Health Care : Current Reviews

EFFECTIVE AND SECURE HEALTH CARE USING ADVERSARIAL LEARNING

7^th World Congress on Healthcare & Technologies

September 26-27, 2016 London, UK

Harry Wechsler

George Mason University, USA

Posters & Accepted Abstracts: Health Care: Current Reviews

Abstract:

It is crucial for both biological (e.g., immune system) and machine-based systems to recognize patterns and messages as a friend or foe and to respond to them appropriately. We consider here the use of adversarial learning to enhance defenses against adaptive, malicious, and persistent offensive threats, and towards such ends, we propose conformal prediction as the principled and a unified learning framework to design, develop, and deploy a multi-faceted protective and self-managing defensive shield to detect, disrupt, and deny intrusive attacks, hostile and malicious behavior, and subterfuge. Conformal prediction supports a multitude of functional blocks that address the major challenges faced by adversarial learning, including denial and deception, adequate message representation and classification, and platform vulnerabilities, deliberate or not, affecting learning, training, and annotation. The solutions proffered are built around active learning, meta-reasoning, randomness, immunity, semantics and stratification, and most important and above all, around adaptive Oracles that are effective and valid regarding model selection and prediction. The motivation for using conformal prediction and its immediate offspring, those of semi-supervised learning and transduction, comes from them supporting discriminative and non-parametric methods using likelihood ratios; demarcation using cohorts, local estimation, and non-conformity measures; randomness for hypothesis testing and inference using sensitivity analysis; reliability indices on prediction outcomes using credibility and confidence; open set recognition including the reject option and negative selection; and consensus reasoning to upend questionable label annotation, deliberate or not, using aggregation and importance sampling.

Biography :

Email: wechsler@gmu.edu