Publications

Self-Supervised Learning of Speech Representation via Redundancy Reduction

Published in Gesellschaft für Informatik e.V.. pp. 11-19. Doctoral Consortium at KI 2023. Berlin. 45195, 2023

This work aims to investigate a novel self-supervised learning (SSL) method for speech representation that leverages redundancy reduction techniques to learn robust representations capturing speaker characteristics. Our proposed approach builds upon the Barlow Twins framework, introduced in computer vision, and we adapt it to speech processing. The primary objective is to assess the quality of the learned representations through comprehensive evaluations of various downstream tasks, including speaker identification, gender recognition, emotion recognition, and more. By exploiting the statistical relationships between different views of the same speech input, the proposed method encourages the model to capture speaker-specific information while attenuating the impact of irrelevant variations. This enables extraction of features invariant to non-speaker-related factors, such as language content or background noise. Read more

Recommended citation: Brima, Yusuf (2023): Self-Supervised Learning of Speech Representation via Redundancy Reduction. DC@KI2023: Proceedings of Doctoral Consortium at KI 2023. DOI: 10.18420/ki2023-dc-02. Gesellschaft für Informatik e.V.. pp. 11-19. Doctoral Consortium at KI 2023. Berlin. 45195 https://doi.org/10.18420/ki2023-dc-02

Understanding Self-Supervised Learning of Speech Representation via Invariance and Redundancy Reduction

Published in arXiv preprint arXiv.2309.03619, 2023

In this study, we investigate the impact of different formulations of the Barlow Twins (BT) objective on downstream task performance for speech data, proposing Modified Barlow Twins (MBT) with normalized latents to enforce scale-invariance, and our results demonstrate that MBT improves representation generalization over original BT, particularly when fine-tuning with limited target data, highlighting the importance of designing objectives that encourage invariant and transferable representations for self-supervised speech representations. Read more

Recommended citation: Brima, Yusuf, et al. "Understanding Self-Supervised Learning of Speech Representation via Invariance and Redundancy Reduction." arXiv preprint arXiv:2309.03619 (2023). https://doi.org/10.48550/arXiv.2309.03619

Visual Interpretable and Explainable Deep Learning Models for Brain Tumor MRI and COVID-19 Chest X-ray Images

Published in arXiv preprint arXiv.2208.00953, 2023

We evaluated attribution methods to illuminate how deep neural networks analyze medical images, using adaptive path-based gradient integration to attribute predictions from brain tumor MRI and COVID-19 chest X-ray datasets, highlighting possible biomarkers, exposing model biases, and offering insights into the links between input and prediction, demonstrating the method’s ability to elucidate model reasoning and improve deep learning transparency for domain experts by revealing the rationale behind predictions. Read more

Recommended citation: Brima, Yusuf and Marcellin Atemkeng. “Visual Interpretable and Explainable Deep Learning Models for Brain Tumor MRI and COVID-19 Chest X-ray Images.” (2022). https://doi.org/10.48550/arXiv.2208.00953

A Mathematical Framework for Understanding Recognition Systems

Published in arXiv preprint biorxiv:2023.06.08.544240, 2023

We propose a theoretical framework for biological recognition, integrating category theory, information theory, dynamical systems modeling, and optimization theory. This framework defines individual recognition (IR) and class-level recognition (CR) as functors between stimuli and responses. We identify five conditions for robust IR systems, termed “signature systems,” and model them as attractor states, offering insights into communication and language evolution. Read more

Recommended citation: Brima, Yusuf. "A Mathematical Framework for Understanding Recognition Systems" biorxiv preprint biorxiv:2023.06.08.544240 (2023). https://doi.org/10.1101/2023.06.08.544240

Transfer Learning for the Detection and Diagnosis of Types of Pneumonia including Pneumonia Induced by COVID-19 from Chest X-ray Images

Published in MDPI Special Issue on Machine Learning Applications for COVID-19 and Its Complications: Screening, Diagnosis, Treatment, and Prognosis, 2021

Accurate early diagnosis of COVID-19 viral pneumonia, primarily in asymptomatic people, is essential to reduce the spread of the disease, the burden on healthcare capacity, and the overall death rate. It is essential to design affordable and accessible solutions to distinguish pneumonia caused by COVID-19 from other types of pneumonia. In this work, we propose a reliable approach based on deep transfer learning that requires few computations and converges faster. Experimental results demonstrate that our proposed framework for transfer learning is a potential and effective approach to detect and diagnose types of pneumonia from chest X-ray images with a test accuracy of 94.0%. Read more

Recommended citation: Yusuf Brima. (2021). "Deep Transfer Learning for the Detection and Diagnosis of Types of Pneumonia including Pneumonia Induced by COVID-19 from Chest X-ray Images." MDPI Special Issue on Machine Learning Applications for COVID-19 and Its Complications: Screening, Diagnosis, Treatment, and Prognosis. 1(1). https://doi.org/10.3390/diagnostics11081480

Deep Transfer Learning for Brain Magnetic Resonance Image Multi-class Classification

Published in IEEE/ACM Transactions on Computational Biology and Bioinformatics, 2021

In our research, we utilized Magnetic Resonance Imaging (MRI) as a principal diagnostic tool in radiology to detect soft tissue abnormalities, particularly in the brain. To address the laborious process of manually interpreting a large number of MRI scans, we developed a framework using Machine Learning methodologies. Our approach, based on Deep Transfer Learning and the Deep Residual Convolutional Neural Network (ResNet50) architecture, achieved remarkable classification accuracies of 86.40% on our dataset, 93.80% on the Harvard Whole Brain Atlas dataset, and an impressive 97.05% accuracy on the School of Biomedical Engineering dataset. These results demonstrate the effectiveness of our proposed transfer learning framework for multi-classification of brain tumors in MRI images, offering a potential and impactful method to improve medical diagnosis in radiology. Read more

Recommended citation: Yusuf Brima, Mossadek Hossain Kamal, Upama Kabir, and Tariqul Islam (2019). "Deep Transfer Learning for Brain Magnetic Resonance Image Multi-class Classification." IEEE/ACM Transactions on Computational Biology and Bioinformatics. 1(1). https://doi.org/10.48550/arXiv.2106.07333