Using machine learning to improve neutron identification in water Cherenkov detectors
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Water Cherenkov detectors like Super-Kamiokande, and the next generation Hyper-Kamiokande are adding gadolinium to their water to improve the detection of neutrons. By detecting neutrons in addition to the leptons in neutrino interactions, an improved separation between neutrino and anti-neutrinos, and reduced backgrounds for proton decay searches can be expected. The neutron signal itself is still small and can be confused with muon spallation and other background sources. In this paper, machine learning techniques are employed to optimize the neutron capture detection capability in the new intermediate water Cherenkov detector (IWCD) for Hyper-K. In particular, boosted decision tree (XGBoost), graph convolutional network (GCN), and dynamic graph convolutional neural network (DGCNN) models are developed and benchmarked against a statistical likelihood-based approach, achieving up to a 10% increase in classification accuracy. Characteristic features are also engineered from the datasets and analyzed using SHAP (SHapley Additive exPlanations) to provide insight into the pivotal factors influencing event type outcomes. The dataset used in this research consisted of roughly 1.6 million simulated particle gun events, divided nearly evenly between neutron capture and a background electron source. The current samples used for training are representative only, and more realistic samples will need to be made for the analyses of real data. The current class split is 50/50, but there is expected to be a difference between the classes in the real experiment, and one might consider using resampling techniques to address the issue of serious imbalances in the class distribution in real data if necessary.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 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 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it