A5060434920- Domain Adaptation and Few-Shot Learning
- AI in cancer detection
- Machine Learning in Healthcare
- Adversarial Robustness in Machine Learning
- Generative Adversarial Networks and Image Synthesis
- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Advanced Text Analysis Techniques
- Anomaly Detection Techniques and Applications
- Radiomics and Machine Learning in Medical Imaging
- Topic Modeling
- Medical Image Segmentation Techniques
- Bayesian Modeling and Causal Inference
- Natural Language Processing Techniques
- COVID-19 diagnosis using AI
- Computer Graphics and Visualization Techniques
- Brain Tumor Detection and Classification
- earthquake and tectonic studies
- Advanced Causal Inference Techniques
- Diatoms and Algae Research
- Digital Media Forensic Detection
- Frailty in Older Adults
- Transportation Planning and Optimization
- Explainable Artificial Intelligence (XAI)
- Transportation and Mobility Innovations
University of Edinburgh
2022-2024
University of Illinois Urbana-Champaign
2024
Recent advances in generative AI have brought incredible breakthroughs several areas, including medical imaging. These models tremendous potential not only to help safely share data via synthetic datasets but also perform an array of diverse applications, such as anomaly detection, image-to-image translation, denoising, and MRI reconstruction. However, due the complexity these models, their implementation reproducibility can be difficult. This hinder progress, act a use barrier, dissuade...
Abstract Although accurately classifying signals from earthquakes and explosions at local distance (<250 km) remains an important task for seismic network operations, the growing volume of available data presents a challenge analysts using traditional source discrimination techniques. In recent years, deep-learning models have proven effective discriminating between low-magnitude measured distances, but it is not clear how well these are capable generalizing across different...
Diatoms are microscopic organisms belonging to the algae kingdom. They adapt ecosystem and modify their shape texture depending on hundreds of variables. Hence, these micro-organism considered as most accurate indicator measure water quality. Commonly, recognition class diatoms in a image has always been done by expert biologists knowledgeable about morphometric characteristics organisms. This work proposes new automatic diatom genus system from images using state-of-the-art deep CNNs. In...
We present a comprehensive evaluation of Parameter-Efficient Fine-Tuning (PEFT) techniques for diverse medical image analysis tasks. PEFT is increasingly exploited as valuable approach knowledge transfer from pre-trained models in natural language processing, vision, speech, and cross-modal tasks, such vision-language text-to-image generation. However, its application remains relatively unexplored. As foundation are the domain, it crucial to investigate comparatively assess various...
Deep learning models often need sufficient supervision (i.e. labelled data) in order to be trained effectively. By contrast, humans can swiftly learn identify important anatomy medical images like MRI and CT scans, with minimal guidance. This recognition capability easily generalises new from different facilities tasks settings. rapid generalisable ability is largely due the compositional structure of image patterns human brain, which are not well represented current models. In this paper,...
Age has important implications for health, and understanding how age manifests in the human body is first step a potential intervention. This becomes especially cardiac since main risk factor development of cardiovascular disease. Data-driven modeling progression been conducted successfully diverse applications such as face or brain aging. While longitudinal data preferred option training deep learning models, collecting dataset usually very costly, medical imaging. In this work, conditional...
Localizing the exact pathological regions in a given medical scan is an important imaging problem that traditionally requires large amount of bounding box ground truth annotations to be accurately solved. However, there exist alternative, potentially weaker, forms supervision, such as accompanying free-text reports, which are readily available. The task performing localization with textual guidance commonly referred phrase grounding. In this work, we use publicly available Foundation Model,...
Counterfactual image generation is pivotal for understanding the causal relations of variables, with applications in interpretability and unbiased synthetic data. However, evaluating a long-standing challenge itself. The need to evaluate counterfactual compounds on this challenge, precisely because counterfactuals, by definition, are hypothetical scenarios without observable ground truths. In paper, we present novel comprehensive framework aimed at benchmarking methods. We incorporate...
Localizing the exact pathological regions in a given medical scan is an important imaging problem that requires large amount of bounding box ground truth annotations to be accurately solved. However, there exist alternative, potentially weaker, forms supervision, such as accompanying free-text reports, which are readily available. The task performing localization with textual guidance commonly referred phrase grounding. In this work, we use publicly available Foundation Model, namely Latent...
Deep learning models often need sufficient supervision (i.e. labelled data) in order to be trained effectively. By contrast, humans can swiftly learn identify important anatomy medical images like MRI and CT scans, with minimal guidance. This recognition capability easily generalises new from different facilities tasks settings. rapid generalisable ability is largely due the compositional structure of image patterns human brain, which are not well represented current models. In this paper,...
Unsupervised representation learning with variational inference relies heavily on independence assumptions over latent variables. Causal (CRL), however, argues that factors of variation in a dataset are, fact, causally related. Allowing variables to be correlated, as consequence causal relationships, is more realistic and generalisable. So far, provably identifiable methods rely on: auxiliary information, weak labels, interventional or even counterfactual data. Inspired by discovery...
Knowledge distillation enables fast and effective transfer of features learned from a bigger model to smaller one. However, objectives are susceptible sub-population shifts, common scenario in medical imaging analysis which refers groups/domains data that underrepresented the training set. For instance, models on health acquired multiple scanners or hospitals can yield subpar performance for minority groups. In this paper, inspired by distributionally robust optimization (DRO) techniques, we...