Paramveer S. Dhillon

Combining appearance and motion for human action classification in videos

An important cue to high level scene understanding is to analyze the objects in the scene and their behavior and interactions. In this paper, we study the problem of classification of activities in videos, as this is an integral component of any scene understanding system, and present a novel approach for recognizing human action categories in videos by combining information from appearance and motion of human body parts. Our approach is based on tracking human body parts by using mixture particle filters and then clustering the particles using local non - parametric clustering, hence associating a local set of particles to each cluster mode. The trajectory of these cluster modes provides the “motion” information and the “appearance” information is provided by the statistical information about the relative motion of these local set of particles over a number of frames. Later we use a “Bag of Words” model to build one histogram per video sequence from the set of these robust appearance and motion descriptors. These histograms provide us characteristic information which helps us to discriminate among various human actions which ultimately helps us in better understanding of the complete scene. We tested our approach on the standard KTH and Weizmann human action datasets and the results were comparable to the state of the art methods. Additionally our approach is able to distinguish between activities that involve the motion of complete body from those in which only certain body parts move. In other words, our method discriminates well between activities with “global body motion” like running, jogging etc. and “local motion” like waving, boxing etc.

Eigenanatomy improves detection power for longitudinal cortical change.

We contribute a novel and interpretable dimensionality reduction strategy, eigenanatomy, that is tuned for neuroimaging data. The method approximates the eigendecomposition of an image set with basis functions (the eigenanatomy vectors) that are sparse, unsigned and are anatomically clustered. We employ the eigenanatomy vectors as anatomical predictors to improve detection power in morphometry. Standard voxel-based morphometry (VBM) analyzes imaging data voxel-by-voxel--and follows this with cluster-based or voxel-wise multiple comparisons correction methods to determine significance. Eigenanatomy reverses the standard order of operations by first clustering the voxel data and then using standard linear regression in this reduced dimensionality space. As with traditional region-of-interest (ROI) analysis, this strategy can greatly improve detection power. Our results show that eigenanatomy provides a principled objective function that leads to localized, data-driven regions of interest. These regions improve our ability to quantify biologically plausible rates of cortical change in two distinct forms of neurodegeneration. We detail the algorithm and show experimental evidence of its efficacy.

Transfer Learning, Feature Selection and Word Sense Disambiguation

We propose a novel approach for improving Feature Selection for Word Sense Disambiguation by incorporating a feature relevance prior for each word indicating which features are more likely to be selected. We use transfer of knowledge from similar words to learn this prior over the features, which permits us to learn higher accuracy models, particularly for the rarer word senses. Results on the OntoNotes verb data show significant improvement over the baseline feature selection algorithm and results that are comparable to or better than other state-of-the-art methods.

Subject-specific functional parcellation via prior based eigenanatomy.

We present a new framework for prior-constrained sparse decomposition of matrices derived from the neuroimaging data and apply this method to functional network analysis of a clinically relevant population. Matrix decomposition methods are powerful dimensionality reduction tools that have found widespread use in neuroimaging. However, the unconstrained nature of these totally data-driven techniques makes it difficult to interpret the results in a domain where network-specific hypotheses may exist. We propose a novel approach, Prior Based Eigenanatomy (p-Eigen), which seeks to identify a data-driven matrix decomposition but at the same time constrains the individual components by spatial anatomical priors (probabilistic ROIs). We formulate our novel solution in terms of prior-constrained ℓ1 penalized (sparse) principal component analysis. p-Eigen starts with a common functional parcellation for all the subjects and refines it with subject-specific information. This enables modeling of the inter-subject variability in the functional parcel boundaries and allows us to construct subject-specific networks with reduced sensitivity to ROI placement. We show that while still maintaining correspondence across subjects, p-Eigen extracts biologically-relevant and patient-specific functional parcels that facilitate hypothesis-driven network analysis. We construct default mode network (DMN) connectivity graphs using p-Eigen refined ROIs and use them in a classification paradigm. Our results show that the functional connectivity graphs derived from p-Eigen significantly aid classification of mild cognitive impairment (MCI) as well as the prediction of scores in a Delayed Recall memory task when compared to graph metrics derived from 1) standard registration-based seed ROI definitions, 2) totally data-driven ROIs, 3) a model based on standard demographics plus hippocampal volume as covariates, and 4) Ward Clustering based data-driven ROIs. In summary, p-Eigen incarnates a new class of prior-constrained dimensionality reduction tools that may improve our understanding of the relationship between MCI and functional connectivity.

Efficient Feature Selection in the Presence of Multiple Feature Classes

We present an information theoretic approach to feature selection when the data possesses feature classes. Feature classes are pervasive in real data. For example, in gene expression data, the genes which serve as features may be divided into classes based on their membership in gene families or pathways. When doing word sense disambiguation or named entity extraction, features fall into classes including adjacent words, their parts of speech, and the topic and venue of the document the word is in. When predictive features occur predominantly in a small number of feature classes, our information theoretic approach significantly improves feature selection. Experiments on real and synthetic data demonstrate substantial improvement in predictive accuracy over the standard L0 penalty-based stepwise and stream wise feature selection methods as well as over Lasso and Elastic Nets, all of which are oblivious to the existence of feature classes.

