Dictionary learning and sparse coding on Grassmann manifolds : an extrinsic solution

Recent advances in computer vision and machine learning suggest that a wide range of problems can be addressed more appropriately by considering non-Euclidean geometry. In this paper we explore sparse dictionary learning over the space of linear subspaces, which form Riemannian structures known as Grassmann manifolds. To this end, we propose to embed Grassmann manifolds into the space of symmetric matrices by an isometric mapping, which enables us to devise a closed-form solution for updating a Grassmann dictionary, atom by atom. Furthermore, to handle non-linearity in data, we propose a kernelised version of the dictionary learning algorithm. Experiments on several classification tasks (face recognition, action recognition, dynamic texture classification) show that the proposed approach achieves considerable improvements in discrimination accuracy, in comparison to state-of-the-art methods such as kernelised Affine Hull Method and graph-embedding Grassmann discriminant analysis.

Multi-shot person re-identification via relational stein divergence

Person re-identification is particularly challenging due to significant appearance changes across separate camera views. In order to re-identify people, a representative human signature should effectively handle differences in illumination, pose and camera parameters. While general appearance-based methods are modelled in Euclidean spaces, it has been argued that some applications in image and video analysis are better modelled via non-Euclidean manifold geometry. To this end, recent approaches represent images as covariance matrices, and interpret such matrices as points on Riemannian manifolds. As direct classification on such manifolds can be difficult, in this paper we propose to represent each manifold point as a vector of similarities to class representers, via a recently introduced form of Bregman matrix divergence known as the Stein divergence. This is followed by using a discriminative mapping of similarity vectors for final classification. The use of similarity vectors is in contrast to the traditional approach of embedding manifolds into tangent spaces, which can suffer from representing the manifold structure inaccurately. Comparative evaluations on benchmark ETHZ and iLIDS datasets for the person re-identification task show that the proposed approach obtains better performance than recent techniques such as Histogram Plus Epitome, Partial Least Squares, and Symmetry-Driven Accumulation of Local Features.

Comparison of block and event-related experimental designs in diffusion-weighted functional MRI

PURPOSE To compare diffusion-weighted functional magnetic resonance imaging (DfMRI), a novel alternative to the blood oxygenation level-dependent (BOLD) contrast, in a functional MRI experiment. MATERIALS AND METHODS Nine participants viewed contrast reversing (7.5 Hz) black-and-white checkerboard stimuli using block and event-related paradigms. DfMRI (b = 1800 mm/s2 ) and BOLD sequences were acquired. Four parameters describing the observed signal were assessed: percent signal change, spatial extent of the activation, the Euclidean distance between peak voxel locations, and the time-to-peak of the best fitting impulse response for different paradigms and sequences. RESULTS The BOLD conditions showed a higher percent signal change relative to DfMRI; however, event-related DfMRI showed the strongest group activation (t = 21.23, P < 0.0005). Activation was more diffuse and spatially closer to the BOLD response for DfMRI when the block design was used. DfMRIevent showed the shortest TTP (4.4 +/- 0.88 sec). CONCLUSION The hemodynamic contribution to DfMRI may increase with the use of block designs.

Absent interiors

Anuradha Mathur and Dilip da Cunha theorise in their work on cities and flooding that it is not the floodwaters that threaten lives and homes, the real cause of danger in natural disaster is the fixity of modern civilisation. Their work traces the fluidity of the boundaries between 'dry' and 'wet' land challenging the deficiencies of traditional cartography in representing the extents of bodies of water. Mathur and da Cunha propose a process of unthinking to address the redevelopment of communities in the aftermath of natural disaster. By documenting the path of floodwaters in non-Euclidean space they propose a more appropriate response to flooding. This research focuses on the documentation of flooding in the interior of dwellings, which is an extreme condition of damage by external conditions in an environment designed to protect from these very elements. Because the floodwaters don't discriminate between the interior and the exterior, they move between structures with disregard for the systems of space we have in place. With the rapid clean up that follows flood damage, little material evidence is left for post mortem examination. This is especially the case for the flood damaged interior, piles of materials susceptible to the elements, furniture, joinery and personal objects line curbsides awaiting disposal. There is a missed opportunity in examining the interior in the after math of flood, in the way that Mathur and Dilip investigate floods and the design of cities, the flooded interior proffers an undersigned interior to study. In the absence of intact flood damaged interior, this research relies on two artists' documentation of the flooded interior. The first case study is the mimetic scenographic interiors of a flood-damaged office exhibited in the Bangkok art gallery by the group _Proxy in 2011. The second case study is Robert Polidori's photographic exhibition in New Orleans, described by Julianna Preston as, 'a series of interiors undetected by satellite imaging or storm radar. More telling, more dramatic, more unnerving, more alarming, they force a disturbance of what is familiar'.

