A deployed real-time end-to-end deep learning algorithm for fast radio burst detection
Bibliographic record
Abstract
Context . Over the past decade, fast radio bursts (FRBs) have attracted substantial interest in the field of astrophysics due to their extremely energetic nature, drawing considerable speculation regarding the mechanisms that are behind these fast transient events. To further our understanding of FRBs, it is essential to develop fast and efficient analysis pipelines to recover more of these events in radio astronomy observations. Aims . We developed a fast end-to-end deep learning based FRB detection pipeline capable of handling ~100 Gb/s of real-time data throughput without applying dedispersion techniques. Methods . We introduced a modified masked ResNet-38 model designed for FRB detection tasks. Using synthetic injections, we demonstrated that our trained end-to-end model matches and surpasses current established pipelines (on injections) with a 7% gain in accuracy without the need for dedispersion or radio frequency interference masking. We deployed this model in a real-time setting at the Allen Telescope Array. Utilizing Nvidia Holoscan, a new GPU-accelerated sensor processing platform along with model optimizations, our pipeline successfully executed an end-to-end FRB detection on beam-formed spectrograms. Results . We report that our end-to-end pipeline achieves a latency of 150× faster than real-time production constraints compared to current state-of-the-art dedispersion + ML assisted FRB search pipelines at the Allen Telescope Array, which is three times slower than real-time constraints. We demonstrate the full functionality of our pipeline by successfully recovering giant pulses from PSR B0531+21 in a real-time setting as well as from FRB20240114 A in an offline setting. This study highlights the promise of future real-time deep-learning-accelerated radio astronomy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".