accurate methodology for capturing and predicting individual-level spatial patterns in the presence of noisy and sparse data.

Scaling up the evaluation of psychotherapy: evaluating motivational interviewing fidelity via statistical text classification

Abstract: Background: Behavioral interventions such as psychotherapy are leading, evidence-based practices for a variety of problems (e.g., substance abuse), but the evaluation of provider fidelity to behavioral interventions is limited by the need for human judgment. The current study evaluated the accuracy of statistical text classification in replicating human-based judgments of provider fidelity in one specific psychotherapy—motivational interviewing (MI). Method: Participants (n = 148) came from five previously conducted randomized trials and were either primary care patients at a safety-net hospital or university students. To be eligible for the original studies, participants met criteria for either problematic drug or alcohol use. All participants received a type of brief motivational interview, an evidence-based intervention for alcohol and substance use disorders. The Motivational Interviewing Skills Code is a standard measure of MI provider fidelity based on human ratings that was used to evaluate all therapy sessions. A text classification approach called a labeled topic model was used to learn associations between human-based fidelity ratings and MI session transcripts. It was then used to generate codes for new sessions. The primary comparison was the accuracy of model-based codes with human-based codes. Results: Receiver operating characteristic (ROC) analyses of model-based codes showed reasonably strong sensitivity and specificity with those from human raters (range of area under ROC curve (AUC) scores: 0.62 – 0.81; average AUC: 0.72). Agreement with human raters was evaluated based on talk turns as well as code tallies for an entire session. Generated codes had higher reliability with human codes for session tallies and also varied strongly by individual code. Conclusion: To scale up the evaluation of behavioral interventions, technological solutions will be required. The current study demonstrated preliminary, encouraging findings regarding the utility of statistical text classification in bridging this methodological gap.

The co-factor of LIM domains (CLIM/LDB/NLI) maintains basal mammary epithelial stem cells and promotes breast tumorigenesis.

Abstract: Mammary gland branching morphogenesis and ductal homeostasis relies on mammary stem cell function for the maintenance of basal and luminal cell compartments. The mechanisms of transcriptional regulation of the basal cell compartment are currently unknown. We explored these mechanisms in the basal cell compartment and identified the Co-factor of LIM domains (CLIM/LDB/NLI) as a transcriptional regulator that maintains these cells. Clims act within the basal cell compartment to promote branching morphogenesis by maintaining the number and proliferative potential of basal mammary epithelial stem cells. Clim2, in a complex with LMO4, supports mammary stem cells by directly targeting the Fgfr2 promoter in basal cells to increase its expression. Strikingly, Clims also coordinate basal-specific transcriptional programs to preserve luminal cell identity. These basal-derived cues inhibit epidermis-like differentiation of the luminal cell compartment and enhance the expression of luminal cell-specific oncogenes ErbB2 and ErbB3. Consistently, basal-expressed Clims promote the initiation and progression of breast cancer in the MMTV-PyMT tumor model, and the Clim-regulated branching morphogenesis gene network is a prognostic indicator of poor breast cancer outcome in humans.

Beyond MAP Estimation With the Track-Oriented Multiple Hypothesis Tracker

Abstract: The track-oriented multiple hypothesis tracker (TOMHT) is a popular algorithm for tracking multiple targets in a cluttered environment. In tracking parlance it is known as a multi-scan, maximum a posteriori (MAP) estimator-multi-scan because it enumerates possible data associations jointly over several scans, and MAP because it seeks the most likely data association conditioned on the observations. This paper extends the TOMHT, building on its internal representation to support probabilistic queries other than MAP estimation. Specifically, by summing over the TOMHT's pruned space of data association hypotheses one can compute marginal probabilities of individual tracks. Since this summation is generally intractable, any practical implementation must replace it with an approximation. We introduce a factor graph representation of the TOMHT's data association posterior and use variational message-passing to approximate track marginals. In an empirical evaluation, we show that marginal estimates computed through message-passing compare favorably to those computed through explicit summation over the k-best hypotheses, especially as the number of possible hypotheses increases. We also show that track marginals enable parameter estimation in the TOMHT via a natural extension of the expectation maximization algorithm used in single-target tracking. In our experiments, online EM updates using approximate marginals significantly increased tracker robustness to poor initial parameter specification.

Approximate Slice Sampling for Bayesian Posterior Inference

Abstract: In this paper, we advance the theory of large scale Bayesian posterior inference by introducing a new approximate slice sampler that uses only small mini-batches of data in every iteration. While this introduces a bias in the stationary distribution, the computational savings allow us to draw more samples in a given amount of time and reduce sampling variance. We empirically verify on three different models that the approximate slice sampling algorithm can significantly outperform a traditional slice sampler if we are allowed only a fixed amount of computing time for our simulations.

Hot Swapping for Online Adaptation of Optimization Hyperparameters

Abstract: We describe a general framework for online adaptation of optimization hyperparameters by `hot swapping' their values during learning. We investigate this approach in the context of adaptive learning rate selection using an explore-exploit strategy from the multi-armed bandit literature. Experiments on a benchmark neural network show that the hot swapping approach leads to consistently better solutions compared to well-known alternatives such as AdaDelta and stochastic gradient with exhaustive hyperparameter search.

Annealing paths for the evaluation of topic models

Abstract: Statistical topic models such as latent Dirich-let allocation have become enormously popular in the past decade, with dozens of learning algorithms and extensions being proposed each year. As these models and algorithms continue to be developed, it becomes increasingly important to evaluate them relative to previous techniques. However, evaluating the predictive performance of a topic model is a computationally difficult task. Annealed importance sampling (AIS), a Monte Carlo technique which operates by annealing between two distributions, has previously been successfully used for topic model evaluation (Wallach et al., 2009b). This technique estimates the likelihood of a held-out document by simulating an annealing process from the prior to the posterior for the latent topic assignments, and using this simulation as an importance sampling proposal distribution. In this paper we introduce new AIS annealing paths which instead anneal from one topic model to another, thereby estimating the relative performance of the models. This strategy can exhibit much lower empirical variance than previous approaches, facilitating reliable per-document comparisons of topic models. We then show how to use these paths to evaluate the predictive performance of topic model learning algorithms by efficiently estimating the likelihood at each iteration of the training procedure. The proposed method achieves better held-out likelihood estimates for this task than previous algorithms with, in some cases, an order of magnitude less computation.

Approximate Slice Sampling for Bayesian Posterior Inference

Abstract: In this paper, we advance the theory of large scale Bayesian posterior inference by introducing a new approximate slice sampler that uses only small mini-batches of data in every iteration. While this introduces a bias in the stationary distribution, the computational savings allow us to draw more samples in a given amount of time and reduce sampling variance. We empirically verify on three dierent models that the approximate slice sampling algorithm can signicantly outper

Hot Swapping for Online Adaptation of Optimization Hyperparameters

Abstract: We describe a general framework for online adaptation of optimization hyperparameters by `hot swapping' their values during learning. We investigate this approach in the context of adaptive learning rate selection using an explore-exploit strategy from the multi-armed bandit literature. Experiments on a benchmark neural network show that the hot swapping approach leads to consistently better solutions compared to well-known alternatives such as AdaDelta and stochastic gradient with exhaustive hyperparameter search.