Projects
Atlas Analytics Lab is currently addressing multiple projects in the field of deep learning and computational pathology
Our lab is currently working on the integration of vision-language models and efficient fine-tuning algorithms on histopathology datasets, seeking to integrate multimodal information for improved medical image analysis while minimizing the consumption of high computational resources.
Our lab is researching test-time training and meta-learning to enhance model adaptability and performance in dynamic environments, allowing for rapid adjustment to new data and tasks, improving robustness in real-world applications, and facilitating continual learning and optimization across diverse scenarios.
Atlas Analytics Lab explores the use of parameter-efficient fine-tuning techniques on foundation models that have been pretrained on extensive datasets. By implementing these methods in computational pathology, we aim to significantly enhance model performance on new, target datasets without the need to fine-tune the entire model. This strategy seeks to optimize resource efficiency while maintaining high accuracy in diagnostic applications.
Our research explores the application of Diffusion Generative Models (DGMs) for studying and synthesizing histopathology images. This work includes a comprehensive comparison of various diffusion generative methods, analyzing the characteristics they impart to the generated images. Additionally, we showcase the capability of DGMs to learn patch resolution of histopathology image patches, highlighting their utility in medical imaging applications.
We are currently advancing the field of optimization for deep neural networks by developing and evaluating a new adaptive second-order optimizer, AdaFisher, across a variety of tasks, including audio processing with SpeechBrain and image segmentation.
Our lab is advancing self-supervised learning techniques tailored for colorectal polyp screening to significantly enhance diagnostic accuracy. By strategically leveraging limited regions of interest annotations, our work aims to streamline the diagnostic workflow of pathologists, allowing for more efficient and precise identification of abnormalities.
Our research on novel architectures for computer vision encompasses pioneering advancements in the development of innovative models and algorithms. Specifically, we focus on integrating cutting-edge self-supervised learning techniques designed for computational pathology, with the overarching goal of enhancing diagnostic precision and efficiency in the analysis of medical images.