Yusuf Osmanlıoğlu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Yusuf Osmanlıoğlu has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
Ali Shokoufandeh
Drexel University
10 shared publications- Ali Shokoufandeh
Drexel University
7 shared publications Drew Parker
University of Pennsylvania
7 shared publications- Daniel Schwartz
Drexel University
5 shared publications Ragini Verma
University of Pennsylvania
5 shared publications- Ragini Verma
University of Pennsylvania, Philadelphia, PA, USA
5 shared publications
Latest Publications
- Open Access
- 29 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Human-Computer Interaction
- Vol. 10 (4) , 1-25
Resource-Efficient Gesture Recognition through Convexified Attention EICS006
Wearable e-textile interfaces require gesture recognition capabilities but face severe constraints in power consumption, computational capacity, and form factor that make traditional deep learning impractical. While lightweight architectures like MobileNet improve efficiency, they still demand thousands of parameters, limiting deployment on textile-integrated platforms. We introduce a convexified attention mechanism for wearable applications that dynamically weights features while preserving convexity through nonexpansive simplex projection and convex loss functions. Unlike conventional attention mechanisms using non-convex softmax operations, our approach employs Euclidean projection onto the probability simplex combined with multi-class hinge loss, ensuring global convergence guarantees. Implemented on a textile-based capacitive sensor with four connection points, our approach achieves 100.00% accuracy on tap gestures and 100.00% on swipe gestures—consistent across 10-fold cross-validation and held-out test evaluation—while requiring only 120–360 parameters, a 97% reduction compared to conventional approaches. With sub-millisecond inference times (290–296 μ s) and minimal storage requirements (< 7KB), our method enables gesture interfaces directly within e-textiles without external processing. Our evaluation, conducted in controlled laboratory conditions with a single-user dataset, demonstrates feasibility for basic gesture interactions. Real-world deployment would require validation across multiple users, environmental conditions, and more complex gesture vocabularies. These results demonstrate how convex optimization can enable efficient on-device machine learning for textile interfaces.
- 29 June 2026
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Human-Computer Interaction
- Vol. 10 (4) , 1-25
Open AccessResource-Efficient Gesture Recognition through Convexified Attention EICS006
Wearable e-textile interfaces require gesture recognition capabilities but face severe constraints in power consumption, computational capacity, and form factor that make traditional deep learning impractical. While lightweight architectures like MobileNet improve efficiency, they still demand thousands of parameters, limiting deployment on textile-integrated platforms. We introduce a convexified attention mechanism for wearable applications that dynamically weights features while preserving convexity through nonexpansive simplex projection and convex loss functions. Unlike conventional attention mechanisms using non-convex softmax operations, our approach employs Euclidean projection onto the probability simplex combined with multi-class hinge loss, ensuring global convergence guarantees. Implemented on a textile-based capacitive sensor with four connection points, our approach achieves 100.00% accuracy on tap gestures and 100.00% on swipe gestures—consistent across 10-fold cross-validation and held-out test evaluation—while requiring only 120–360 parameters, a 97% reduction compared to conventional approaches. With sub-millisecond inference times (290–296 μ s) and minimal storage requirements (< 7KB), our method enables gesture interfaces directly within e-textiles without external processing. Our evaluation, conducted in controlled laboratory conditions with a single-user dataset, demonstrates feasibility for basic gesture interactions. Real-world deployment would require validation across multiple users, environmental conditions, and more complex gesture vocabularies. These results demonstrate how convex optimization can enable efficient on-device machine learning for textile interfaces.
- Open Access
Resource-Efficient Gesture Recognition through Convexified Attention
Wearable e-textile interfaces require gesture recognition capabilities but face severe constraints in power consumption, computational capacity, and form factor that make traditional deep learning impractical. While lightweight architectures like MobileNet improve efficiency, they still demand thousands of parameters, limiting deployment on textile-integrated platforms. We introduce a convexified attention mechanism for wearable applications that dynamically weights features while preserving convexity through nonexpansive simplex projection and convex loss functions. Unlike conventional attention mechanisms using non-convex softmax operations, our approach employs Euclidean projection onto the probability simplex combined with multi-class hinge loss, ensuring global convergence guarantees. Implemented on a textile-based capacitive sensor with four connection points, our approach achieves 100.00\% accuracy on tap gestures and 100.00\% on swipe gestures -- consistent across 10-fold cross-validation and held-out test evaluation -- while requiring only 120--360 parameters, a 97\% reduction compared to conventional approaches. With sub-millisecond inference times (290--296$μ$s) and minimal storage requirements ($<$7KB), our method enables gesture interfaces directly within e-textiles without external processing. Our evaluation, conducted in controlled laboratory conditions with a single-user dataset, demonstrates feasibility for basic gesture interactions. Real-world deployment would require validation across multiple users, environmental conditions, and more complex gesture vocabularies. These results demonstrate how convex optimization can enable efficient on-device machine learning for textile interfaces.