Robust clustering of multi-type relational data via a heterogeneous manifold ensemble

High-Order Co-Clustering (HOCC) methods have attracted high attention in recent years because of their ability to cluster multiple types of objects simultaneously using all available information. During the clustering process, HOCC methods exploit object co-occurrence information, i.e., inter-type relationships amongst different types of objects as well as object affinity information, i.e., intra-type relationships amongst the same types of objects. However, it is difficult to learn accurate intra-type relationships in the presence of noise and outliers. Existing HOCC methods consider the p nearest neighbours based on Euclidean distance for the intra-type relationships, which leads to incomplete and inaccurate intra-type relationships. In this paper, we propose a novel HOCC method that incorporates multiple subspace learning with a heterogeneous manifold ensemble to learn complete and accurate intra-type relationships. Multiple subspace learning reconstructs the similarity between any pair of objects that belong to the same subspace. The heterogeneous manifold ensemble is created based on two-types of intra-type relationships learnt using p-nearest-neighbour graph and multiple subspaces learning. Moreover, in order to make sure the robustness of clustering process, we introduce a sparse error matrix into matrix decomposition and develop a novel iterative algorithm. Empirical experiments show that the proposed method achieves improved results over the state-of-art HOCC methods for FScore and NMI.

Efficient minimax strategies for square loss games

We consider online prediction problems where the loss between the prediction and the outcome is measured by the squared Euclidean distance and its generalization, the squared Mahalanobis distance. We derive the minimax solutions for the case where the prediction and action spaces are the simplex (this setup is sometimes called the Brier game) and the \ell_2 ball (this setup is related to Gaussian density estimation). We show that in both cases the value of each sub-game is a quadratic function of a simple statistic of the state, with coefficients that can be efficiently computed using an explicit recurrence relation. The resulting deterministic minimax strategy and randomized maximin strategy are linear functions of the statistic.

Genetic influences on sulcal patterns of the brain

We present a shape-space approach for analyzing genetic influences on the shapes of the sulcal folding patterns on the cortex. Sulci are represented as continuously parameterized functions in a shape space, and shape differences between sulci are obtained via geodesics between them. The resulting statistical shape analysis framework is used not only to construct populations averages, but also used to compute meaningful correlations within and across groups of sulcal shapes. More importantly, we present a new algorithm that extends the traditional Euclidean estimate of the intra-class correlation to the geometric shape space, thereby allowing us to study heritability of sulcal shape traits for a population of 193 twin pairs. This new methodology reveals strong genetic influences on the sulcal geometry of the cortex.

Comparison of fractional and geodesic anisotropy in diffusion tensor images of 90 monozygotic and dizygotic twins

We used diffusion tensor magnetic resonance imaging (DTI) to reveal the extent of genetic effects on brain fiber microstructure, based on tensor-derived measures, in 22 pairs of monozygotic (MZ) twins and 23 pairs of dizygotic (DZ) twins (90 scans). After Log-Euclidean denoising to remove rank-deficient tensors, DTI volumes were fluidly registered by high-dimensional mapping of co-registered MP-RAGE scans to a geometrically-centered mean neuroanatomical template. After tensor reorientation using the strain of the 3D fluid transformation, we computed two widely used scalar measures of fiber integrity: fractional anisotropy (FA), and geodesic anisotropy (GA), which measures the geodesic distance between tensors in the symmetric positive-definite tensor manifold. Spatial maps of intraclass correlations (r) between MZ and DZ twins were compared to compute maps of Falconer's heritability statistics, i.e. the proportion of population variance explainable by genetic differences among individuals. Cumulative distribution plots (CDF) of effect sizes showed that the manifold measure, GA, comparably the Euclidean measure, FA, in detecting genetic correlations. While maps were relatively noisy, the CDFs showed promise for detecting genetic influences on brain fiber integrity as the current sample expands.

