Laboratory for Applied Machine Learning Algorithms (LAMA)

Prerequisites

Basic programming skills are required. Knowledge of the fundamentals of information technology, signal and system theory and probability theory is also required.

Contents

The relevance of machine learning and artificial intelligence (AI) in modern society is undeniable. Advances in computing power, combined with new technologies such as AI accelerators, GPUs and FPGAs, have paved the way for efficient, parallelized frameworks and algorithms.
More recently, Large Language Models (LLMs) such as GPT and multimodal networks have fundamentally changed the AI landscape. These systems, which can process speech, image and other modalities, open up new possibilities in human-machine interaction and data analysis.
This course offers electrical engineering students a sound introduction to the methods and tools of machine learning, including current developments:

• Implementation of basic algorithms (e.g. Perceptron, Decision Trees)
• Application of industry-relevant tools (Keras, Tensorflow, Pytorch)
• Insights into the architecture and functionality of LLMs and multimodal networks
• Use of powerful workstations for practical exercises
• Working on real problems from different domains

The course covers a broad spectrum - from the analysis of medical imaging to speech processing and complex prediction models. In the "Into-the-Wild" section, students can pursue their own project ideas or participate in current research questions, possibly also in the field of LLMs or multimodal systems.

The aim is to impart the necessary skills to understand and utilize the opportunities and challenges of AI in the future professional field. The skills acquired form a solid basis for further work in the field of machine learning during the course.

Organizational matters

Attendance is compulsory for all dates and the preliminary meeting.

• You will work on the tasks in teams of two or three.
• The assessment is made up of the submitted task sheets, the "Into the wild" part and a colloquium.
• Further details on the assessment will be explained in the introductory event.
• Six ETCS points are awarded.

Course content

• Preliminary discussion and group allocation
• Processing and analysis of data sets
• Evaluation of ML systems
• Basics of supervised learning
• Unsupervised learning
• Neural networks
• Evolutionary algorithms
• Convolutional Neural Networks
• Recurrent Neural Networks and Transformer Models
• Into the Wild...

Procedure of the lab sessions 1-8

• During the individual lab sessions, students work on predefined tasks. These include programming tasks as well as tasks to be answered in text form.
• The task sheets are issued in the form of Jupyter notebooks. With this interactive development environment, you can test program codes directly in the task sheet, display solutions and complete the documentation or answers.
• After each laboratory session, the completed task sheet (the corresponding Jupyter notebook) is handed in.
• It is sufficient to hand in one copy of the notebook per group.
• Successful completion and submission of the task sheets is a prerequisite for participation in the oral colloquium at the end of the semester. Participants who fail to hand in their work will not be admitted to the colloquium!

Into the Wild...

• At the beginning of the second part, various data sets are presented, which are available for processing.
• Each group chooses one of the presented data sets, of course several groups can use the same data set. The problem is defined by the group itself. It is also possible to create your own problems for your own data sets.
• A concept for solving the problem is developed on the basis of what has been learned previously. This is implemented and tested.
• In addition to working on the data, it is also possible to optimize an existing approach for a data set in terms of runtime or latency. Among other things, an attempt can be made to achieve this using suitable hardware.
• Finally, each group prepares a presentation that introduces the developed concept and presents the results. Possible next steps should also be identified by critically reflecting on the decisions made previously.

Application procedure

This year, 30 places are expected to be offered for the lab.
You can register for the lab via the WiWi portal at the following link: https://portal.wiwi.kit.edu/ys/8270

Materials

The necessary materials for the LAMA will be made available via ILIAS. After successfully registering for LAMA, you will receive access to the tasks, data sets and additional information material.
The material is provided in English, but can be edited in German.

Tutors wanted

For former participants or other Master's students who would like to work on the topics, there are opportunities to be a tutor during the course. If you are interested, please send an e-mail with the appropriate subject to the contact address: lama∂itiv.kit.edu.