Open AccessResource-Efficient Gesture Recognition through Convexified Attention
Wearable e-textile interfaces require gesture recognition capabilities but face severe constraints in power consumption, computational capacity, and form factor that make traditional deep learning impractical. While lightweight architectures like MobileNet improve efficiency, they still demand thousands of parameters, limiting deployment on textile-integrated platforms. We introduce a convexified attention mechanism for wearable applications that dynamically weights features while preserving convexity through nonexpansive simplex projection and convex loss functions. Unlike conventional attention mechanisms using non-convex softmax operations, our approach employs Euclidean projection onto the probability simplex combined with multi-class hinge loss, ensuring global convergence guarantees. Implemented on a textile-based capacitive sensor with four connection points, our approach achieves 100.00\% accuracy on tap gestures and 100.00\% on swipe gestures -- consistent across 10-fold cross-validation and held-out test evaluation -- while requiring only 120--360 parameters, a 97\% reduction compared to conventional approaches. With sub-millisecond inference times (290--296$μ$s) and minimal storage requirements ($<$7KB), our method enables gesture interfaces directly within e-textiles without external processing. Our evaluation, conducted in controlled laboratory conditions with a single-user dataset, demonstrates feasibility for basic gesture interactions. Real-world deployment would require validation across multiple users, environmental conditions, and more complex gesture vocabularies. These results demonstrate how convex optimization can enable efficient on-device machine learning for textile interfaces.
- Open Access
- 9 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Minimalist Neural Networks for Gesture Recognition on Wearable Capacitive Touch Textiles With Comparative User Study
- 9 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessMinimalist Neural Networks for Gesture Recognition on Wearable Capacitive Touch Textiles With Comparative User Study
- 1 September 2025
- Published by Springer Nature
- p. 46-59
A Minimal Neural Network for Reproducible Gesture Recognition on Knitted Capacitive Touch Sensors
- 1 September 2025
- Published by Springer Nature
- p. 46-59
A Minimal Neural Network for Reproducible Gesture Recognition on Knitted Capacitive Touch Sensors
- 10 May 2025
- Published by Springer Nature
- p. 32-46
Efficient Combinatorial Alignment for Improved Graph Generation Using GraphVAEs
- 10 May 2025
- Published by Springer Nature
- p. 32-46
Efficient Combinatorial Alignment for Improved Graph Generation Using GraphVAEs
- Open Access
Subnet Communicability: Diffusive Communication Across the Brain Through a Backbone Subnetwork
One of the fundamental challenges in modern neuroscience is understanding the interplay between the brain’s functional activity and its underlying structural pathways. To address this question, we propose a novel communication pattern called subnet communicability , which models diffusive communication between pairs of regions through a small, intermediary subnetwork of brain regions as opposed to spreading messages through the entire network. We demonstrate that subnet communicability strengthens coupling between the structural and functional connectomes better than previous models, including communicability. Over two large datasets, we show that the optimal subnetwork is consistent across the population. Subnet communicability provides new insights into structure-function coupling in the brain and offers a balance between redundancy in message passing and economy of brain wiring.
Open AccessSubnet Communicability: Diffusive Communication Across the Brain Through a Backbone Subnetwork
One of the fundamental challenges in modern neuroscience is understanding the interplay between the brain’s functional activity and its underlying structural pathways. To address this question, we propose a novel communication pattern called subnet communicability , which models diffusive communication between pairs of regions through a small, intermediary subnetwork of brain regions as opposed to spreading messages through the entire network. We demonstrate that subnet communicability strengthens coupling between the structural and functional connectomes better than previous models, including communicability. Over two large datasets, we show that the optimal subnetwork is consistent across the population. Subnet communicability provides new insights into structure-function coupling in the brain and offers a balance between redundancy in message passing and economy of brain wiring.
- Open Access
Choice of Parcellation Atlas Might Not be Too Critical for Connectomic Analysis
Connectomics has been a rapidly growing discipline in neuroimaging and neuroscience that evolved our understanding of the brain. Connectomics involves representing the brain as a network of regions, where the parcellation of the brain into regions using a template atlas is an integral part of the analysis. Over developmental and young adult cohorts of healthy individuals, we investigated how choosing parcellation atlases at certain resolutions affect sex classification and age prediction tasks performed using deep learning on structural connectomes. Datasets were processed on a total of 35 parcellations, where the only significant difference was observed for age prediction on the developmental cohort with a slight improvement on higher resolutions. This indicates that choice of parcellation scheme is generally not critical for deep learningbased age prediction and sex classification. Therefore, results between studies using different parcellation schemes could be comparable and repeating analyses on multiple atlases might be unnecessary.