Tensor-based analysis of genetic influences on brain integrity using DTI in 100 twins

Information from the full diffusion tensor (DT) was used to compute voxel-wise genetic contributions to brain fiber microstructure. First, we designed a new multivariate intraclass correlation formula in the log-Euclidean framework. We then analyzed used the full multivariate structure of the tensor in a multivariate version of a voxel-wise maximum-likelihood structural equation model (SEM) that computes the variance contributions in the DTs from genetic (A), common environmental (C) and unique environmental (E) factors. Our algorithm was tested on DT images from 25 identical and 25 fraternal twin pairs. After linear and fluid registration to a mean template, we computed the intraclass correlation and Falconer's heritability statistic for several scalar DT-derived measures and for the full multivariate tensors. Covariance matrices were found from the DTs, and inputted into SEM. Analyzing the full DT enhanced the detection of A and C effects. This approach should empower imaging genetics studies that use DTI.

The multivariate A/C/E model and the genetics of fiber architecture

We present a new algorithm to compute the voxel-wise genetic contribution to brain fiber microstructure using diffusion tensor imaging (DTI) in a dataset of 25 monozygotic (MZ) twins and 25 dizygotic (DZ) twin pairs (100 subjects total). First, the structural and DT scans were linearly co-registered. Structural MR scans were nonlinearly mapped via a 3D fluid transformation to a geometrically centered mean template, and the deformation fields were applied to the DTI volumes. After tensor re-orientation to realign them to the anatomy, we computed several scalar and multivariate DT-derived measures including the geodesic anisotropy (GA), the tensor eigenvalues and the full diffusion tensors. A covariance-weighted distance was measured between twins in the Log-Euclidean framework [2], and used as input to a maximum-likelihood based algorithm to compute the contributions from genetics (A), common environmental factors (C) and unique environmental ones (E) to fiber architecture. Quanititative genetic studies can take advantage of the full information in the diffusion tensor, using covariance weighted distances and statistics on the tensor manifold.

Best individual template selection from deformation tensor minimization

We study the influence of the choice of template in tensor-based morphometry. Using 3D brain MR images from 10 monozygotic twin pairs, we defined a tensor-based distance in the log-Euclidean framework [1] between each image pair in the study. Relative to this metric, twin pairs were found to be closer to each other on average than random pairings, consistent with evidence that brain structure is under strong genetic control. We also computed the intraclass correlation and associated permutation p-value at each voxel for the determinant of the Jacobian matrix of the transformation. The cumulative distribution function (cdf) of the p-values was found at each voxel for each of the templates and compared to the null distribution. Surprisingly, there was very little difference between CDFs of statistics computed from analyses using different templates. As the brain with least log-Euclidean deformation cost, the mean template defined here avoids the blurring caused by creating a synthetic image from a population, and when selected from a large population, avoids bias by being geometrically centered, in a metric that is sensitive enough to anatomical similarity that it can even detect genetic affinity among anatomies.

Complex symbolic sequence clustering and multiple classifiers for predictive process monitoring

This paper addresses the following predictive business process monitoring problem: Given the execution trace of an ongoing case,and given a set of traces of historical (completed) cases, predict the most likely outcome of the ongoing case. In this context, a trace refers to a sequence of events with corresponding payloads, where a payload consists of a set of attribute-value pairs. Meanwhile, an outcome refers to a label associated to completed cases, like, for example, a label indicating that a given case completed “on time” (with respect to a given desired duration) or “late”, or a label indicating that a given case led to a customer complaint or not. The paper tackles this problem via a two-phased approach. In the first phase, prefixes of historical cases are encoded using complex symbolic sequences and clustered. In the second phase, a classifier is built for each of the clusters. To predict the outcome of an ongoing case at runtime given its (uncompleted) trace, we select the closest cluster(s) to the trace in question and apply the respective classifier(s), taking into account the Euclidean distance of the trace from the center of the clusters. We consider two families of clustering algorithms – hierarchical clustering and k-medoids – and use random forests for classification. The approach was evaluated on four real-life datasets.