2025
Subject Invariant Contrastive Learning For Human Activity Recognition
Yavuz Yarici, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Workshop on Machine Learning for Signals Processing 2025
In this work, we introduce additional regularization based on subject identity to overcome inherent biases within action recognition pipelines.
Subject Invariant Contrastive Learning For Human Activity Recognition
Yavuz Yarici, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Workshop on Machine Learning for Signals Processing 2025
In this work, we introduce additional regularization based on subject identity to overcome inherent biases within action recognition pipelines.
AdaDim: Dimensionality Adaptation for SSL Representational Dynamics
Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
Under review.
In this work, we study SSL training dynamics in terms of both mutual information between projection and representation spaces as well as in terms of overall dimensionality.
AdaDim: Dimensionality Adaptation for SSL Representational Dynamics
Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
Under review.
In this work, we study SSL training dynamics in terms of both mutual information between projection and representation spaces as well as in terms of overall dimensionality.
HEX: Hierarchical Emergence Exploitation in Self-Supervised Algorithms
Kiran Kokilepersaud, Seulgi Kim, Mohit Prabhushankar, Ghassan AlRegib
Winter Applications on Computer Vision (WACV) 2025 ORAL
In this work, we introduce an SSL regularization strategy based on localized hierarchical relationships. We show that this method can be added on to a wide variety of SSL approaches to improve performance.
HEX: Hierarchical Emergence Exploitation in Self-Supervised Algorithms
Kiran Kokilepersaud, Seulgi Kim, Mohit Prabhushankar, Ghassan AlRegib
Winter Applications on Computer Vision (WACV) 2025 ORAL
In this work, we introduce an SSL regularization strategy based on localized hierarchical relationships. We show that this method can be added on to a wide variety of SSL approaches to improve performance.
2024
Taxes Are All You Need: Integration of Taxonomical Hierarchy Relationships into the Contrastive Loss
Kiran Kokilepersaud, Yavuz Yarici, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Conference on Image Processing (ICIP) 2024
In this work, we present a supervised representation learning strategy that integrates taxonomy relationships into the contrastive loss.
Taxes Are All You Need: Integration of Taxonomical Hierarchy Relationships into the Contrastive Loss
Kiran Kokilepersaud, Yavuz Yarici, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Conference on Image Processing (ICIP) 2024
In this work, we present a supervised representation learning strategy that integrates taxonomy relationships into the contrastive loss.
Explaining Representation Learning With Perceptual Components
Yavuz Yarici, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Conference on Image Processing (ICIP) 2024
In this work, we present an explainability setup for SSL algorithms based on perceptual factors like texture and color.
Explaining Representation Learning With Perceptual Components
Yavuz Yarici, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
IEEE International Conference on Image Processing (ICIP) 2024
In this work, we present an explainability setup for SSL algorithms based on perceptual factors like texture and color.
CRACKS: Crowdsourcing Resources for Analysis and Categorization of Key Subsurface faults
Mohit Prabhushankar, Kiran Kokilepersaud*, Jorge Quesada*, Yavuz Yarici*, Chen Zhou, Mohammad Alotaibi, Ghassan AlRegib, Ahmad Mustafa, Yusufjon Kumakov (* equal contribution)
Under review.
In this work, we develop a dataset with annotations of seismic faults across different levels of annotator expertise.
CRACKS: Crowdsourcing Resources for Analysis and Categorization of Key Subsurface faults
Mohit Prabhushankar, Kiran Kokilepersaud*, Jorge Quesada*, Yavuz Yarici*, Chen Zhou, Mohammad Alotaibi, Ghassan AlRegib, Ahmad Mustafa, Yusufjon Kumakov (* equal contribution)
Under review.
In this work, we develop a dataset with annotations of seismic faults across different levels of annotator expertise.
Ophthalmic Biomarker Detection: Highlights from the IEEE Video and Image Processing Cup 2023 Student Competition
Ghassan AlRegib, Mohit Prabhushankar, Kiran Kokilepersaud, Prithwijit Chowdhury, Zoe Fowler, Stephanie Trejo Corona, Lucas Thomaz, Angshul Majumdar
IEEE Signals Processing Magazine 2024
In this work, we present analyses on the methods performed by participants in the 2023 VIP Cup Competition organized by our lab.
Ophthalmic Biomarker Detection: Highlights from the IEEE Video and Image Processing Cup 2023 Student Competition
Ghassan AlRegib, Mohit Prabhushankar, Kiran Kokilepersaud, Prithwijit Chowdhury, Zoe Fowler, Stephanie Trejo Corona, Lucas Thomaz, Angshul Majumdar
IEEE Signals Processing Magazine 2024
In this work, we present analyses on the methods performed by participants in the 2023 VIP Cup Competition organized by our lab.
2023
FOCAL: A Cost-Aware Video Dataset for Active Learning
Kiran Kokilepersaud*, Yash-Yee Logan*, Ryan Benkert, Chen Zhou, Mohit Prabhushankar, Ghassan AlRegib, Enrique Corona, Kunjan Singh, Armin Parchami (* equal contribution)
IEEE International Conference on Big Data 2023
In this work, we develop a dataset that measures the time it takes an annotator to measure video sequences. Through this information we are able to benchmark active learning algorithms based on their true associated cost savings.