Open AccessChoice of Parcellation Atlas Might Not be Too Critical for Connectomic Analysis
Connectomics has been a rapidly growing discipline in neuroimaging and neuroscience that evolved our understanding of the brain. Connectomics involves representing the brain as a network of regions, where the parcellation of the brain into regions using a template atlas is an integral part of the analysis. Over developmental and young adult cohorts of healthy individuals, we investigated how choosing parcellation atlases at certain resolutions affect sex classification and age prediction tasks performed using deep learning on structural connectomes. Datasets were processed on a total of 35 parcellations, where the only significant difference was observed for age prediction on the developmental cohort with a slight improvement on higher resolutions. This indicates that choice of parcellation scheme is generally not critical for deep learningbased age prediction and sex classification. Therefore, results between studies using different parcellation schemes could be comparable and repeating analyses on multiple atlases might be unnecessary.
- Open Access
Analysis of structural network topology alterations in autism and its relationship with autism severity
Autism Spectrum Disorder (ASD) is a public health concern ranging along a continuum of severity. The neurobiological motives behind ASD have been widely explored with reports about aberrant brain anatomy and functional connectivity. However, research on the underlying structural connectivity alterations is limited. We propose the application of a novel connectomic measure called Network Normality Score (NNS) to identify brain abnormalities and quantity topological dissimilarities in individuals with ASD. We show that the network topology of structural connectivity is altered in ASD brains relative to healthy controls at the global and system levels. We demonstrate that structural connectivity differences are more pronounced in certain subnetworks. Finally, we quantify the association between network similarity and behavioral autism severity to show the efficacy of NNS as a neuroimaging measure.
Open AccessAnalysis of structural network topology alterations in autism and its relationship with autism severity
Autism Spectrum Disorder (ASD) is a public health concern ranging along a continuum of severity. The neurobiological motives behind ASD have been widely explored with reports about aberrant brain anatomy and functional connectivity. However, research on the underlying structural connectivity alterations is limited. We propose the application of a novel connectomic measure called Network Normality Score (NNS) to identify brain abnormalities and quantity topological dissimilarities in individuals with ASD. We show that the network topology of structural connectivity is altered in ASD brains relative to healthy controls at the global and system levels. We demonstrate that structural connectivity differences are more pronounced in certain subnetworks. Finally, we quantify the association between network similarity and behavioral autism severity to show the efficacy of NNS as a neuroimaging measure.
- Open Access
- 29 April 2022
- Published by Wiley in Human Brain Mapping
- Vol. 43 (13) , 3944-3957
Connectomic assessment of injury burden and longitudinal structural network alterations in moderate‐to‐severe traumatic brain injury
Traumatic brain injury (TBI) is a major public health problem. Caused by external mechanical forces, a major characteristic of TBI is the shearing of axons across the white matter, which causes structural connectivity disruptions between brain regions. This diffuse injury leads to cognitive deficits, frequently requiring rehabilitation. Heterogeneity is another characteristic of TBI as severity and cognitive sequelae of the disease have a wide variation across patients, posing a big challenge for treatment. Thus, measures assessing network-wide structural connectivity disruptions in TBI are necessary to quantify injury burden of individuals, which would help in achieving personalized treatment, patient monitoring, and rehabilitation planning. Despite TBI being a disconnectivity syndrome, connectomic assessment of structural disconnectivity has been relatively limited. In this study, we propose a novel connectomic measure that we call network normality score (NNS) to capture the integrity of structural connectivity in TBI patients by leveraging two major characteristics of the disease: diffuseness of axonal injury and heterogeneity of the disease. Over a longitudinal cohort of moderate-to-severe TBI patients, we demonstrate that structural network topology of patients is more heterogeneous and significantly different than that of healthy controls at 3 months postinjury, where dissimilarity further increases up to 12 months. We also show that NNS captures injury burden as quantified by posttraumatic amnesia and that alterations in the structural brain network is not related to cognitive recovery. Finally, we compare NNS to major graph theory measures used in TBI literature and demonstrate the superiority of NNS in characterizing the disease.
