SoC Research Publications
Permanent URI for this collection
Browse
Recent Submissions
Item Non-parametric classification via expand-and-sparsify representation(Neural information processing systems foundation, 2024-12-09) Sinha, Kaushik; Globerson A.; Mackey L.; Belgrave D.; Fan A.; Paquet U.; Tomczak J.; Zhang C.In expand-and-sparsify (EaS) representation, a data point in Sd-1 is first randomly mapped to higher dimension ℝm, where m > d, followed by a sparsification operation where the informative k ≪ m of the m coordinates are set to one and the rest are set to zero. We propose two algorithms for non-parametric classification using such EaS representation. For our first algorithm, we use winners-take-all operation for the sparsification step and show that the proposed classifier admits the form of a locally weighted average classifier and establish its consistency via Stone's Theorem. Further, assuming that the conditional probability function P(y = 1|x) = η(x) is Hölder continuous and for optimal choice of m, we show that the convergence rate of this classifier is minimax-optimal. For our second algorithm, we use empirical k-thresholding operation for the sparsification step, and under the assumption that data lie on a low dimensional manifold of dimension d0 ≪ d, we show that the convergence rate of this classifier depends only on d0 and is again minimax-optimal. Empirical evaluations performed on real-world datasets corroborate our theoretical results. © 2024 Neural information processing systems foundation. All rights reserved.Item Leveraging diffusion and Flow Matching Models for demographic bias mitigation of facial attribute classifiers(Elsevier B.V., 2025-03-23) Ramachandran, Sreeraj; Rattani, AjitaPublished research highlights the presence of demographic bias in automated facial attribute classification algorithms, notably impacting women and individuals with darker skin tones. Proposed bias mitigation techniques are not generalizable, need demographic annotations, are application-specific, and often obtain fairness by reducing overall classification accuracy. In response to these challenges, this paper proposes a novel bias mitigation technique that systematically integrates diffusion and flow-matching models with a base classifier with minimal additional computational overhead. These generative models are chosen for their extreme success in capturing diverse data distributions and their inherent stochasticity. Our proposed approach augments the base classifier's accuracy across all demographic sub-groups with enhanced fairness. Further, the stochastic nature of these generative models is harnessed to quantify prediction uncertainty, allowing for test-time rejection, which further enhances fairness. Additionally, novel solvers are proposed to significantly reduce the computational overhead of generative model inference. An exhaustive evaluation carried out on facial attribute annotated datasets substantiates the efficacy of our approach in enhancing the accuracy and fairness of facial attribute classifiers by 0.5%−3% and 0.5%−5% across datasets over SOTA mitigation techniques. Thus, obtaining state-of-the-art performance. Further, our proposal does not need a demographically annotated training set and is generalizable to any downstream classification task. © 2025 Elsevier B.V.Item Omnisurface: Common reality for intuitive human-robot collaboration(Springer Science and Business Media Deutschland GmbH, 2025-02-07) Zaman, Akhlak Uz; Li, Hui; Yan, Fujian; Zhang, Yinlong; He, Hongsheng; Li,Haizhou; Zhu, Jian; Schultz, Tanja; Bi, Yalei; He, Hongsheng; Ma, Jun; Cai, Siqi; Ge, Shuzhi Sam; Jiang, WanyueEffective communication and information projection are essential for human-robot teaming. The projection of images on non-planar surfaces using a conventional projector is challenging due to the inherent problem of distortion. The projection distortion occurs due to the variations in depth across the surface of the teaming workspace. As a result, the projected image, information, or symbols lose their original shape and create confusion during human-robot teaming. In this paper, we presented an innovative approach to perform distortion-free projections in the teaming workspace. A pre-warped image is constructed based on the surface geometry that the projector displays and accurately replicates the original projection image. Beyond the technical achievement, this research highlights the social acceptance of improved spatial augmented reality in human-robot teams. It fosters better teamwork, trust, and efficiency by enabling more intuitive and reliable interactions. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.Item Synergistic machine learning approaches for early lung cancer detection and improved prognostics(Institute of Electrical and Electronics Engineers Inc., 2024-11-26) Ahamed, Imtiaj Uddin; Hossain, Al-Amin; Gupta, Uchchas Das; Ahamed, Imtiaj Uddin; Anika, Ayvee Nusreen; Islam, AlveeIn the battle against lung cancer, early detection remains the best weapon, yet it's a field where success has been limited by the shadows of delayed diagnosis. This study explores the field of advanced machine learning, embracing the potential of state-of-the-art algorithms to revolutionize lung cancer diagnoses. By utilizing a meticulously selected dataset, we reveal the seamless synergy between advanced computational methods: the accuracy of Bayesian Optimized ExtraTrees and the flexibility of LightGBM. Upon thorough examination of the data, we have unearthed a remarkable discovery. The Bayesian Optimized ExtraTrees model has exhibited exceptional accuracy, with a score of 97%, and an impressive ROC-AUC score of 99.50%. This discovery indicates a new era of diagnostic precision that holds the promise of revolutionizing the field. This leap in performance illuminates a path forward, suggesting that the fusion of advanced machine-learning methods can be a game-changer in the timely detection of lung cancer, thereby kindling the flames of hope for improved patient outcomes. Our exploration underscores the revolutionary impact of weaving complex analytical threads into the fabric of medical diagnostics, charting a course for future breakthroughs in the early detection and treatment paradigms of cancer. Future research should investigate larger, more diverse datasets, explore deep neural networks, and incorporate feder-ated learning to address privacy concerns. © 2024 IEEE.Item Exploring sleep patterns and behavioral components for stress detection: A machine learning approach(Institute of Electrical and Electronics Engineers Inc., 2024-11-26) Hossain, Al-Amin; Ahamed, Imtiaj Uddin; Gupta, Uchchas Das; Anika, Ayvee Nusreen; Islam, Alvee; Saha, UtshaStress, sadness, and anxiety have become prevalent issues in our everyday lives. The primary aim of this research is to employ Machine Learning (ML) techniques to identify stress in individuals, ultimately aiming to improve their day-to-day well-being. Utilizing the publicly available multi modal dataset SayoPillow, we propose various ML models for detecting stress in humans during sleep. With advancements in wearable technology, numerous physiological factors can now be continuously monitored. The significance of stress detection has increased notably, as early identification can aid in better health management and mitigate the adverse effects of prolonged stress exposure. This study focuses on identifying optimal ML models for assessing stress levels during sleep. We have implemented several machine learning-based models to forecast individuals' stress levels. To provide a comprehensive overview of the approaches in this domain, we initially identify sleep patterns and key components of sleep assessment. We scrutinize the principal technical advancements in sleep behavior analysis, data collection, and monitoring, outlining their advantages and disadvantages. Additionally, we discuss potential research challenges and future directions. The performance metrics such as F1-score, recall, and accuracy were calculated and compared across various ML models. © 2024 IEEE.