Lectures

Lecture 1/3 “Self-supervised Learning 1”

Abstract TBA

Lecture 2/3 “Self-supervised Learning 2”

Abstract TBA

Lecture 3/3 “Vision-Language Learning”

Abstract TBA

Yuki Asano

Lecture 1/3 “The AI-driven Hospital of the Future”

Abstract TBA

Lecture 2/3 “Foundations of Attention Mechanisms and Transformers”

Abstract TBA

Lecture 3/3 “AI Safety: Challenges and Solutions”

Abstract TBA

Pierre Baldi

Lecture 1/4 “Transformers Part 1”

Abstract TBA

Lecture 2/4 “Transformers Part 2”

Abstract TBA

Lecture 3/4 “Vision and VLMs Part 1”

Abstract TBA

Lecture 4/4 “Vision and VLMs Part 2”

Abstract TBA

Lucas Beyer

Lecture 1/3 “Foundation Models for Earth Systems 1/3”

Abstract TBA

Lecture 2/3 “Foundation Models for Earth Systems 2/3”

Abstract TBA

Lecture 3/3 “Foundation Models for Earth Systems 3/3”

Abstract TBA

Wessel Bruinsma

Lecture 1/3

Abstract TBA

Lecture 2/3

Abstract TBA

Lecture 3/3

Abstract TBA

Sam Buchanan

Lecture 1/3 “Foundations of GNN Expressiveness”

Abstract TBA

Lecture 2/3 “Beyond Standard GNNs: Increasing Expressiveness”

Abstract TBA

Lecture 3/3 “Expressiveness of GNNs in Practice”

Abstract TBA

Floris Geerts

Lecture 1/3 “From Large Language Models to Reasoning Models”

Abstract TBA

Lecture 2/3 “Multi-Agent System”

Abstract TBA

Lecture 3/3 “Applications of Foundation Models”

Abstract TBA

Sven Giesselbach

Lecture 1/3 “Foundation of Generative Models”

Abstract TBA

Lecture 2/3 “Physics for Generative Modeling: Diffusion Models”

Abstract TBA

Lecture 3/3 “Generative Models for Molecular Science”

Abstract TBA

Chin-Wei Huang

Lecture 1/3 “From GANs to Diffusion Models for Image Synthesis “

Abstract TBA

Lecture 2/3 “Control and Guidance in Diffusion Models”

Abstract TBA

Lecture 3/3 “Conditional Generation of Multimodal Data “

Abstract TBA

Vicky Kalogeiton

Lecture 1/3 “Introduction to Graph Neural Networks”

Abstract TBA

Lecture 2/3 “Graph Neural Networks for Physical Simulation”

Abstract TBA

Lecture 3/3 “Compositional World Models”

Abstract TBA

Thomas Kipf

Lecture 1/3 “Principles of Deep Representation Learning”

Abstract TBA

Lecture 2/3 “Principles of Deep Representation Learning”

Abstract TBA

Lecture 3/3 “Principles of Deep Representation Learning”

Abstract TBA

Yi Ma

Lecture TBA

Abstract TBA

Raniero Romagnoli

Lecture 1/3 “Detecting AI-Generated Text”

1. Detecting AI-generated text
2. SynthID: a watermark for LLM-generated text
3. Neural text generation: history to present

Lecture 2/3 “SynthID: a Watermark for LLM-Generated Text”

1. Detecting AI-generated text
2. SynthID: a watermark for LLM-generated text
3. Neural text generation: history to present

Sumanth Dathathri, Abigail See, …, Demis Hassabis & Pushmeet Kohli, “Scalable watermarking for identifying large language model outputs“, Nature volume 634, pages 818–823 (2024)

https://www.nature.com/articles/s41586-024-08025-4

Lecture 3/3 “Neural Text Generation: History to Present”

1. Detecting AI-generated text
2. SynthID: a watermark for LLM-generated text
3. Neural text generation: history to present

Abigail See

Lecture 1/3 “Thinking about thinking: Metacognitive Capabilities of LLMs”

Metacognitive knowledge refers to humans’ intuitive knowledge of their own thinking and reasoning processes. Today’s best LLMs clearly possess some reasoning processes. The paper gives evidence that they also have metacognitive knowledge, including ability to name skills and procedures to apply given a task. We explore this primarily in the context of math reasoning, developing a prompt-guided interaction procedure to get a powerful LLM to assign sensible skill labels to math questions, followed by having it perform semantic clustering to obtain coarser families of skill labels. These coarse skill labels look interpretable to humans.

To validate that these skill labels are meaningful and relevant to the LLM’s reasoning processes we perform the following experiments. (a) We ask GPT-4 to assign skill labels to training questions in math datasets GSM8K and MATH. (b) When using an LLM to solve the test questions, we present it with the full list of skill labels and ask it to identify the skill needed. Then it is presented with randomly selected exemplar solved questions associated with that skill label. This improves accuracy on GSM8k and MATH for several strong LLMs, including code-assisted models. The methodology presented is domain-agnostic, even though this article applies it to math problems.