- 29 April 2022
- Published by Wiley in Human Brain Mapping
- Vol. 43 (13) , 3944-3957
Open AccessConnectomic assessment of injury burden and longitudinal structural network alterations in moderate‐to‐severe traumatic brain injury
Traumatic brain injury (TBI) is a major public health problem. Caused by external mechanical forces, a major characteristic of TBI is the shearing of axons across the white matter, which causes structural connectivity disruptions between brain regions. This diffuse injury leads to cognitive deficits, frequently requiring rehabilitation. Heterogeneity is another characteristic of TBI as severity and cognitive sequelae of the disease have a wide variation across patients, posing a big challenge for treatment. Thus, measures assessing network-wide structural connectivity disruptions in TBI are necessary to quantify injury burden of individuals, which would help in achieving personalized treatment, patient monitoring, and rehabilitation planning. Despite TBI being a disconnectivity syndrome, connectomic assessment of structural disconnectivity has been relatively limited. In this study, we propose a novel connectomic measure that we call network normality score (NNS) to capture the integrity of structural connectivity in TBI patients by leveraging two major characteristics of the disease: diffuseness of axonal injury and heterogeneity of the disease. Over a longitudinal cohort of moderate-to-severe TBI patients, we demonstrate that structural network topology of patients is more heterogeneous and significantly different than that of healthy controls at 3 months postinjury, where dissimilarity further increases up to 12 months. We also show that NNS captures injury burden as quantified by posttraumatic amnesia and that alterations in the structural brain network is not related to cognitive recovery. Finally, we compare NNS to major graph theory measures used in TBI literature and demonstrate the superiority of NNS in characterizing the disease.
- 16 July 2021
- Published by Drexel University Libraries
On the Applications of Metric Trees and Metric Labeling to Hard Combinatorial Optimization Problems
Matching of metric distributions is a fundamental problem in computer science having numerous real life applications including computer vision, pattern recognition, and natural language processing. The problem can be reduced to graph matching since arbitrary metrics can be represented using graphs. Exact graph matching is known to be computationally intractable which motivates inexact matching approaches. Although graphs structurally represent metric data without losing information, processing data over graphs is prone to entail performance problems. Specifically, running time of many graph algorithms depend on the number of edges and vertices of the input graph which increases quadratically by the number of nodes on the extreme case of complete graphs. Thus, it is desirable to obtain a sparse representation of the data while preserving the quality of information. A common technique to achieve this is through representing graphs by metric trees which recently became defacto metric structures for embedding problems. In this dissertation, we focus on problems involving data that can be represented by graphs. As a general theme, we concentrate on approximation of graphs by trees for improving the performance of certain algorithms using topological structure of trees. We also focus on the fundamental problem of inexact graph matching, efficient approximation algorithms for the problem and its applications. Specifically, we present an inexact graph matching problem referred to as multilayer matching, which utilizes the structure of hierarchical metric trees. We represent graphs as trees and achieve matching over the trees to improve the performance and accuracy of inexact matching. We also establish a relationship between the well known metric labeling problem and inexact graph matching. We propose efficient approximation algorithms for both multilayer matching and metric labeling using the primal-dual approximation scheme. We provide application of the proposed methods to image matching, pattern recognition, and question answering problems. Finally, we present a novel motion segmentation method utilizing metric trees which provides tracking of objects in a video sequence without a priori knowledge of number of objects in the scene.
- 16 July 2021
- Published by Drexel University Libraries
On the Applications of Metric Trees and Metric Labeling to Hard Combinatorial Optimization Problems
Matching of metric distributions is a fundamental problem in computer science having numerous real life applications including computer vision, pattern recognition, and natural language processing. The problem can be reduced to graph matching since arbitrary metrics can be represented using graphs. Exact graph matching is known to be computationally intractable which motivates inexact matching approaches. Although graphs structurally represent metric data without losing information, processing data over graphs is prone to entail performance problems. Specifically, running time of many graph algorithms depend on the number of edges and vertices of the input graph which increases quadratically by the number of nodes on the extreme case of complete graphs. Thus, it is desirable to obtain a sparse representation of the data while preserving the quality of information. A common technique to achieve this is through representing graphs by metric trees which recently became defacto metric structures for embedding problems. In this dissertation, we focus on problems involving data that can be represented by graphs. As a general theme, we concentrate on approximation of graphs by trees for improving the performance of certain algorithms using topological structure of trees. We also focus on the fundamental problem of inexact graph matching, efficient approximation algorithms for the problem and its applications. Specifically, we present an inexact graph matching problem referred to as multilayer matching, which utilizes the structure of hierarchical metric trees. We represent graphs as trees and achieve matching over the trees to improve the performance and accuracy of inexact matching. We also establish a relationship between the well known metric labeling problem and inexact graph matching. We propose efficient approximation algorithms for both multilayer matching and metric labeling using the primal-dual approximation scheme. We provide application of the proposed methods to image matching, pattern recognition, and question answering problems. Finally, we present a novel motion segmentation method utilizing metric trees which provides tracking of objects in a video sequence without a priori knowledge of number of objects in the scene.