We propose Diverse Ensemble Entropy Minimization (DEnEM), questioning the effectiveness of current TTA methods on ultrasound data. We show that these methods, although outperforming baselines, are suboptimal due to relying on neural networks output probabilities, which could be uncalibrated, or relying on data augmentation, which is not straightforward to define on ultrasound data. Our results show a significant improvement of 5% to 7% in AUROC over the existing methods and 3% to 5% over TTA methods, demonstrating the advantage of DEnEM in addressing distribution shift.

This study presents LensePro, a unified method that not only excels in label efficiency but also demonstrates robustness against label noise and OOD data. LensePro comprises two key stages: first, self-supervised learning to extract high-quality feature representations from abundant unlabeled TRUS data, and second, label noise-tolerant prototype-based learning to classify the extracted features.

Prostate ultrasound data is highly heterogeneous due to variability in tissue types and the cancer appearance in images, making the development of robust models that can generalize to new patients and clinical settings, challenging. A viable model requires not only accurate cancer predictions but also effective uncertainty calibration, i.e. understanding when the predictions are likely to be correct. Previous studies on uncertainty calibration for PCa detection often rely on limited or single center datasets which may not adequately capture real-world data variability. We aim to establish realistic benchmarks for PCa detection using a multi-centre, diverse dataset to better assess the clinical utility of this tool.

We present a detailed study of several Image Transformer architectures for both ROI-scale and multi-scale classification, and a comparison of the performance of CNNs and Transformers for ultrasound-based prostate cancer classification. We evaluate 3 vision transformers on ROI-scale cancer classification then use the strongest model to tune a multi-scale classifier using multi-objective learning. We compare our results in both settings to a baseline convolutional architecture typically used in computer vision tasks. We evaluate all our models using nested k-fold cross-validation. Our core-wise multi-objective model achieves a 77.9% AUROC, a sensitivity of 75.9%, and a specificity of 66.3%, a considerable improvement over baseline ROI-scale classification.

In this work, we explore a novel approach based on probabilistic planning and dynamic graph analysis to model the spread of COVID-19 in indoor spaces. We also endow the planner with means to control the spread of the disease trough non-pharmaceutical interventions (NPIs) such as mandating masks, vaccines, capacity limits, and social distancing. We demonstrate that the use of automated probabilistic planning is effective in predicting the amount of infections that are likely to occur in shared spaces, and that they are capable of designing competent interventions to limit the spread of the disease.

We explore RNN and CodeBERT deep learning models that highlight errors in student submissions to Python coding problems. We find that a standard automatic metric like AUC does not correspond well to human evaluation, and that the scale of the benefits of transfer learning and pre-training are only seen when using human evaluation.