This dissertation comprises five papers that focus on a novel paradigm of privacy attack, i.e., model inversion (MI) attack, where the adversarial goal is to infer or reconstruct training samples. In particular, these works are aligned with investigating MI privacy attacks, designing novel realistic MI attacks under restricted realistic capabilities, and introducing novel robust defense techniques against these attacks. At first, we focus on the systematization of MI attacks from the literature review (IEEE CSF). This opened up ways to investigate MI attacks on the tabular dataset. We developed novel MI attacks for inferring sensitive private training data, published in USENIX Security. Then, we worked on exploring MI attacks with limited adversarial capabilities (IEEE SaTML), i.e., when adversaries do not have access to the same data distributions as model training data. All these streams of work on privacy attack designing enabled the design of novel defenses against MI attacks. We have developed a novel sparse coding architecture (SCA), which shows 1.1-18.3 times more robustness against MI attacks while not significantly compromising model accuracy. This exciting work has just been published at ECCV 2024 this year and inspires us to improve the defense further by designing systematic techniques to drop highly sensitive features during training that can also provide provable privacy bounds.

The introduction of the Internet of Things (IoT) has made several emerging applications, from financial transactions to property access, possible through IoT-connected smart wearables (smartwatches). This creates an immediate need for an authentication system that can validate a user seamlessly, compared to knowledge-based approaches. In this work, we present an implicit authentication system that utilizes a bag of on-phone artificial neural network (ANN) models to validate a user based on the availability of three soft-biometrics (heart rate, gait, and breathing patterns) collected from smartphones and Fitbits. We find that using all three biometrics we can achieve an average accuracy of up to$.973 \pm . 004$. Next, we implement the bag of models on smartphones using Google's TensorFlow Lite framework-supportedTFL Authapplication, which requires around 56-65 KB memory and can verify a user in 5 seconds. Finally, we evaluate the systemTFL Authusing two cohorts of 25 subjects in total, and we find that the system has average understandability and importance scores of around 4.0 and 4.3 on a 1 – 5 scale.

A crucial module of the widely applied machine learning (ML) model is the model training phase, which involves large-scale training data, often including sensitive private data. ML models trained on these sensitive data suffer from significant privacy concerns since ML models can intentionally or unintendedly leak information about training data. Adversaries can exploit this information to perform privacy attacks, including model extraction, membership inference, and model inversion. While a model extraction attack steals and replicates a trained model functionality, and membership inference infers the data sample's inclusiveness to the training set, a model inversion attack has the goal of inferring the training data sample's sensitive attribute value or reconstructing the training sample (i.e., image/audio/text). Distinct and inconsistent characteristics of model inversion attack make this attack even more challenging and consequential, opening up model inversion attack as a more prominent and increasingly expanding research paradigm. Thereby, to flourish research in this relatively underexplored model inversion domain, we conduct the first-ever systematic literature review of the model inversion attack landscape. We characterize model inversion attacks and provide a comprehensive taxonomy based on different dimensions. We illustrate foundational perspectives emphasizing methodologies and key principles of the existing attacks and defense techniques. Finally, we discuss challenges and open issues in the existing model inversion attacks, focusing on the roadmap for future research directions.

Respiratory diseases, such as chronic obstructive pulmonary disease (COPD) and asthma, are two major reasons for people's death across the globe. In addition to these common inflammatory respiratory diseases, some human transmissible respiratory diseases, such as coronaviruses, cause a global pandemic. One major symptom of these inflammatory respiratory diseases is coughing. Identifying coughing using smartphone-microphone recordings is easily doable from a remote setup and can help physicians and researchers early guess a situation for an individual and a community. However, smartphone-microphone recordings can be affected by environmental noises and that can impact the performance of models that are developed to detect coughing from microphone recording. Thereby, in this work, we present a detailed analysis of noise impacts on cough detection models. We develop models using voluntary coughs and other background sounds obtained from three public datasets and test the performance of those models while detecting various types of coughs, including COPD and COVID-19, obtain from three separate datasets in the presence of background noises.

When high school students leave their homes for a college education, they often face enormous changes and challenges in life, such as meeting new people, more responsibilities in life, and being away from family and their comfort zones. These sudden changes often lead to an elevation of stress and anxiety, affecting a student’s health and well-being. Situations can even get worse in the age of global pandemics, such as COVID-19, when regular life and social activities are significantly disrupted due to lockdown or stay-at-home orders. Therefore, predicting phone call patterns (a measure of social engagement) based on various factors and activities of a person can be helpful to foster social engagement and promote health and well-being during sudden lifestyle changes. In this work, we investigate a cohort of 370 on-campus college students over three consecutive semesters and breaks between them to find various geo-temporal factors and activities that affect students’ phone call behaviors and develop models that can predict the next call duration with a correlation of up to 0.89 between the actual and predicted duration using individual-level generalized linear models. Findings from this work can further be extended to other populations, and thereby, our findings will enable the design and delivery of new smartphone-based health interventions (guided feedback) to help people to adapt and cope up with situations that affect their lifestyle and social activities.

This paper presents a new technique for designing a compact and an efficient quantum divider circuit. A backtracking technique has been incorporated to reduce the depth of the proposed design. In addition, we propose a new quantum full-subtractor circuit and a quantum ANDing circuit to perform the AND operation of the quantum bits. The proposed quantum divider circuit requires a linear depth, whereas the depth of the design of the best known existing quantum divider circuit is exponential. The comparative study shows that the proposed quantum divider improves 67.51% in terms of quantum gates, area, power and 53.33% in terms of constant inputs and 36.36% in terms of garbage outputs over the best known existing quantum divider when the dividend is a 64-qubit.