2381 Data Science and Artificial Intelligence I B

Johann Mitloehner, 2024

29.11. Introduction
- Intelligence - Definitions, Tests, AI past and present ||| nb
- Slides 1 (Learn)
06.12. Machine Learning: reinforcement and connectionist
- Reinforcement Learning ||| nb
- before the break: Timur Sharifullin, Make a Copy of your Friend: Very Meaningless LLM
- after the break: Peter Kovac, AI in Defense
- Neural Net Introduction: One-layer Feed-Forward ||| nb
- Example: Iris dataset with Perceptron and pure Python Single-Layer network ||| nb
13.12. Extending the basic model
- Two-layer Net and Backpropagation ||| nb
- More Examples: networks with mushrooms and sentiment sentences ||| nb
20. 12. Connectionist Machine Learning Framework: PyTorch
- before the break: Patricia Haumer, Gender bias in AI
- Adaptive models with PyTorch ||| nb
10.01. Symbolic AI
- Slides 5 (Learn)
- Answer Set Programming with Clingo ||| nb
- More Challenging Examples ||| nb
- Slides 6 (Learn)
- Knowledge Extraction ||| nb
17.01. Deep learning, word embeddings, and applications: Transformers and Large Language Models
- Embeddings, LSTM ||| nb
- Comparison: Decision Tree
- Mulugeta, The Use of AI in Cybercrime
- Hauer, Schoditsch, Regulating AI: Understanding the EU AI Act and its implications
- Scepka, Zero-Shot-Learning
- Transformers, GPT, BERT ||| nb
- Build your own Transformer ||| nb
- offline and small-scale LLM
24.01. Programming Project Presentations
- (business) application problem solved in symbolic AI, connectionist machine learning, or hybrid approach; dataset(s) of your own choice, e.g. from Kaggle or UCI repository; see ToDo in Learn
- Seiss, AutoKeras
- Mashkovskaia and Al-Serori, Austrian AI strategy

Extra Credit

Give us a talk on a topic of your choice!
- In-depth on something we did not cover in detail, or
- additional subject areas, coding examples, case studies, or
- inspiring videos linked in lecture slides 1 on Learn
Topics are of course within the scope of the course i.e. connectionist and symbolic AI systems, past and present. Let me know your idea and we will schedule your talk within one of the course units.
This can be in addition to, or instead of, your programming project.
Earn credit up to 12 points depending on scope and duration of your talk.
Gain experience preparing and presenting a challenging topic!

Software and other Sources

We only use free Open Source software, so you can install everything on your own computer.

If needed, get Python from python.org and (in Windows) click 'Add to Path' when you run the installer.
Use pip on the command line to install additional Python packages as needed, e.g.
pip install jupyter pandas matplotlib scikit-learn
For the Pytorch package (connectionist part) see pytorch.org for the installation options.
The Clingo software (symbolic part) also has various installation options, see potassco.org/clingo
We also have Jupyter notebook servers ready for you at the Learn web page for this course (Learning Activities).

Books:

Artificial Intelligence: A Modern Approach, by Stuart Russel and Peter Norvig, 3rd ed. 1995, free on archive.org
The Little Book of Deep Learning, by Francois Fleuret, 2024-; free PDF formatted for phone

Datasets:

The UC Irvine machine learning repository has over 600 datasets for studying various adaptive methods; the collection is much smaller than other sites, but the datasets tend to be in good quality and format, and many have a Python interface for accessing them dynamically in your code. Login/register not required.
Hugging Face Datasets, huge collection with over 260k datasets, also offers tutorials and Python package for data loading. Login/register not required.
Kaggle also offers a huge number of datasets, among other things; the site requires registering for download.

More Examples:

Dec 1, 2023: UCI Wine Dataset
Dec 22, 2023: UCI Student Performance

Workshop "A Paradigm Shift in Computer Science", Nov 28th - 29th 2024, TU Wien
Live stream videos are now available for viewing

Screencast videos: