PhD Research Seminar: AutoML methods and tools, Deep Learning and biological images

First talk: Methods and Software Tools of Automated Machine Learning   
Speaker: Andrey Kuzminykh, second-year PhD student, Faculty of Computer Science

Automating the end-to-end process of machine learning paves the way for efficient construction of models that often outperform models built manually. We will review current methods and software tools that automate the construction of machine learning models in classification and ranking problems. We will discuss such software tools as H20 AutoML, Google AutoML, and AutoKeras, among others.
We will also highlight AutoML-related issues that remain to be solved.


Second talk: Deep Learning and Biological Images     
Speaker: Nikita Moshkov, third-year PhD student, Faculty of Computer Science

The talk is devoted to the analysis of biological images and techniques used to solve this task such as dataset augmentation for image segmentation, representation learning, partial annotations, and active learning. We apply image-to-image translation (GANs) for training set augmentation and test-time augmentation to improve segmentation. Deep learning and representation learning are used  for extracting phenotypic information from microscopy images of cells. In order to be able to use less annotated data, we are working on an active learning strategy that would allow us to learn a model for image segmentation of quality comparable with that of a model trained on full data.