Galaxy mergers in UNIONS – I. A simulation-driven hybrid deep learning ensemble for pure galaxy merger classification
Bibliographic record
Abstract
ABSTRACT Merging and interactions can radically transform galaxies. However, identifying these events based solely on structure is challenging as the status of observed mergers is not easily accessible. Fortunately, cosmological simulations are now able to produce more realistic galaxy morphologies, allowing us to directly trace galaxy transformation throughout the merger sequence. To advance the potential of observational analysis closer to what is possible in simulations, we introduce a supervised deep learning convolutional neural network and vision transformer hybrid framework, Mummi (MUlti Model Merger Identifier). Mummi is trained on realism-added synthetic data from IllustrisTNG100-1, and is comprised of a multistep ensemble of models to identify mergers and non-mergers, and to subsequently classify the mergers as interacting pairs or post-mergers. To train this ensemble of models, we generate a large imaging data set of 6.4 million images targeting UNIONS with RealSimCFIS. We show that Mummi offers a significant improvement over many previous machine learning classifiers, achieving 95 per cent pure classifications even at Gyr long time-scales when using a jury-based decision-making process, mitigating class imbalance issues that arise when identifying real galaxy mergers from $z=0$ to 0.3. Additionally, we can divide the identified mergers into pairs and post-mergers at 96 per cent success rate. We drastically decrease the false positive rate in galaxy merger samples by 75 per cent. By applying Mummi to the UNIONS DR5-SDSS DR7 overlap, we report a catalogue of 13 448 high-confidence galaxy merger candidates. Finally, we demonstrate that Mummi produces powerful representations solely using supervised learning, which can be used to bridge galaxy morphologies in simulations and observations.
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 | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 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 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".