ERGO-ML: comparing IllustrisTNG and HSC galaxy images via contrastive learning
Bibliographic record
Abstract
ABSTRACT Modern cosmological hydrodynamical galaxy simulations provide tens of thousands of reasonably realistic synthetic galaxies across cosmic time. However, quantitatively assessing the level of realism of simulated universes in comparison to the real one is difficult. In this paper of the Extracting Reality from Galaxy Observables with Machine Learning series, we utilize contrastive learning to directly compare a large sample of simulated and observed galaxies based on their stellar-light images. This eliminates the need to specify summary statistics and allows to exploit the whole information content of the observations. We produce survey-realistic galaxy mock data sets resembling real Hyper Suprime-Cam (HSC) observations using the cosmological simulations TNG50 and TNG100. Our focus is on galaxies with stellar masses between 109 and 1012 M⊙ at z = 0.1–0.4. This allows us to evaluate the realism of the simulated TNG galaxies in comparison to actual HSC observations. We apply the self-supervised contrastive learning method Nearest Neighbour Contrastive Learning to the images from both simulated and observed data sets (g-, r-, i-bands). This results in a 256-dimensional representation space, encoding all relevant observable galaxy properties. First, this allows us to identify simulated galaxies that closely resemble real ones by seeking similar images in this multidimensional space. Even more powerful, we quantify the alignment between the representations of these two image sets, finding that the majority (≳ 70 per cent) of the TNG galaxies align well with observed HSC images. However, a subset of simulated galaxies with larger sizes, steeper Sérsic profiles, smaller Sérsic ellipticities, and larger asymmetries appears unrealistic. We also demonstrate the utility of our derived image representations by inferring properties of real HSC galaxies using simulated TNG galaxies as the ground truth.
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 | 0.002 | 0.005 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.002 | 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) | 0.002 | 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".