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.

SAFE: A Declarative Trust-Agile System with Linked Credentials

Secure Access For Everyone (SAFE), is an integrated system for managing trust

using a logic-based declarative language. Logical trust systems authorize each

request by constructing a proof from a context---a set of authenticated logic

statements representing credentials and policies issued by various principals

in a networked system. A key barrier to practical use of logical trust systems

is the problem of managing proof contexts: identifying, validating, and

assembling the credentials and policies that are relevant to each trust

decision.

SAFE addresses this challenge by (i) proposing a distributed authenticated data

repository for storing the credentials and policies; (ii) introducing a

programmable credential discovery and assembly layer that generates the

appropriate tailored context for a given request. The authenticated data

repository is built upon a scalable key-value store with its contents named by

secure identifiers and certified by the issuing principal. The SAFE language

provides scripting primitives to generate and organize logic sets representing

credentials and policies, materialize the logic sets as certificates, and link

them to reflect delegation patterns in the application. The authorizer fetches

the logic sets on demand, then validates and caches them locally for further

use. Upon each request, the authorizer constructs the tailored proof context

and provides it to the SAFE inference for certified validation.

Delegation-driven credential linking with certified data distribution provides

flexible and dynamic policy control enabling security and trust infrastructure

to be agile, while addressing the perennial problems related to today's

certificate infrastructure: automated credential discovery, scalable

revocation, and issuing credentials without relying on centralized authority.

We envision SAFE as a new foundation for building secure network systems. We

used SAFE to build secure services based on case studies drawn from practice:

(i) a secure name service resolver similar to DNS that resolves a name across

multi-domain federated systems; (ii) a secure proxy shim to delegate access

control decisions in a key-value store; (iii) an authorization module for a

networked infrastructure-as-a-service system with a federated trust structure

(NSF GENI initiative); and (iv) a secure cooperative data analytics service

that adheres to individual secrecy constraints while disclosing the data. We

present empirical evaluation based on these case studies and demonstrate that

SAFE supports a wide range of applications with low overhead.

Measuring the impact of a quality improvement collaboration to decrease maternal mortality in a Ghanaian regional hospital.

OBJECTIVE: To evaluate the performance of a continuous quality improvement collaboration at Ridge Regional Hospital, Accra, Ghana, that aimed to halve maternal and neonatal deaths. METHODS: In a quasi-experimental, pre- and post-intervention analysis, system deficiencies were analyzed and 97 improvement activities were implemented from January 2007 to December 2011. Data were collected on outcomes and implementation rates of improvement activities. Severity-adjustment models were used to calculate counterfactual mortality ratios. Regression analysis was used to determine the association between improvement activities, staffing, and maternal mortality. RESULTS: Maternal mortality decreased by 22.4% between 2007 and 2011, from 496 to 385 per 100000 deliveries, despite a 50% increase in deliveries and five- and three-fold increases in the proportion of pregnancies complicated by obstetric hemorrhage and hypertensive disorders of pregnancy, respectively. Case fatality rates for obstetric hemorrhage and hypertensive disorders of pregnancy decreased from 14.8% to 1.6% and 3.1% to 1.1%, respectively. The mean implementation score was 68% for the 97 improvement processes. Overall, 43 maternal deaths were prevented by the intervention; however, risk severity-adjustment models indicated that an even greater number of deaths was averted. Mortality reduction was correlated with 26 continuous quality improvement activities, and with the number of anesthesia nurses and labor midwives. CONCLUSION: The implementation of quality improvement activities was closely correlated with improved maternal mortality.