DM3L 1st Annual Symposium 2025
sOrganized by: N. Trolese, R. Furrer
Home | Registration | Programme | Speakers | Pictures
Speakers
Prof. Damian Borth"Weight Space Learning: How to learn Representations of Neural Networks"
As the number of trained neural network models continues to grow, a fascinating opportunity has emerged to learn from these diverse model populations, known as "model zoos". In this talk, we will explore the recent advances in "weight spaces learning" aiming to learn representations of model weights and their applications to discriminative and generative downstream tasks. On the discriminative side, such representations enable advanced model analysis such as e.g., the prediction of model performance without requiring access to test data. On the generative side, these representations support the sampling of new, high-performing models for tasks like initialization, transfer learning, and ensemble creation. Attendees will gain insights into not only how weight space learning could be applied in real-world machine learning tasks such as image classification, but also how it could pave a path towards the training of a foundation model of neural networks.
Prof. Rama Cont
"Generative models for financial scenario simulation"
Generative models have become the focus of a huge body of research in Machine Learning, and many researchers have deployed these methods on financial time series in an attempt to build 'market generators'. However, in absence of a clear benchmarking methodology, the competitiveness of such models compared to model-based Monte Carlo simulation remains unclear and model validation remains a stumbling block for large-scale deployment of GenAI in finance and risk management.
We propose a systematic and interpretable approach to the validation of generative models in finance and illustrate how these validation critera may be used to design interpretable machine learning algorithms for financial applications.
References:
R Cont, M Cucuringu, Renyuan Xu, Chao Zhang (2025) Tail-GAN: Learning to Simulate Tail Risk Scenarios. Management Science..
M Vuletic, R Cont (2024) VOLGAN: a generative model for arbitrage-free implied volatility surfaces, Applied Mathematical Finance, 31:4, 203-238.
Prof. Martin Jaggi
"Training a highly multilingual open-data open-weights large language model"
During the last year, many students and researchers from across Switzerland have contributed to the training of a new large language model. The model comes in 8B and 70B parameter sizes, is fully transparent in its training process, as all training data and code recipes are made openly available, while being compliant with European and Swiss regulations. The model supports over 1000 languages, many of which are under-represented in current commercial models. In this talk, we will share experiences from different aspects of the project, ranging from scientific to engineering and organizational challenges, and discuss promising directions for LLM research in several dimensions in pre-training and post-training, including efficiency, data curation, evaluation and multi-lingual capabilities.
Prof. Fariba Karimi
"Inequality and Fairness in social networks and algorithms"
While algorithms promise many benefits, including efficiency, objectivity, and accuracy, they may also introduce or amplify biases. In this talk, I show how biases in our social networks are fed into and amplified by ranking and recommender systems. Drawing from social theories and fairness literature, we argue that biases in social connections need to be taken into consideration when designing people recommender systems.
Prof. Andrea Wulzer
"Statistical Learning for Statistical Inference"
Particle collider physics offers a privileged environment characterized by abundant and complex data that can be modeled accurately and with well-characterized uncertainties by the Standard Model of fundamental particles and interactions. Assessing the Standard Model validity is our goal, to be accomplished by a sophisticated and principled statistical analysis of the data where Machine Learning methodologies are deployed at many stages. The talk will focus on two Machine Learning ideas for statistical inference that aim at improving the amount of information that can be extracted from the data. One is the systematic deployment of Optimal Inference by accessing the Likelihood ratio. The other is the model-agnostic comparison of the data with the Standard Model within a Goodness-of-Fit setup.