Spring 2023

DL Seminar

@ University of Wrocław

We like deep learning and we like to listen about it

Monday 16:15

Room 119

deeplearning@cs.uni.wroc.pl

The seminar is aimed at students interested in understanding the latest deep learning techniques. The seminar primarily consists of students' presentations, but in addition to that, we plan to have invited talks given by experts in the field.

Supervised by Marek Adamczyk

Invited Speakers

Łukasz Kaiser

OpenAI

Petar Veličković

DeepMind

Konstantinos Vougioukas

Meta Research

Sponsorship

Schedule

13.03 Karol Kuczmarz

Transformers 101

20.03 Łukasz Kaiser

OpenAI invited talk

27.03 Petar Veličković

DeepMind invited talk

03.04 Piotr Słowik

Recurrent Neural Networks and LSTMs from scratch

17.04 Jacek Kowalski

LSTMs with attention

24.04 Jakub Kuciński

An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

8.05 Oskar Bujacz

Chess Transformers

15.05 Mateusz Basiak

Computational Power of Recurrent Neural Nets and Transformers

22.05 Konstantinos Vougioukas

Meta Research invited talk

5.06 Mikołaj Słupiński

Hidden Markov Models

12.06 Maciej Malicki

Tutorial on Graph Neural Network

19.06 Artur Suchwałko

Quantup sponsored talk

Machine Learning for Biotech Startups. Case Study: Bioavlee S.A. and Bacterial Strain Identification

Abstract: Bioavlee offers a device for automatic identification of bacteria with high accuracy, low cost, and short identification time. The approach is based on a patented method of analyzing spectra from laser diffraction. Bacterial strains are identified using computer vision (CV) and machine learning (ML) methods. QuantUp assisted Bioavlee in building an identification "engine" for bacteria.

During the presentation, I will share our experiences in supporting product development for Biotech companies throughout the entire product development cycle: what proved to be crucial, what risks emerged, and how we dealt with them.

The development of the breakthrough device required not only a solid understanding of mathematical methods but also knowledge of laboratory work realities and domain expertise.

The solution built by the QuantUp team ensures high-quality and fast identification of bacteria.

The insights from the project and presentation are applicable to the construction of various types of products that heavily leverage mathematical methods.