EveNet: Towards a Generalist Event Transformer for Unified Understanding and Generation of Collider Data

With the scale of machine learning (ML) models and datasets expanding, foundation models have revolutionized the application of ML to real-world challenges. Multimodal language models such as ChatGPT and Llama have expanded their capabilities for specialized tasks through shared pre-training. In high-energy physics (HEP), analyses frequently encounter persistent challenges that necessitate scalable, data-driven approaches. In this talk, we introduce a foundation model tailored for high-energy physics. EveNet harnesses vast simulated datasets during pre-training to tackle common tasks across various analyses, providing a cohesive foundation for domain-specific adaptations. We showcase the advantages of this pre-trained model in enhancing search sensitivity, anomaly detection, event reconstruction, feature engineering, and more. By capitalizing on the strengths of pre-trained architectures, we aim to advance the frontiers of discovery with heightened efficiency and deeper insights.