Learning from Geometry

Subspaces and manifolds are two powerful models for high dimensional signals. Subspaces model linear correlation and are a good fit to signals generated by physical systems, such as frontal images of human faces and multiple sources impinging at an antenna array. Manifolds model sources that are not linearly correlated, but where signals are determined by a small number of parameters. Examples are images of human faces under different poses or expressions, and handwritten digits with varying styles. However, there will always be some degree of model mismatch between the subspace or manifold model and the true statistics of the source. This dissertation exploits subspace and manifold models as prior information in various signal processing and machine learning tasks.

A near-low-rank Gaussian mixture model measures proximity to a union of linear or affine subspaces. This simple model can effectively capture the signal distribution when each class is near a subspace. This dissertation studies how the pairwise geometry between these subspaces affects classification performance. When model mismatch is vanishingly small, the probability of misclassification is determined by the product of the sines of the principal angles between subspaces. When the model mismatch is more significant, the probability of misclassification is determined by the sum of the squares of the sines of the principal angles. Reliability of classification is derived in terms of the distribution of signal energy across principal vectors. Larger principal angles lead to smaller classification error, motivating a linear transform that optimizes principal angles. This linear transformation, termed TRAIT, also preserves some specific features in each class, being complementary to a recently developed Low Rank Transform (LRT). Moreover, when the model mismatch is more significant, TRAIT shows superior performance compared to LRT.

The manifold model enforces a constraint on the freedom of data variation. Learning features that are robust to data variation is very important, especially when the size of the training set is small. A learning machine with large numbers of parameters, e.g., deep neural network, can well describe a very complicated data distribution. However, it is also more likely to be sensitive to small perturbations of the data, and to suffer from suffer from degraded performance when generalizing to unseen (test) data.

From the perspective of complexity of function classes, such a learning machine has a huge capacity (complexity), which tends to overfit. The manifold model provides us with a way of regularizing the learning machine, so as to reduce the generalization error, therefore mitigate overfiting. Two different overfiting-preventing approaches are proposed, one from the perspective of data variation, the other from capacity/complexity control. In the first approach, the learning machine is encouraged to make decisions that vary smoothly for data points in local neighborhoods on the manifold. In the second approach, a graph adjacency matrix is derived for the manifold, and the learned features are encouraged to be aligned with the principal components of this adjacency matrix. Experimental results on benchmark datasets are demonstrated, showing an obvious advantage of the proposed approaches when the training set is small.

Stochastic optimization makes it possible to track a slowly varying subspace underlying streaming data. By approximating local neighborhoods using affine subspaces, a slowly varying manifold can be efficiently tracked as well, even with corrupted and noisy data. The more the local neighborhoods, the better the approximation, but the higher the computational complexity. A multiscale approximation scheme is proposed, where the local approximating subspaces are organized in a tree structure. Splitting and merging of the tree nodes then allows efficient control of the number of neighbourhoods. Deviation (of each datum) from the learned model is estimated, yielding a series of statistics for anomaly detection. This framework extends the classical {\em changepoint detection} technique, which only works for one dimensional signals. Simulations and experiments highlight the robustness and efficacy of the proposed approach in detecting an abrupt change in an otherwise slowly varying low-dimensional manifold.

Nucleolar organization, ribosomal DNA array stability, and acrocentric chromosome integrity are linked to telomere function.

The short arms of the ten acrocentric human chromosomes share several repetitive DNAs, including ribosomal RNA genes (rDNA). The rDNA arrays correspond to nucleolar organizing regions that coalesce each cell cycle to form the nucleolus. Telomere disruption by expressing a mutant version of telomere binding protein TRF2 (dnTRF2) causes non-random acrocentric fusions, as well as large-scale nucleolar defects. The mechanisms responsible for acrocentric chromosome sensitivity to dysfunctional telomeres are unclear. In this study, we show that TRF2 normally associates with the nucleolus and rDNA. However, when telomeres are crippled by dnTRF2 or RNAi knockdown of TRF2, gross nucleolar and chromosomal changes occur. We used the controllable dnTRF2 system to precisely dissect the timing and progression of nucleolar and chromosomal instability induced by telomere dysfunction, demonstrating that nucleolar changes precede the DNA damage and morphological changes that occur at acrocentric short arms. The rDNA repeat arrays on the short arms decondense, and are coated by RNA polymerase I transcription binding factor UBF, physically linking acrocentrics to one another as they become fusogenic. These results highlight the importance of telomere function in nucleolar stability and structural integrity of acrocentric chromosomes, particularly the rDNA arrays. Telomeric stress is widely accepted to cause DNA damage at chromosome ends, but our findings suggest that it also disrupts chromosome structure beyond the telomere region, specifically within the rDNA arrays located on acrocentric chromosomes. These results have relevance for Robertsonian translocation formation in humans and mechanisms by which acrocentric-acrocentric fusions are promoted by DNA damage and repair.

Acoustic radiation force impulse imaging (ARFI) on an IVUS circular array.

Our long-term goal is the detection and characterization of vulnerable plaque in the coronary arteries of the heart using intravascular ultrasound (IVUS) catheters. Vulnerable plaque, characterized by a thin fibrous cap and a soft, lipid-rich necrotic core is a precursor to heart attack and stroke. Early detection of such plaques may potentially alter the course of treatment of the patient to prevent ischemic events. We have previously described the characterization of carotid plaques using external linear arrays operating at 9 MHz. In addition, we previously modified circular array IVUS catheters by short-circuiting several neighboring elements to produce fixed beamwidths for intravascular hyperthermia applications. In this paper, we modified Volcano Visions 8.2 French, 9 MHz catheters and Volcano Platinum 3.5 French, 20 MHz catheters by short-circuiting portions of the array for acoustic radiation force impulse imaging (ARFI) applications. The catheters had an effective transmit aperture size of 2 mm and 1.5 mm, respectively. The catheters were connected to a Verasonics scanner and driven with pushing pulses of 180 V p-p to acquire ARFI data from a soft gel phantom with a Young's modulus of 2.9 kPa. The dynamic response of the tissue-mimicking material demonstrates a typical ARFI motion of 1 to 2 microns as the gel phantom displaces away and recovers back to its normal position. The hardware modifications applied to our IVUS catheters mimic potential beamforming modifications that could be implemented on IVUS scanners. Our results demonstrate that the generation of radiation force from IVUS catheters and the development of intravascular ARFI may be feasible.