Interactive Syllabus
News
- The course is a [i]regular [re]search seminar, mentored in a "family" manner. In this term, the seminar [sub]topic is Tame the complexity. The enrolled student must give a presentation on an agreed-upon topic (of interest, or her thesis, or he will talk about a research paper or area) once during the term. Topics of presentations focus primarily (but not necessarily) on those related to machine learning, representation learning, and scientific visualization or about our subtopic (tackling complexity by LLM agents, proving P=NP, etc. ;-).
- There is a discussion group with official course information and a communication channel in addition to the
course outline below: watch both frequently!
Topics and Course Outline
Week 1– 19.9. canceled due to floods
- Readings: Motivating video: DEK's advice to young students.
Week 2 – Introduction, research strategy, evaluation, and course schedule planning
Join us at A502, Faculty of Informatics MU, on September 26th at 10 AM (CET) [or on Zoom, on-demand only].
- Why? Put readers in your place! Specifics of CS research and doctoral studies and their evaluation at FI MU: CS conference rankings
- How? to write
- What (and where)? It is important to "sell the ideas and work," pick the right topics and questions, research "big issues," and pick the proper publication forums (in CS and NLP). An h-index as a measure of impact. The danger of Tyranny of metric.
Week 3 – Taming the complexity in/of your projects
Join us at A502, Faculty of Informatics MU, on Oct 3rd at 9:50 AM (catering preparation) and 10 AM (Invitation of newcomers, questions on readings, and a summary of Week 2). To join via Zoom, ask for a password in advance.
We will present the research projects we are working on, in an elevator-pitch style. Presenting complex projects under these time constraints puts pressure on the compact style of presentation where each word or diagram matters and is challenging.
Week 4 – 10. 10. canceled
Week 5 – 17. 10. Ondřej Sojka
Join us at A502, Faculty of Informatics MU, on Oct 3rd at 9:50 AM (catering preparation) and 10 AM (Invitation of newcomers, questions on readings, and speaker introduction). To join via Zoom, ask for a password in advance.
Hyphenation patterns play a vital role in enhancing the readability and aesthetics of text, especially for Slavic languages. Current hyphenation systems for many Slavic languages are outdated, sometimes relying on manually created patterns with limited effectiveness. We explore the transfer learning of syllabic hyphenation patterns across multiple Slavic languages to develop improved, data-driven hyphenation systems. By using the International Phonetic Alphabet (IPA) as an intermediary, this research transfers hyphenation patterns between related Slavic languages, creating a unified set of IPA-based rules. These IPA patterns are then used to generate language-specific hyphenation patterns for each target language. The proposed approach aims to develop reliable hyphenation patterns using machine learning methods, improving syllabification across multiple languages. Although the work is ongoing, early results indicate promising improvements, particularly for Ukrainians. The new patterns are intended to be practical and easy to reproduce, ultimately contributing to better text layout quality for Slavic languages.
Week 6 – 24. 10. No lecture
Week 7 – 31. 10. No lecture
Week 8 – 7. 11. Tereza Vrabcová and Marek Kadlčík
Join us at A502, Faculty of Informatics MU, on Oct 7th at 9:50 AM (catering preparation) and 10 AM (lectures). To join via Zoom, ask for a password _in advance_.
Human communication is complex. With its many rules and components, implicit and explicit meanings of words and sentences, within the Computer Science field it has been long researched by the area of Natural Language Processing (NLP). Though we have made strides in making the "computers" understand us, one of the key elements of communication still remains unsatisfactorily unresolved - the problem of negation. In this presentation, we will delve into the role of negation in human communication, the ability (or rather inability) of large language models to tackle negation, current approaches to this problem, and the possible research directions for solving this problem.
