Junayed Naushad

Junayed Naushad

Doctoral Student

Student, St Anne's College

Wolfson Building, Parks Road, Oxford OX1 3QD

Interests

My research interests include:

• Medical image analysis
• Uncertainty estimation
• Representation learning
• Multimodal learning

My DPhil focuses on developing trustworthy machine learning algorithms for medical image analysis. I am working on methods that provide reliable uncertainty estimates for deep learning models so that challenging failure cases or out of distribution inputs can be detected when models are deployed in safety-critical settings (Super-TrustScore). I am also interested in multimodal learning and how different strategies for fusing modalities affect the learned representations, model uncertainty, and bias.

Representation Learning

Comparing the learned embedding spaces of the popular computer vision dataset CIFAR-10 with a skin lesion dataset illustrates the "entangled" nature of medical image data and highlights the challenges in performing uncertainty estimation in this space. 

Personal Pages

GitHub

Google Scholar

Selected Publications

View AllManage publications

• Super−TrustScore: Reliable Failure Detection for Automated Skin Lesion Diagnosis

  Junayed Naushad and Irina Voiculescu

  In IEEE International Symposium on Biomedical Imaging (ISBI). May, 2024.

  Details about Super−TrustScore: Reliable Failure Detection for Automated Skin Lesion Diagnosis | BibTeX data for Super−TrustScore: Reliable Failure Detection for Automated Skin Lesion Diagnosis | Link to Super−TrustScore: Reliable Failure Detection for Automated Skin Lesion Diagnosis

Supervisor