Invited Speakers

Large Language Models are having a major impact on the world. Even though these LLMs are impressive, it is unclear if you can really trust them. This talk will present ongoing research from the EU project TrustLLM which has the goal of developing more factual and trustworthy large language models. To achieve the ambitious objectives of this project, TrustLLM will tackle the full range of challenges of LLM development, from ensuring sufficient quality and quantity of multilingual training data, to sustainable efficiency and effectiveness of model training, to enhancements and refinements for factual correctness, transparency, and trustworthiness, to a suite of holistic evaluation benchmarks validating the multi-dimensional objectives.

Reliable forecasting of the Earth system is essential for mitigating natural disasters and supporting human progress. Traditional numerical models, although powerful, are extremely computationally expensive1. Recent advances in artificial intelligence (AI) have shown promise in improving both predictive performance and efficiency, yet their potential remains underexplored in many Earth system domains. Here we introduce Aurora, a large-scale foundation model trained on more than one million hours of diverse geophysical data. Aurora outperforms operational forecasts in predicting air quality, ocean waves, tropical cyclone tracks and high-resolution weather, all at orders of magnitude lower computational cost. With the ability to be fine-tuned for diverse applications at modest expense, Aurora represents a notable step towards democratizing accurate and efficient Earth system predictions. These results highlight the transformative potential of AI in environmental forecasting and pave the way for broader accessibility to high-quality climate and weather information.

Causal inference aims to determine how changes in one variable affect others and is crucial for evaluating the impact of interventions in fields such as healthcare, marketing, and policing. Moreover, it is becoming increasingly important for developing fair and explainable machine learning systems. In real-world scenarios, however, empirical trials are often infeasible

Time series classification is a fundamental data science task, interpreting dynamic processes as they evolve over time. Convolutional kernels provide an effective method for extracting a wide range of different forms of information from time series data. I present the Rocket family of time series classification technologies that utilize convolutional kernels to achieve state-of-the-art accuracy with many orders of magnitude greater efficiency and scalability than any alternative.  These make time series classification feasible at hitherto unattainable scale.  The methods also have potential application across many other forms of time series analysis, including extrinsic regression, clustering, anomaly detection, segmentation and forecasting.