AI/ML showcase notebook and the corresponding lecture video:

Pv287 lecture 02

Articles to be studied in groups:

Theme A - Gene-disease associations

• paper: The DisGeNET knowledge platform for disease genomics: 2019 update
• group: Sofiia Serediuk, Ondřej Sloup, Samuel Šuĺan

Theme B - Biomedical text mining

• paper: BioBERT: a pre-trained biomedical language representation model for biomedical text mining
• group: Patrik Begáň, Jan Halas, Tomášs Houfek

Theme C - Drug interactions

• paper: Modeling polypharmacy side effects with graph convolutional networks
• group: Marek Toma

Theme D - Skin lesion classification

• paper: Dermatologist-level classification of skin cancer with deep neural networks
• group: Jana Dudášová, Petr Kadlec, Ondřej Lošťák

Theme E - Denoising imaging data

• paper: PET image denoising using unsupervised deep learning
• group: Kateřina Čížková, Luboš Řehounek (theme proposer), Hana Vanovčanová

How and what to study, what to present before the first hackathon:

• At first, each member of the group will thoroughly study the paper on their own (ca. the first two weeks). Feel free to reach out to each other and sync if you want to, though.
• During this period of primarily independent work, each of you will prepare a "lightning talk" (max. 3 minutes).
• These will be presented in one of the regular group meetings (in week 05).
• Points to be addressed by the lightning talks:
  • What is the problem addressed by the paper, and its broader context?
    
  • Why is the problem relevant? 
  • Why hadn't the problem been solved before?
  • What is the presented solution about?
  • How does it differ from related works?
  • How can the approach be extended and/or used in your own work?
• After the lightning talks, you'll jointly work on a "big" presentation (max. 15 minutes, including discussion).
• This will be expected to introduce the paper to a general audience (meaning you should assume your audience has very little context  and may have limited understanding of the corresponding domain and/or technical issues) - this is harder than it seems, so don't underestimate it!
• Possible ways to distribute the workload (all done iteratively and in a collaborative manner, using for instance Google Slides):
  • One person defines the overall structure and drafts the presentation
  • One person deals with the graphics
  • One person fills in the details and presents
  • Each of you plays a "devil's advocate" towards the work of the others.
• Last but not least, you'll jointly identify an interesting problem based on the studied paper as a possible theme of your project and suggest one or more papers that may help you solve the problem (this is to conclude the paper presentation and motivate your hackathon efforts).