Discovery Challenges
Predictive Online Digital Sales (PODS) and Marketing
Digital advertisements of products and services are commonplace in almost every online e-commerce platform. The objective in this Discovery Challenge is to optimize sponsored ad targeting in e-commerce platforms where ads show up in response to keyword-based search by the users. The first task involves predicting future Click Through Rate (CTR) for a keyword based on campaign performance data (e.g. keyword bid, cost-per-click) for thousands of related keywords. The second task is to predict future ad-conversion. Participants will develop scalable algorithms that can be used for large scale online campaign management. Agnik is releasing campaign management data for the first time to support this competition and advance machine learning research in this emerging field. The winner will receive free registration for the ECML-PKDD 2025. Moreover, we will offer prize money to the top three winners. The winning team will receive 500€, the second-place 300€, and the third-place 200€.
Contact Email: pods2025@agnik.com
Colliding with Adversaries: A Challenge on Robust Learning in High Energy Physics (CARL-HEP)
Adversarial machine learning has become a key area of research for improving model robustness and understanding model behavior. While much of the focus has been on domains like image recognition and natural language processing, adversarial attacks on tabular data — common in fields such as medicine and High Energy Physics (HEP) — have received less attention. This challenge seeks to address that gap by applying adversarial techniques to tabular data, a domain where adversarial vulnerabilities have been less explored despite their potential to improve model robustness. By focusing on tasks related to generating adversarial examples and creating models resilient to them, participants will explore innovative methods that could enhance robustness in fields such as particle physics. This challenge not only advances the development of more reliable machine learning systems but also offers opportunities to improve model explainability, performance under data scarcity, and inspire new approaches to adversarial robustness in various scientific fields.
Contact Email: collidingadversaries@googlegroups.com
In-silico Genomics Benchmarking for Neural Models (OGB)
RNA molecules are crucial for cellular processes, and accurately predicting their structure and function remains challenging due to RNA's flexibility and limited experimental data. This competition focuses on advancing RNA-oriented foundation models (GFMs) to improve RNA structure prediction, functional characterization, and molecular design. The challenge encourages participants to enhance existing GFM models, develop new architectures, or integrate traditional machine learning methods to address key issues in RNA sequence behaviour, structural analysis, and functional inference. By benchmarking RNA GFMs, this competition aims to drive innovations in computational genomics, facilitate the design of RNA-based therapeutics, and improve our understanding of RNA biology. Success in this challenge will accelerate research in biotechnology, personalized medicine, and the development of RNA-targeted therapies for diseases like cancer and viral infections, ultimately enhancing both predictive capabilities and experimental methodologies in the field.
Contact Email k.li@exeter.ac.uk
Atmosphere Machine Learning Emulation Challenge (AMLEC)
Atmospheric Radiative Transfer Models (RTMs) are essential tools in climate and Earth sciences but are computationally intensive, limiting their direct use in operational settings. Common solutions like look-up table (LUT) interpolation reduce this burden but require large, memory-heavy datasets and lack generalization. These limitations are especially critical for hyperspectral satellite missions, where data volume grows exponentially. Emulation offers a promising alternative by replacing costly simulations with fast, accurate statistical models that replicate RTM behavior. This enables real-time data processing, improved atmospheric correction, and efficient climate modeling. However, emulating RTMs is challenging due to high-dimensional inputs and complex physics. The Atmosphere Machine Learning Emulation Challenge (AMLEC) aims to advance surrogate modeling and physics-aware AI, accelerating progress in remote sensing, weather forecasting, and climate research.
Contact Email: jorge.vicent@uv.es