Week 9 – 14. 11. Tomáš Gregor
In recent years, enormous progress has been made in studying and developing artificial neural networks and machine learning models that can approximate any well-behaved function to an arbitrary precision. These models can perform even superhuman tasks, such as predicting the spatial structure of any protein just from its sequence of amino acids. Another cutting-edge research area is quantum computing, which studies using the quantum properties of polarized light or supercooled materials for computation. Researchers hope to use the properties of these quantum computations to solve problems that would take more than the universe's lifetime to compute on a classical computer. At the intersection of these two research areas lie quantum neural networks, a rapidly growing research topic with much promise but little concrete results thus far. In my presentation, I will lay out the theory behind the components of quantum computing: qubits, quantum circuits, and quantum algorithms. Armed with the foundations of quantum computing, the architecture of quantum neural networks, the methods used to train these models, results, problems, and advantages over classical neural networks will be discussed.
Week 10 – 21. 11. Martin Kňažovič
As the potential of large language models (LLMs) grows, we can help businesses automate previously human-dependent processes. One such process that is crucial for many businesses is the processing of RFPs (requests for proposals). This process usually involves reading clients' emails, searching for the necessary details, and generating proposals for potential clients. This project aims to reduce the manual work involved in the RFP process. We believe that with the help of a clever AI system, only a fraction of man-hours will be needed to accomplish what teams of people spend many hours every week. Specifically, our solution utilizes LLM to process emails and their attachments to extract product-related information that humans only need to verify and price.
Week 11 – 28. 11. Michal Štefánik
Join us at A502, Faculty of Informatics MU, on November 28th at 10 AM (CET) [and on Zoom].
EMNLP is a top-tier NLP conference where leading experts in NLP and AI publish and meet together. Michal will report on the main take-home messages he brought from Miami.
Week 12 – 5. 12. Frank Mittelbach
Join us at A502, Faculty of Informatics MU, on December 5th at 10 AM (CET) or [or on Zoom].
An overview presentation of a general framework for globally optimized pagination of linear text, as well as for text plus floating objects, such as figures and tables. The framework uses a flexible constraint model that allows for the implementation of typical typographic rules that can be weighted against each other to support different application scenarios. In this context, "flexible" means that the rules of the typographic presentation of a document are not fixed but can be (to some extent) adjusted to different typographic requirements. It is easy to see that without restrictions, the float placement possibilities grow exponentially if the number of floats is linearly related to the document size. It is, therefore, important to restrict the objective function used for optimization in a way that the algorithm does not have to evaluate all theoretically possible placements while still being guaranteed to find an optimal solution. The goal is, therefore, to define a framework that is both rich in the expressiveness of modeling a large class of pagination applications and, at the same time, is capable of solving the optimization problem in an acceptable time for realistic input data.
Week 13 – 12. 12. Jakub Pekár (and Merry Christmas)
Join us at A502, Faculty of Informatics MU, on December 12th at 10 AM (CET) or [or on Zoom].
The detection of metastasis in lymph nodes is a critical step in determining the appropriate course of treatment for cancer patients. While advanced-stage metastases are often visibly evident, early-stage detection remains a challenging task. Traditionally, pathologists have relied on microscopic examination of tissue samples, but advancements in whole-slide imaging technology have made digital pathology increasingly the standard. This shift enables the application of machine learning (ML) models to enhance the accuracy, speed, and consistency of metastasis detection. Although ML models have demonstrated success in identifying certain cancer types, such as prostate cancer, detecting lymph node metastases presents unique challenges due to the variability in tissue morphology and staining patterns. This project focuses on developing an ML-based model for detecting lymph node metastases in hematoxylin and eosin (H&E)-stained digital tissue samples. However, due to the limited availability of high-quality annotated data, we propose a multi-step approach rather than tackling the problem directly. This presentation will outline the methodology, the current progress in addressing these challenges, and the future directions for this research.
Tips for readings, discussions, and presentation preparations:
- Top2Vec towardsdatascience.com/top2vec-new-way-of-topic-modelling
- How to Speak by Patrick Winston (YouTube video)
Žákovi, který se hrozil chyb, Mistr řekl: "Ti, kdo nedělají chyby, chybují nejvíc ze všech – nepokoušejí se o nic nového." Anthony de Mello: O cestě
To a student in danger, the Master said: "Those who do not make mistakes most of all – they do not try anything new." Anthony de Mello