ARUNESH SINHA
Research Interest
My research interest spans game theory and machine learning with a particular focus on adversarial settings. My work is driven by real world problems with the goal of applying principled methods in practice.
Current and Past Projects
  • Robust Learning: In adversarial settings learning can be exploited to fool defenders. But, even without adversaries, in past work we have shown that learning followed by optimization does not result in optimal outcomes and the learning goal needs to be modified to rectify this mismatch [AAMAS 16, IJCAI 16]. We have also shown how the adversary''s fooling of Bayesian update in repeated games can be thwarted by the defender to a large extent [AAAI 19]. I continue to explore this area especially from the deception angle.
  • Incentive Design in Adversarial Settings: I have worked on problems of optimal price design in private data marketplaces in presence of privacy adversaries, where the solution is to sell data with additive noise and charge a premium for high quality (low noise) data [EC 19]. I have also worked on scoring mechanism design for competitions in which the goal is to test how well the competitors collaborate to improve the social welfare [AAMAS 19]. An example of such a competition is the DARPA SC2 competition.
  • Threat Screening: The Transport Security Administration has launched a new initiative known as Dynamic Aviation Risk Management Solution (DARMS) which aims to enhance aviation security. Dubbed as the "future of aviation security", this ambitious project aims to provide a mathematically sound basis for intelligent screening of passengers. The TSA utilizes a number of screening resources for screening passengers, e.g., x-ray machines, metal detectors, pat-downs, etc. However, the most effective screening resources cannot be used to screen every passenger as there exists capacity bounds on the usage of each screening resource with lower capacity for more effective resources. Thus, given the risk category of each passenger, the number of passengers and the efficacy of resources in detecting threats, the goal is to maximize the quality of security screening subject to the underlying passenger flow constraints. As part of passenger screening component of the DARMS projects, I proposed a novel Stackelberg game model for screening of threats (Threat Screening Games). Combining team formation aspects for the capacity-constrained screening resources with the need to screen every passenger in every risk category as effectively as possible resulted in a huge (exponentially large) linear program (LP) required to compute equilibrium of the Stackelberg game. I provided techniques for scaling up the equilibrium computational; the techniques includes a novel temporal decomposition of the game and a novel alternate representation of the LP obtained by using a convex combination of a number of polytopes to represent the feasible space of the LP. This work has been tested in US airports.
  • Violation Screening: The HIPAA privacy law was passed in 1996 to prevent inappropriate flow of private health information. Given the undesirable consequences of restricting flow of health information (e.g., in a medical emergency), post-hoc audits of health record accesses is the preferred means of detection and remediation of HIPAA breaches in hospitals. However, the complicated privacy policies do not permit completely automated auditing, thus, the few human auditors are faced with a large number of suspicious cases to audit. Indeed, ad-hoc audit practices has resulted in many HIPAA breaches. I proposed a Stackelberg game model [IJCAI 13, AAAI 15] of the interaction between the auditor and auditee. An important component of this auditor-auditee interaction is punishments; judicious use of punishments can shape the behavior of the auditee resulting in desirable outcomes. I provided scalable algorithms to compute the solution of the non-convex optimization involved in computing the Stackelberg equilibrium of the game.
  • Crime Screening: The Department of Public Safety (DPS) at University of Southern California (USC) is tasked with ensuring security in and around USC campus, and similar to other security agencies, the number of patrol officers available is quite limited. DPS has records for past crime and patrol allocations over multiple years, and their goal is to conduct preventive patrols intelligently by predicting crime. I recognized that it is almost impossible to estimate utilities for petty criminals and moreover it had been shown empirically that criminals do not exhibit strategic behavior in practice. Also, while it is amazing that regret based approaches provide guarantees in an arbitrary adversary setting, I also recognized that adversaries may not be completely arbitrary and are more likely to exhibit patterns in their behavior which can be exploited to obtain better guarantees. Thus, rather than using poor estimates of utility and uncertain models of bounded rationality or regret based approaches, I proposed a technique that directly learns the behavior of the criminals. I proposed a Dynamic Bayesian Network (DBN) based model [AAMAS 15] of predicting crimes from available crime and patrol data; the parameters of the model capture the behavior of the criminals.