Susmita Sur-Kolay

Bio: Susmita Sur-Kolay is an academic researcher from Indian Statistical Institute. The author has contributed to research in topics: Floorplan & Qubit. The author has an hindex of 19, co-authored 121 publications receiving 1274 citations. Previous affiliations of Susmita Sur-Kolay include Jadavpur University.

Systematic Poisoning Attacks on and Defenses for Machine Learning in Healthcare

Abstract: Machine learning is being used in a wide range of application domains to discover patterns in large datasets. Increasingly, the results of machine learning drive critical decisions in applications related to healthcare and biomedicine. Such health-related applications are often sensitive, and thus, any security breach would be catastrophic. Naturally, the integrity of the results computed by machine learning is of great importance. Recent research has shown that some machine-learning algorithms can be compromised by augmenting their training datasets with malicious data, leading to a new class of attacks called poisoning attacks. Hindrance of a diagnosis may have life-threatening consequences and could cause distrust. On the other hand, not only may a false diagnosis prompt users to distrust the machine-learning algorithm and even abandon the entire system but also such a false positive classification may cause patient distress. In this paper, we present a systematic, algorithm-independent approach for mounting poisoning attacks across a wide range of machine-learning algorithms and healthcare datasets. The proposed attack procedure generates input data, which, when added to the training set, can either cause the results of machine learning to have targeted errors (e.g., increase the likelihood of classification into a specific class), or simply introduce arbitrary errors (incorrect classification). These attacks may be applied to both fixed and evolving datasets. They can be applied even when only statistics of the training dataset are available or, in some cases, even without access to the training dataset, although at a lower efficacy. We establish the effectiveness of the proposed attacks using a suite of six machine-learning algorithms and five healthcare datasets. Finally, we present countermeasures against the proposed generic attacks that are based on tracking and detecting deviations in various accuracy metrics, and benchmark their effectiveness.

Wearable Medical Sensor-Based System Design: A Survey

Abstract: Wearable medical sensors (WMSs) are garnering ever-increasing attention from both the scientific community and the industry. Driven by technological advances in sensing, wireless communication, and machine learning, WMS-based systems have begun transforming our daily lives. Although WMSs were initially developed to enable low-cost solutions for continuous health monitoring, the applications of WMS-based systems now range far beyond health care. Several research efforts have proposed the use of such systems in diverse application domains, e.g., education, human-computer interaction, and security. Even though the number of such research studies has grown drastically in the last few years, the potential challenges associated with their design, development, and implementation are neither well-studied nor well-recognized. This article discusses various services, applications, and systems that have been developed based on WMSs and sheds light on their design goals and challenges. We first provide a brief history of WMSs and discuss how their market is growing. We then discuss the scope of applications of WMS-based systems. Next, we describe the architecture of a typical WMS-based system and the components that constitute such a system, and their limitations. Thereafter, we suggest a list of desirable design goals that WMS-based systems should satisfy. Finally, we discuss various research directions related to WMSs and how previous research studies have attempted to address the limitations of the components used in WMS-based systems and satisfy the desirable design goals.

Energy-Efficient Long-term Continuous Personal Health Monitoring

Abstract: Continuous health monitoring using wireless body area networks of implantable and wearable medical devices (IWMDs) is envisioned as a transformative approach to healthcare. Rapid advances in biomedical sensors, low-power electronics, and wireless communications have brought this vision to the verge of reality. However, key challenges still remain to be addressed. The constrained sizes of IWMDs imply that they are designed with very limited processing, storage, and battery capacities. Therefore, there is a very strong need for efficiency in data collection, analysis, storage, and communication. In this paper, we first quantify the energy and storage requirements of a continuous personal health monitoring system that uses eight biomedical sensors: (1) heart rate, (2) blood pressure, (3) oxygen saturation, (4) body temperature, (5) blood glucose, (6) accelerometer, (7) electrocardiogram (ECG), and (8) electroencephalogram (EEG). Our analysis suggests that there exists a significant gap between the energy and storage requirements for long-term continuous monitoring and the capabilities of current devices. To enable energy-efficient continuous health monitoring, we propose schemes for sample aggregation, anomaly-driven transmission, and compressive sensing to reduce the overheads of wirelessly transmitting, storing, and encrypting/authenticating the data. We evaluate these techniques and demonstrate that they result in two to three orders-of-magnitude improvements in energy and storage requirements, and can help realize the potential of long-term continuous health monitoring.

A modeling approach for addressing power supply switching noise related failures of integrated circuits

Abstract: Power density of high-end microprocessors has been increasing by approximately 80% per technology generation, while the voltage is scaling by a factor of 0.8. This leads to 225% increase in current per unit area in successive generation of technologies. The cost of maintaining the same IR drop becomes too high. This leads to compromise in power delivery and power grid becomes a performance limiter. Traditional performance related test techniques with transition and path delay fault models focus on testing the logic but not the power delivery. In this paper we view power grid as performance limiter and develop a fault model to address the problem of vector generation for delay faults arising out of power delivery problems. A fault extraction methodology applied to a microprocessor design block is explained.

PAQCS: Physical Design-Aware Fault-Tolerant Quantum Circuit Synthesis

Abstract: Quantum circuits consist of a cascade of quantum gates. In a physical design-unaware quantum logic circuit, a gate is assumed to operate on an arbitrary set of quantum bits (qubits), without considering the physical location of the qubits. However, in reality, physical qubits have to be placed on a grid. Each node of the grid represents a qubit. The grid implements the architecture of the quantum computer. A physical constraint often imposed is that quantum gates can only operate on adjacent qubits on the grid. Hence, a communication channel needs to be built if the qubits in the logical circuit are not adjacent. In this paper, we introduce a tool called the physical design-aware fault-tolerant quantum circuit synthesis (PAQCS). It contains two algorithms: one for physical qubit placement and another for routing of communications. With the help of these two algorithms, the overhead of converting a logical to a physical circuit is reduced by 30.1%, on an average, relative to previous work. The optimization algorithms in PAQCS are evaluated on circuits implemented using quantum operations supported by two different quantum physical machine descriptions and three quantum error-correcting codes. They reduce the number of primitive operations by 11.5%–68.6%, and the number of execution cycles by 16.9%–59.4%.