FOCAL: A Cost-Aware Video Dataset for Active Learning
Kiran Kokilepersaud*, Yash-Yee Logan*, Ryan Benkert, Chen Zhou, Mohit Prabhushankar, Ghassan AlRegib, Enrique Corona, Kunjan Singh, Armin Parchami (* equal contribution)
IEEE International Conference on Big Data 2023
In this work, we develop a dataset that measures the time it takes an annotator to measure video sequences. Through this information we are able to benchmark active learning algorithms based on their true associated cost savings.
Clinical Trial Active Learning
Zoe Fowler, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
ACM International Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB) 2023
In this work, we develop an active learning framework that considers the collection process of real-world clinical trial routines.
Clinical Trial Active Learning
Zoe Fowler, Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib
ACM International Conference on Bioinformatics, Computational Biology and Health Informatics (ACM-BCB) 2023
In this work, we develop an active learning framework that considers the collection process of real-world clinical trial routines.
Clinically Labeled Contrastive Learning for OCT Biomarker Classification
Kiran Kokilepersaud, Stephanie Trejo Corona, Mohit Prabhushankar, Ghassan AlRegib, Charles Wykoff
IEEE Journal of Bio Health Informatics 2023
In this work, we demonstrate that clinical information can be used in SSL algorithms to inform the related task of biomarker detection. This shows that one modality can act as guiding information for a contrastive learning algorithm.
Clinically Labeled Contrastive Learning for OCT Biomarker Classification
Kiran Kokilepersaud, Stephanie Trejo Corona, Mohit Prabhushankar, Ghassan AlRegib, Charles Wykoff
IEEE Journal of Bio Health Informatics 2023
In this work, we demonstrate that clinical information can be used in SSL algorithms to inform the related task of biomarker detection. This shows that one modality can act as guiding information for a contrastive learning algorithm.
Exploiting the Distortion-Semantic Interaction in Fisheye Data
Kiran Kokilepersaud, Yavuz Yarici, Mohit Prabhushankar, Ghassan AlRegib, Armin Parchami
IEEE Open Journal of Signals Processing (OJSP) 2023
In this work, we develop a representation learning strategy to overcome distortion in fisheye camera images. We show how this method improves object detection performance in distorted regions.
Exploiting the Distortion-Semantic Interaction in Fisheye Data
Kiran Kokilepersaud, Yavuz Yarici, Mohit Prabhushankar, Ghassan AlRegib, Armin Parchami
IEEE Open Journal of Signals Processing (OJSP) 2023
In this work, we develop a representation learning strategy to overcome distortion in fisheye camera images. We show how this method improves object detection performance in distorted regions.
2022
OLIVES Dataset: Ophthalmic Labels for Investigating Visual Eye Semantics
Mohit Prabhushankar, Kiran Kokilepersaud*, Yash-Yee Logan*, Stephanie Trejo Corona, Ghassan AlRegib, Charles Wykoff (* equal contribution)
Neural Information Processing Systems (NeurIPS) 2022
We collaborated with a medical institution to develop one of the first multimodal time series clinical trial datasets. We demonstrate its utility to the machine learning community across tasks such as multi-modal learning, time series prediction, and self supervised learning.
OLIVES Dataset: Ophthalmic Labels for Investigating Visual Eye Semantics
Mohit Prabhushankar, Kiran Kokilepersaud*, Yash-Yee Logan*, Stephanie Trejo Corona, Ghassan AlRegib, Charles Wykoff (* equal contribution)
Neural Information Processing Systems (NeurIPS) 2022
We collaborated with a medical institution to develop one of the first multimodal time series clinical trial datasets. We demonstrate its utility to the machine learning community across tasks such as multi-modal learning, time series prediction, and self supervised learning.
Gradient-Based Severity Labeling for OCT Biomarker Classification
Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib, Stephanie Trejo Corona, Charles Wykoff
IEEE International Conference on Image Processing (ICIP) 2022
In this work, we measure gradient responses from a network trained on healthy data to rank images with a pseudo-severity score. This score is then used to group images for a contrastive learning objective.
Gradient-Based Severity Labeling for OCT Biomarker Classification
Kiran Kokilepersaud, Mohit Prabhushankar, Ghassan AlRegib, Stephanie Trejo Corona, Charles Wykoff
IEEE International Conference on Image Processing (ICIP) 2022
In this work, we measure gradient responses from a network trained on healthy data to rank images with a pseudo-severity score. This score is then used to group images for a contrastive learning objective.
Volumetric Supervised Contrastive Learning for Seismic Semantic Segmentation
Kiran Kokilepersaud, Ghassan AlRegib
International Meeting for Applied Geoscience and Energy 2022
In this work, we propose a self supervised methodology that considers volumetric positions during the learning of seismic representations.
Volumetric Supervised Contrastive Learning for Seismic Semantic Segmentation
Kiran Kokilepersaud, Ghassan AlRegib
International Meeting for Applied Geoscience and Energy 2022
In this work, we propose a self supervised methodology that considers volumetric positions during the learning of seismic representations.