Research Interests

I am primarily interested in the application of deep learning techniques to classification tasks in medical image analysis, with a particular emphasis on analyzing skin lesion images. My research interests revolve around advancing the field by improving the lifelong learning, generalization capabilities and fairness of deep neural networks.

Improving Lifelong Learning of Deep Networks

In real-world applications, particularly in medical imaging, models must learn from an ever-evolving stream of data. The challenge lies in ensuring that these models retain previously acquired knowledge while seamlessly adapting to new information. Traditional learning methods often suffer from catastrophic forgetting, where the model loses past knowledge when updated with new data. This issue can have severe consequences in critical applications like healthcare, where outdated or incorrect knowledge could impact patient outcomes. My research focuses on developing continual learning strategies that maintain high performance and ensure models remain adaptable, reliable, and efficient over time.

Improving Generalization of Deep Networks on out-of-distribution (OOD) Data

Models trained in controlled environments often struggle when deployed in real-world settings where data distributions differ significantly from the training set. This domain shift is particularly problematic in medical imaging, where variations in equipment, patient demographics, and imaging conditions can lead to substantial performance drops. The ability to generalize across domains without retraining on new data is essential for ensuring consistent and accurate diagnoses. My research tackles these challenges by exploring methods that learn robust, domain-invariant features, paving the way for models that are resilient and applicable in diverse healthcare environments.

Fairness and Bias Mitigation

Machine learning models are prone to biases that can result in unequal performance across demographic groups, which is especially concerning in fields like healthcare, where disparities can directly affect patient care. Biases can be introduced from imbalanced datasets, label inaccuracies, or model architectures that disproportionately favor certain groups. Addressing these issues is crucial to ensure that models make fair and equitable decisions. My research focuses on developing strategies to identify and mitigate these biases, using advanced debiasing techniques that promote fairness and improve outcomes for underrepresented or vulnerable populations, ultimately making AI systems more trustworthy and inclusive.