PyData Trójmiasto #21

We are happy to announce our first meetup in 2023!

This time we are welcoming everyone to join for two great talks from Jan Glinko and Franciszek Górski.

Talk#1

Meta-learning for fast Neural Network fine-tuning

(PyTorch, Higher, custom data loader, training script preparation, batch normalization issues)

The speed of learning, whether it is pattern recognition or the acquisition of new skills, is a characteristic of human intelligence. Meta-learning is a field of machine learning that attempts to replicate how humans learn new tasks. This type of fast and flexible learning, benefiting from previous experience, differs from the standard approach to training deep neural networks. In summary, our goal is no longer a model that generalizes well, but becomes one that adapts well.

Key points:

• When is it worth using meta-learning?
• Types of meta-learning
• Using Model-Agnostic Meta-Learning to solve a regression problem

Talk#2

Machine learning algorithms for cancer classification of liquid biopsy data

Early cancer detecetion is a key step in a successfull treatment. Recent development of a liquid biopsy method, which is a simple noninvasive method and discovering a Tumor Educated Platelets mechanism boosted the studies on using machine learning algorithms for cancer classification task.

In my presentation I would speak about my works during collaboration with Center for Biostatistics and Bioinformatics from Gumed with such liquid biopsy data, designed for cancer classification. During these works we fit and validated various machine learning algorithms like neural networks, kNNs or gradient boosted trees. We trained them for binary and multiclass classification problems. I would speak about our results and various attemps for processing of our data like dimensionality reduction, attemps to creating a graphs or using autoencoders for classification task.