Current Trends in AI

Advanced GenAI (17 April 2024, 12.30-15.30h)

Dr. Thomas Winters

These days, we are surrounded by creative text and image generators like GPT and diffusion models that seem to be able to generate anything we want. But how do we ensure that these types of AI truly aid us in overcoming our unique challenges? This talk sheds light on several techniques for controlling such generative models. We look at several powerful prompt engineering techniques – the art of enhancing our communication with AI – and useful ways of connecting these generators to other systems. We dive into the world of autoregressive text generators, learn their inner mechanisms and which training phases they went through to get to the current state-of-the-art text generators. These insights help us understand why certain prompt engineering techniques (such as few-shot prompting, role-prompting and chain-of-thought prompting) are able to outperform simpler prompting methods. We also briefly look at several other techniques to overcome the limitations of such models, such as retrieval-augmented generation and function calling. Similarly, we uncover the workings of diffusion models and show several techniques to gain more control over the generated images. We show how even some of AI's classic hard problems, such as humour generation, become even more within reach thanks to these large language models and their prompt engineering techniques.

Knowledge Graphs (17 April 2024, 16.00-19.00h)

Prof. Pieter Bonte and prof. Anastasia Dimou

Knowledge graphs have become the ultimate technology for unlocking the full potential of your data, illuminating the connections between entities, attributes, and relationships with unparalleled clarity. Sharing & exchanging data, harvesting insights, driving innovation, fostering integration, and revolutionizing data exploration, knowledge graphs pave the way for transformative discoveries and understanding.

In this seminar, the following topic will be tackled:

• Introduction to knowledge graphs
• Semantic Web basics: RDF
• From raw data to knowledge graphs with [R2]RML
• Unlocking insights with querying through SPARQL

MLOps (25 April 2024, 12.30-15.30h)

Prof. Mathias Verbeke and drs. Lara Luys

One of the main challenges for industry today is to get machine learning models out of the proof-of-concept phase and into production. This is not an easy task since data in production is not static, due to which the model performance can degrade over time. Machine learning operations or MLOps is a paradigm that aims to address this problem. Being a contraction of Machine Learning and DevOps, MLOps focuses on developing and maintaining machine learning models in production. This includes training, evaluating as well as monitoring the model. In this seminar, the MLOps pipeline and the underlying principles will be explained, illustrated by means of a number of tools that can be used in the MLOps process.

AI@Edge (25 April 2024, 16.00-19.00h)

Prof. Hans Hallez and drs. Gregory De Ruyter

Edge Computing has been proven to be an optimised way to delegate computation from the cloud towards the devices where the sensing takes place. Edge computing on embedded devices mostly limits itself towards compression, filtering or other basic analysis. Recent trends also show that devices near the edge of the sensor network are capable of machine learning algorithms. In this session, we will explore different techniques to bring machine learning towards the edge network and deploy these algorithms. First, we will give an overview of what embedded devices are, and how we can perform machine learning at these devices both in inference and in training. Second, we will give a hands-on demonstration as an inspiration of how machine learning at the edge can be implemented.