In this community review report, we discuss applications and techniques for fast machine learning (ML) in science—the concept of integrating powerful ML methods into the real-time experimental data processing loop to accelerate scientific discovery. The material for the report builds on two workshops held by the Fast ML for Science community and covers three main areas: applications for fast ML across a number of scientific domains; techniques for training and implementing performant and resource-efficient ML algorithms; and computing architectures, platforms, and technologies for deploying these algorithms. We also present overlapping challenges across the multiple scientific domains where common solutions can be found. This community report is intended to give plenty of examples and inspiration for scientific discovery through integrated and accelerated ML solutions. This is followed by a high-level overview and organization of technical advances, including an abundance of pointers to source material, which can enable these breakthroughs.

No abstract

Machine learning methods are integrated into the pipelined first level (L1) track trigger of the upgraded flavor physics experiment Belle II at KEK in Tsukuba, Japan. The novel triggering techniques cope with the severe background from events outside the small collision region provided by the new SuperKEKB asymmetric-energy electron-positron collider. Using the precise drift-time information of the central drift chamber which provides axial and stereo wire layers, a neural network L1 trigger estimates the 3D track parameters of tracks, based on input from the axial wire planes provided by a 2D track finder. An extension of this 2D Hough track finder to a 3D finder is proposed, where the single hit representations in the Hough plane are trained using Monte Carlo. This 3D finder improves the track finding efficiency by including the stereo sense wires as input. The estimated polar track angle allows a specialization of the subsequent neural networks to sectors in the polar angle.

Recently proposed quantum systems use frequency multiplexed qubit technology for readout electronics rather than analog circuitry, to increase cost effectiveness of the system. In order to restore individual channels for further processing, these systems require a demultiplexing or channelization approach which can process high data rates with low latency and uses few hardware resources. In this paper, a low latency, adaptable, FPGA-based channelizer using the Polyphase Filter Bank (PFB) signal processing algorithm is presented. As only a single prototype lowpass filter needs to be designed to process all channels, PFBs can be easily adapted to different requirements and further allow for simplified filter design. Due to reutilization of the same filter for each channel they also reduce hardware resource utilization when compared to the traditional Digital Down Conversion approach. The realized system architecture is extensively generic, allowing the user to select from different numbers of channels, sample bit widths and throughput specifications. For a test setup using a 28 coefficient transpose filter and 4 output channels, the proposed architecture yields a throughput of 12.8 Gb/s with a latency of 7 clock cycles.

The Compressed Baryonic Matter (CBM) experiment is designed to handle interaction rates of up to 10 MHz and up to 1 TB/s of raw data generated. With triggerless streaming data acquisition in the experiment and beam intensity fluctuations, it is expected that occasional data bursts will surpass bandwidth capabilities of the Data Acquisition System (DAQ) system. In order to preserve integrity of event data, the bandwidth of DAQ must be throttled in an organised way with minimum information loss. The Timing and Fast Control (TFC) system provides a latency-optimised datapath for throttling commands and distributes a system clock together with a global timestamp. This paper describes a prototype design of the system with focus on synchronisation and its evaluation.