analiza aiqbench
This research introduces a novel particle-level generative model using residual-quantized tokens to address simulation bottlenecks at the High-Luminosity Large Hadron Collider. By leveraging autoregressive transformers trained on discrete event data, the study establishes an empirical framework where token-level loss effectively predicts downstream physical fidelity. The work demonstrates scalable generation capabilities for collider events, bridging advanced discrete modeling techniques with high-energy physics requirements.
Scaling Collider Event Generation with Residual-Quantized Tokens
arXiv:2610.00569v1 Announce Type: cross Abstract: Full detector simulation and reconstruction of collider events are projected to become major bottlenecks at the High-Luminosity Large Hadron Collider, motivating the development of fast, ML-based surrogates. At the same time, LLMs have driven fast progress in generative discrete modeling: autoregressive transformers trained on tokenized data now represent the state of the art across a range of generative tasks. We extend the discrete modeling paradigm by introducing a particle-level generative model trained on residual-quantized full-event data. We demonstrate the ability of this model family to perform conditional generation from detector-stable particles; we study its scaling behavior across a range of dataset and model sizes, characterize the effects of repeated data exposure and demonstrate that token-level loss systematically predicts downstream physical fidelity. These results provide an empirical framework for scalable collider full-event generation based on residual-quantized representations.
Abstract:Full detector simulation and reconstruction of collider events are projected to become major bottlenecks at the High-Luminosity Large Hadron Collider, motivating the development of fast, ML-based surrogates. At the same time, LLMs have driven fast progress in generative discrete modeling: autoregressive transformers trained on tokenized data now represent the state of the art across a range of generative tasks. We extend the discrete modeling paradigm by introducing a particle-level generative model trained on residual-quantized full-event data. We demonstrate the ability of this model family to perform conditional generation from detector-stable particles; we study its scaling behavior across a range of dataset and model sizes, characterize the effects of repeated data exposure and demonstrate that token-level loss systematically predicts downstream physical fidelity. These results provide an empirical framework for scalable collider full-event generation based on residual-quantized representations.
High Energy Physics - Experiment (hep-ex); Machine Learning (cs.LG); High Energy Physics - Phenomenology (hep-ph); Data Analysis, Statistics and Probability (physics.data-an)
arXiv:2610.00569 [hep-ex]
(or arXiv:2610.00569v1 [hep-ex] for this version)
https://doi.org/10.48550/arXiv.2610.00569 arXiv-issued DOI via DataCite (pending registration)
Text integral sub licența deschisă a publicației · 172 cuvinte
Subiecte conexe
Productionizing TurboQuant on AMD GPUs for KV-Cache-Bound LLM Inference
AMD · 11 iun. 2026
The hard physics and complex economics of AI’s insatiable hunger for power
SiliconANGLE · 11 sept. 2026
Miso Labs Releases MisoTTS: An 8B Emotive Text-to-Speech Model with Open Weights
MarkTechPost · 4 iun. 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · 13 oct. 2026