1 / 7

Fundamental Physics

erhtjhtyhy. Co-Leads Marcel Demarteau – Oak Ridge National Laboratory Torre Weanus – Brookhaven National Laboratory Bronson Messer– Oak Ridge National Laboratory Participants – 29 with full list in the xls. Fundamental Physics. A possible taxonomy…. Adapted from Prabhat.

ecrane
Download Presentation

Fundamental Physics

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. erhtjhtyhy Co-Leads Marcel Demarteau– Oak Ridge National Laboratory Torre Weanus– Brookhaven National Laboratory Bronson Messer– Oak Ridge National Laboratory Participants – 29 with full list in the xls. Fundamental Physics

  2. A possible taxonomy… Adapted from Prabhat

  3. A possible taxonomy… Adapted from Prabhat

  4. A possible taxonomy… CNNs, Graph NNs, RNNs Auto-encoders, PCA, random forest ??? VAEs, GANs RL Adapted from Prabhat

  5. FRIB (2022) • Challenges: Event characterization (physics based), moving beyond tracking-based analysis (TPC, gamma-ray spectroscopy), which rates to measure? • Possible Research Areas: • Regression and classification for combined detector analyses • Detector response deconvolution • 30 orders of magnitude in reaction rates; present approaches not robust to this • Transfer learning for short-duration (O(week)) experiments • Beam tuning and delivery (RL) -- too expensive currently • Analysis: Decision support for level analysis; automation? • Significant validation question; including constraints • Design of experiments: Identify thermonuclear reactions for further study that are critical in cosmic element creation • Push back to simulation: Use models to replace interpolation in derived tables (e.g. equation of state, nuclear networks)

  6. LHC (2021; HL 2026), DUNE • Challenges: Data volumes that are 10x, complexity much greater; event generation and simulation at scale; training at scale; data transfer and workflows; • Possible Research Areas: • ”Training as a (set of) Service(s)” – enabling scalable training for fast-turnaround physics analyses • How to integrate new tools into the (well-established) workflow? • Data models • Moving towards a knowledge base/registry of useful “building blocks” (e.g. in the TaaS workflow) • Anomaly detection – optimized selection and implementation of methods • Training and validation of models with simulated and experimental data • Bias and uncertainty of the model(s) • Use of fast simulation (informed by ML models) to make use of trained models • How to use full simulations to train (active learning)? • Do we need new AI methods/algorithms or enhancements to current approaches to capture rare events? to enable this validation? • Improved trigger processing – more physics and faster (inference) • Training at scale (to decrease turnaround time) is a prerequisite • ASICs (maybe more realistic [and just as useful] for EIC) –latency and power constraints are strong

  7. EIC (2030) • Challenges: One big inverse problem—determine the structure of the nucleon • Integrating data from other facilities (universal analysis) • What does the question mean? What is the way to understand the physics deliverable? Imaging? • Possible Research Areas: • Using ML/DL to find correlations in disparate data sets • EIC can enable new ideas here with respect to data curation (cf. ATLAS) • Using AI or ML to constrain parameter-rich theories • Try to close the experiment-theory gap • What is observable? What do we need to measure? (hypothesis testing) • Fast theoretical calculation (properly validated)  design of experiments • Streaming readout (see LHC; large data volume at high cadence) • Using models to guide HW design for accelerator and detectors

More Related