Anatomically-Constrained PCA for Image Parcellation

Traditionally clinicians and medical researchers have been using either totally data driven approaches like PCA/CCA/ICA or ROI based analysis for exploratory analysis of brain images. However, PCA/CCA/ICA based approaches suffer from lack of interpretability of results and on the other hand ROI based approaches are too rigid and wrongly assume that the signal lies totally within a predefined region. In this paper, we propose a novel approach which stands in stark contrast with both these approaches as it borrows strength from both these paradigms and leads to statistically refined definitions of ROIs based on information from data. Our approach, called Anatomically Constrained PCA (AC-PCA) provides a principled way of incorporating prior information in the form of probabilistic or binary ROIs while still allowing the data to softly modify the original ROI definitions. Experimental results on cortical thickness images show the superiority of AC-PCA for MCI classification compared to ROI and unconstrained PCA (a totally data based approach).

Semi-supervised multi-task learning of structured prediction models for web information extraction

Extracting information from web pages is an important problem; it has several applications such as providing improved search results and construction of databases to serve user queries. In this paper we propose a novel structured prediction method to address two important aspects of the extraction problem: (1) labeled data is available only for a small number of sites and (2) a machine learned global model does not generalize adequately well across many websites. For this purpose, we propose a weight space based graph regularization method. This method has several advantages. First, it can use unlabeled data to address the limited labeled data problem and falls in the class of graph regularization based semi-supervised learning approaches. Second, to address the generalization inadequacy of a global model, this method builds a local model for each website. Viewing the problem of building a local model for each website as a task, we learn the models for a collection of sites jointly; thus our method can also be seen as a graph regularization based multi-task learning approach. Learning the models jointly with the proposed method is very useful in two ways: (1) learning a local model for a website can be effectively influenced by labeled and unlabeled data from other websites; and (2) even for a website with only unlabeled examples it is possible to learn a decent local model. We demonstrate the efficacy of our method on several real-life data; experimental results show that significant performance improvement can be obtained by combining semi-supervised and multi-task learning in a single framework.

Mapping of pain circuitry in early post-natal development using manganese-enhanced MRI in rats

Premature or ill full-term infants are subject to a number of noxious procedures as part of their necessary medical care. Although we know that human infants show neural changes in response to such procedures, we know little of the sensory or affective brain circuitry activated by pain. In rodent models, the focus has been on spinal cord and, more recently, midbrain and medulla. The present study assesses activation of brain circuits using manganese-enhanced magnetic resonance imaging (MEMRI). Uptake of manganese, a paramagnetic contrast agent that is transported across active synapses and along axons, was measured in response to a hindpaw injection of dilute formalin in 12-day-old rat pups, the age at which rats begin to show aversion learning and which is roughly the equivalent of full-term human infants. Formalin induced the oft-reported biphasic response at this age and induced a conditioned aversion to cues associated with its injection, thus demonstrating the aversiveness of the stimulation. Morphometric analyses, structural equation modeling and co-expression analysis showed that limbic and sensory paths were activated, the most prominent of which were the prefrontal and anterior cingulate cortices, nucleus accumbens, amygdala, hypothalamus, several brainstem structures, and the cerebellum. Therefore, both sensory and affective circuits, which are activated by pain in the adult, can also be activated by noxious stimulation in 12-day-old rat pups.

Social influence maximization under empirical influence models

Social influence maximization models aim to identify the smallest number of influential individuals (seed nodes) that can maximize the diffusion of information or behaviours through a social network. However, while empirical experimental evidence has shown that network assortativity and the joint distribution of influence and susceptibility are important mechanisms shaping social influence, most current influence maximization models do not incorporate these features. Here, we specify a class of empirically motivated influence models and study their implications for influence maximization in six synthetic and six real social networks of varying sizes and structures. We find that ignoring assortativity and the joint distribution of influence and susceptibility leads traditional models to underestimate influence propagation by 21.7% on average, for a fixed seed set size. The traditional models and the empirical types that we specify here also identify substantially different seed sets, with only 19.8% overlap between them. The optimal seeds chosen under empirical influence models are relatively less well-connected and less central nodes, and they have more cohesive, embedded ties with their contacts. Hence, empirically motivated influence models have the potential to identify more realistic sets of key influencers in a social network and inform intervention designs that disseminate information or change attitudes and behaviours.Aral and Dhillon specify a class of empirically motivated influence maximization models that incorporate more realistic features of real-world social networks and predict substantially greater influence propagation compared with traditional models.

Partial sparse canonical correlation analysis (PSCCA) for population studies in medical imaging

We propose a new multivariate method, partial sparse canonical correlation analysis (PSCCA), for computing the statistical comparisons needed by population studies in medical imaging. PSCCA is a multivariate generalization of linear regression that allows one to statistically parameterize imaging studies in terms of multiple views of the population (e.g., the full collection of measurements taken from an image set along with batteries of cognitive or genetic data) while controlling for nuisance variables. This paper develops the theory of PSCCA, provides an algorithm and illustrates PSCCA performance on both simulated and real datasets. We show, as a first application and evaluation of this new methodology, that PSCCA can improve detection power over mass univariate approaches while retaining the interpretability and biological plausibility of the estimated effects. We also discuss the strengths, limitations and future potential of this methodology.