Adversarial Examples: Attacks and Defenses for Deep Learning

Abstract: With rapid progress and significant successes in a wide spectrum of applications, deep learning is being applied in many safety-critical environments. However, deep neural networks (DNNs) have been recently found vulnerable to well-designed input samples called adversarial examples . Adversarial perturbations are imperceptible to human but can easily fool DNNs in the testing/deploying stage. The vulnerability to adversarial examples becomes one of the major risks for applying DNNs in safety-critical environments. Therefore, attacks and defenses on adversarial examples draw great attention. In this paper, we review recent findings on adversarial examples for DNNs, summarize the methods for generating adversarial examples, and propose a taxonomy of these methods. Under the taxonomy, applications for adversarial examples are investigated. We further elaborate on countermeasures for adversarial examples. In addition, three major challenges in adversarial examples and the potential solutions are discussed.

Neural Cleanse: Identifying and Mitigating Backdoor Attacks in Neural Networks

Abstract: Lack of transparency in deep neural networks (DNNs) make them susceptible to backdoor attacks, where hidden associations or triggers override normal classification to produce unexpected results. For example, a model with a backdoor always identifies a face as Bill Gates if a specific symbol is present in the input. Backdoors can stay hidden indefinitely until activated by an input, and present a serious security risk to many security or safety related applications, e.g. biometric authentication systems or self-driving cars. We present the first robust and generalizable detection and mitigation system for DNN backdoor attacks. Our techniques identify backdoors and reconstruct possible triggers. We identify multiple mitigation techniques via input filters, neuron pruning and unlearning. We demonstrate their efficacy via extensive experiments on a variety of DNNs, against two types of backdoor injection methods identified by prior work. Our techniques also prove robust against a number of variants of the backdoor attack.

A Simulated Annealing-Based Multiobjective Optimization Algorithm: AMOSA

Abstract: This paper describes a simulated annealing based multiobjective optimization algorithm that incorporates the concept of archive in order to provide a set of tradeoff solutions for the problem under consideration. To determine the acceptance probability of a new solution vis-a-vis the current solution, an elaborate procedure is followed that takes into account the domination status of the new solution with the current solution, as well as those in the archive. A measure of the amount of domination between two solutions is also used for this purpose. A complexity analysis of the proposed algorithm is provided. An extensive comparative study of the proposed algorithm with two other existing and well-known multiobjective evolutionary algorithms (MOEAs) demonstrate the effectiveness of the former with respect to five existing performance measures, and several test problems of varying degrees of difficulty. In particular, the proposed algorithm is found to be significantly superior for many objective test problems (e.g., 4, 5, 10, and 15 objective problems), while recent studies have indicated that the Pareto ranking-based MOEAs perform poorly for such problems. In a part of the investigation, comparison of the real-coded version of the proposed algorithm is conducted with a very recent multiobjective simulated annealing algorithm, where the performance of the former is found to be generally superior to that of the latter.

Multi-sensor fusion in body sensor networks: State-of-the-art and research challenges

Abstract: Body Sensor Networks (BSNs) have emerged as a revolutionary technology in many application domains in health-care, fitness, smart cities, and many other compelling Internet of Things (IoT) applications. Most commercially available systems assume that a single device monitors a plethora of user information. In reality, BSN technology is transitioning to multi-device synchronous measurement environments; fusion of the data from multiple, potentially heterogeneous, sensor sources is therefore becoming a fundamental yet non-trivial task that directly impacts application performance. Nevertheless, only recently researchers have started developing technical solutions for effective fusion of BSN data. To the best of our knowledge, the community is currently lacking a comprehensive review of the state-of-the-art techniques on multi-sensor fusion in the area of BSN. This survey discusses clear motivations and advantages of multi-sensor data fusion and particularly focuses on physical activity recognition, aiming at providing a systematic categorization and common comparison framework of the literature, by identifying distinctive properties and parameters affecting data fusion design choices at different levels (data, feature, and decision). The survey also covers data fusion in the domains of emotion recognition and general-health and introduce relevant directions and challenges of future research on multi-sensor fusion in the BSN domain.

A Comprehensive Study of Security of Internet-of-Things

Abstract: Internet of Things (IoT), also referred to as the Internet of Objects, is envisioned as a transformative approach for providing numerous services. Compact smart devices constitute an essential part of IoT. They range widely in use, size, energy capacity, and computation power. However, the integration of these smart things into the standard Internet introduces several security challenges because the majority of Internet technologies and communication protocols were not designed to support IoT. Moreover, commercialization of IoT has led to public security concerns, including personal privacy issues, threat of cyber attacks, and organized crime. In order to provide a guideline for those who want to investigate IoT security and contribute to its improvement, this survey attempts to provide a comprehensive list of vulnerabilities and countermeasures against them on the edge-side layer of IoT, which consists of three levels: (i) edge nodes, (ii) communication, and (iii) edge computing. To achieve this goal, we first briefly describe three widely-known IoT reference models and define security in the context of IoT. Second, we discuss the possible applications of IoT and potential motivations of the attackers who target this new paradigm. Third, we discuss different attacks and threats. Fourth, we describe possible countermeasures against these attacks. Finally, we introduce two emerging security challenges not yet explained in detail in previous literature.