Hsu et al. present EVENET, a foundation model pretrained on simulated particle collisions using both self-supervised learning and physics labels, which transfers across many analysis tasks and works o...

Data-intensive collider experiments increasingly require models that transfer across high-dimensional, partially observed measurements, yet most analyses still optimize separate models for each task. Here, to address this bottleneck, we introduce EVENET, an event-level foundation model pretrained on 500 million simulated collision events using self-supervised learning and physics-informed supervision. The model constructs a shared particle-cloud representation from the variable-length set of physics objects in each event. Across searches for heavy resonances and exotic Higgs decays, EVENET outperforms both established and state-of-the-art baselines, particularly in low-statistics regimes. We validate transfer to experimental data by rediscovering the ϒ meson in CMS Open Data. The pretrained model also supports precise extraction of quantum-correlation observables that remain stable under the tested systematic variations. Together, these results establish a reusable backbone for diverse

We evaluate EVENET, an end-to-end event-level foundation model designed to learn a unified representation across a heterogeneous set of analysis tasks. EVENET operates directly on variable-length particle-cloud inputs using a single backbone that can be applied across diverse observational regimes, including fully reconstructed, partially reconstructed and noise-perturbed events. The backbone is pretrained with approximately balanced event counts across the process–final-state channels listed in Extended Data Table 1, without cross-section-based event weighting or sampling, thereby avoiding the underrepresentation of rare processes. Pretraining follows the hybrid pretraining strategy described in ‘Pretraining strategy selection’: (1) a self-supervised masking phase, followed by (2) a joint optimization phase combining self-supervised and supervised objectives. We then benchmark a suite of downstream tasks that separately probe generalization (‘Generalization across hypothesis grids for discovery’), simulation-to-data transfer (‘Simulation-to-data transfer with resonance rediscovery’), multi-task reconstruction and robustness to detector-level systematic variations (‘Physics-structured multi-task synergy and robustness’) and training efficiency (‘Training convergence’). Pretraining strategy selection To quantify the contribution of the proposed pretraining strategy, we compare three EVENET variants that differ only in their pretraining objectives while sharing an identical backbone architecture and downstream fine-tuning procedure. Table 1 summarizes the prediction heads used during each pretraining stage and downstream application, and...

We evaluate EVENET, an end-to-end event-level foundation model designed to learn a unified representation across a heterogeneous set of analysis tasks. EVENET operates directly on variable-length particle-cloud inputs using a single backbone that can be applied across diverse observational regimes, including fully reconstructed, partially reconstructed and noise-perturbed events. The backbone is pretrained with approximately balanced event counts across the process–final-state channels listed in Extended Data Table 1, without cross-section-based event weighting or sampling, thereby avoiding the underrepresentation of rare processes. Pretraining follows the hybrid pretraining strategy described in ‘Pretraining strategy selection’: (1) a self-supervised masking phase, followed by (2) a joint optimization phase combining self-supervised and supervised objectives. We then benchmark a suite of downstream tasks that separately probe generalization (‘Generalization across hypothesis grids for discovery’), simulation-to-data transfer (‘Simulation-to-data transfer with resonance rediscovery’), multi-task reconstruction and robustness to detector-level systematic variations (‘Physics-structured multi-task synergy and robustness’) and training efficiency (‘Training convergence’). Pretraining strategy selection To quantify the contribution of the proposed pretraining strategy, we compare three EVENET variants that differ only in their pretraining objectives while sharing an identical backbone architecture and downstream fine-tuning procedure. Table 1 summarizes the prediction heads used during each pretraining stage and downstream application, and...

https://www.gizmodo.jp/article/2610-orca-echo-monitor-evenet-recruitment/?utm_source=twitter&utm_medium=feed

Hey, I am Rei Loire, a White Tiger Furry from Germany (if my accent is not already a dead giveaway). I like to play RPGs, Plattformers and Adventure games and will stream whatever I just feel like dur...

Join us for a free seminar breaking down Gaussian splatting in archaeological and museum contexts!

EVE draws from the bleeding edge of experimental electronica, industrial, and darkwave. In EVE’s world, contrasts are everything: light and dark, beauty and grotesquery, silence and overwhelming noise

Der Vorsitzende der Deutschen Bischofskonferenz, Bätzing, hat antidemokratische Bewegungen innerhalb der katholischen Kirche kritisiert.

going to be the number 1 marvel punisher in blood hunt evenet

going to be the number 1 marvel punisher in blood hunt evenet

Lightweight training and inference package built on top of evenet-core.