Applications and Techniques for Fast Machine Learning in Science

Deiana, A. M.; Tran, N. V.; Agar, Joshua C.; Blott, Michaela; Guglielmo, Giuseppe Di; Duarte, J.; Harris, P.; Hauck, Scott; Liu, Mia; Neubauer, M. S.; Ngadiuba, J.; Ogrenci-Memik, Seda; Pierini, M.; Aarrestad, Thea Klaeboe; Bähr, Steffen; Becker, Jürgen C.; Berthold, A.; Bonventre, Richard; Bravo, Tomas E. Muller; Diefenthaler, M.; Dong, Zhen; Fritzsche, N.; Gholami, Amir; Говоркова, Е.; Hazelwood, Kyle; Herwig, T. C.; Khan, Babar; Kim, Sehoon; Klijnsma, T.; Liu, Yaling; Lo, Kin Ho; Nguyen, Tri Q.; Pezzullo, Gianantonio; Rasoulinezhad, S A; Rivera, R.; Scholberg, K.; Justin, Selig,; Sen, Sougata; Strukov, Dmitri B.; Tang, W. M.; Thais, S. J.; Unger, K.; Vilalta, Ricardo; Krosigk, B. von; Warburton, Thomas; Flechas, Maria Acosta; Aportela, Anthony; Calvet, T. P.; Cristella, L.; Díaz, D.; Doglioni, C.; Galati, Maria Domenica; Khoda, E. E.; Fahim, Farah; Giri, Davide; Hawks, Benjamin; Hoang, Duc; Holzman, B.; Hsu, S.‐C.; Jindariani, S.; Johnson, Iris DeLoach; Kansal, Raghav; Kastner, Ryan; Katsavounidis, E.; Krupa, J.; Li, Pan; Madireddy, Sandeep; Marx, E. J.; McCormack, Patrick; Meza, Andrés; Mitrevski, J.; Attia, Mohammed, Mohammed; Mokhtar, F.; Moreno, Éric I.; Nagu, S.; Narayan, R.; Palladino, N.; Que, Zhiqiang; Park, Sang Eon; Ramamoorthy, Subramanian; Rankin, D.; Rothman, Simon; Sharma, Ashish; Summers, S.; Vischia, P.; Vlimant, J. R.; Weng, Olivia

doi:10.48550/arxiv.2110.13041

Cited by 10 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The current state of the art for heterogeneous digital and analog solutions needs to be pushed further to allow for dynamic and reconfigurable hardware, the usage of heterogeneous systems (e.g. Central Processing Units (CPUs), Graphical Processing Units (GPUs), Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), and Systems-on-Chip (SoCs)) and the possibility to deploy advanced algorithms (Artificial Intelligence (AI), Machine Learning (ML) [1]) on the chosen configuration (see also Section 4). Following the choice of physics selection and filtering algorithms and architectures that are deployed on these commodity systems, it is necessary to investigate the advantages and possible physics impacts of that approach above and beyond the current state-of-the-art.…”

Section: Challenges Faced By Current and Future Physics Facilitiesmentioning

confidence: 99%

See 1 more Smart Citation

Innovations in trigger and data acquisition systems for next-generation physics facilities

Bartoldus¹,

Bernius²,

Miller³

2022

Preprint

View full text Add to dashboard Cite

Data-intensive physics facilities are increasingly reliant on heterogeneous and large-scale data processing and computational systems in order to collect, distribute, process, filter, and analyze the ever increasing huge volumes of data being collected. Moreover, these tasks are often performed in hard real-time or quasi real-time processing pipelines that place extreme constraints on various parameters and design choices for those systems. Consequently, a large number and variety of challenges are faced to design, construct, and operate such facilities. This is especially true at the energy and intensity frontiers of particle physics where bandwidths of raw data can exceed 100 TB/s of heterogeneous, high-dimensional data sourced from 300M+ individual sensors. Data filtering and compression algorithms deployed at these facilities often operate at the level of 1 part in 10 5 , and once executed, these algorithms drive the data curation process, further highlighting the critical roles that these systems have in the physics impact of those endeavors. This White Paper aims to highlight the challenges that these facilities face in the design of the trigger and data acquisition instrumentation and systems, as well as in their installation, commissioning, integration and operation, and in building the domain knowledge and technical expertise required to do so.

show abstract

Section: Challenges Faced By Current and Future Physics Facilitiesmentioning

confidence: 99%

“…Inference for "on-detector" processing with mixed architecture chips is extremely powerful and broadly applicable for any high-speed processing tasks across the intensity, cosmic, energy frontiers of physics, as well as basic energy sciences, nuclear physics, and more [1,25].…”

Section: On-detector Inference and Self-driving Physics Facilitiesmentioning

confidence: 99%

Innovations in trigger and data acquisition systems for next-generation physics facilities

Bartoldus¹,

Bernius²,

Miller³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…For example, the CSVv2 algorithm based on a simple neural network for b-quark identification was adopted for use in the HLT by the CMS experiment [152]. For upcoming data taking periods, plans exist to include machine learning -including deep networks -also at L1T [153][154][155][156][157][158][159], to improve the selection efficiency.…”

Section: A Energy Frontiermentioning

confidence: 99%

Machine Learning in the Search for New Fundamental Physics

Karagiorgi¹,

Kasieczka²,

Kravitz³

et al. 2021

Preprint

View full text Add to dashboard Cite

Machine learning plays a crucial role in enhancing and accelerating the search for new fundamental physics. We review the state of machine learning methods and applications for new physics searches in the context of terrestrial high energy physics experiments, including the Large Hadron Collider, rare event searches, and neutrino experiments. While machine learning has a long history in these fields, the deep learning revolution (early 2010s) has yielded a qualitative shift in terms of the scope and ambition of research. These modern machine learning developments are the focus of the present review.

show abstract

“…Using machine learning (ML) to increase the trigger efficiency is a long-established idea [13], and simple neural networks for jet tagging have been used, for example, in the CMS highlevel trigger [14]. The advent of ML-compatible field-programmable gate arrays (FPGAs) has opened new possibilities for employing such classification networks even at the low-level trigger [15][16][17][18][19][20][21]. ML-inference on FPGAs is making rapid progress [15], but the training of sophisticated networks on such devices is still an active area of research.…”

Section: Introductionmentioning

confidence: 99%

Ephemeral Learning -- Augmenting Triggers with Online-Trained Normalizing Flows

Butter,

Diefenbacher,

Kasieczka

et al. 2022

Preprint

View full text Add to dashboard Cite

The large data rates at the LHC require an online trigger system to select relevant collisions. Rather than compressing individual events, we propose to compress an entire data set at once. We use a normalizing flow as a deep generative model to learn the probability density of the data online. The events are then represented by the generative neural network and can be inspected offline for anomalies or used for other analysis purposes. We demonstrate our new approach for a toy model and a correlation-enhanced bump hunt.

show abstract

Applications and Techniques for Fast Machine Learning in Science

Cited by 10 publications

References 0 publications

Innovations in trigger and data acquisition systems for next-generation physics facilities

Innovations in trigger and data acquisition systems for next-generation physics facilities

Machine Learning in the Search for New Fundamental Physics

Ephemeral Learning -- Augmenting Triggers with Online-Trained Normalizing Flows

Contact Info

Product

Resources

About