Explaining deep learning of galaxy morphology with saliency mapping
Problem
Measuring galactic bar lengths at scale requires automation beyond human visual classification of galaxy morphology.
Why it matters
Automated measurement enables analysis of large astronomical data sets beyond what human classification can cover.
What the AI did
CNNs trained on Galaxy Zoo human classifications predict galaxy morphology, and SmoothGrad saliency mapping extracts the bar, which a bespoke algorithm then measures; contrasted with a CNN directly predicting bar lengths.
Results
Correlation with human measurements of 0.76 (XAI method) vs 0.59 (direct CNN); RMSE 1.69 vs 2.10; XAI outperformed conventional deep learning.
Impact
XAI approach outperformed conventional deep learning for bar measurement and is suggested to extend to other galactic features such as bulge-to-disc ratio.
Record
| Venue | Monthly Notices of the Royal Astronomical Society |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Prabh Bhambra, Benjamin Joachimi, O. Lahav |
| Citations | 28 (OpenAlex, at import time) |
| References | 37 |
| Model or method | convolutional neural networks with SmoothGrad saliency mapping |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1093/mnras/stac368 |
| Open access | hybrid |
| Keywords | galaxy morphology, explainable AI, saliency mapping, CNN, Galaxy Zoo, bar length |
Importance rating
incremental / niche
Automating galactic bar length measurement extends analysis of large astronomy datasets beyond human classification; a niche research task with no stated scale beyond astronomy data processing.
More in Physics & Astronomy
| # | Paper | Year | Citations | Importance |
|---|---|---|---|---|
| 1 | Magnetic control of tokamak plasmas through deep reinforcement learning | 2022 | 739 |
7
|
| 2 | Solving the quantum many-body problem with artificial neural networks | 2017 | 2,205 |
7
|
| 3 | Learning high-accuracy error decoding for quantum processors | 2024 | 90 |
7
|
| 4 | Optimizing Quantum Error Correction Codes with Reinforcement Learning | 2019 | 158 |
7
|
| 5 | Machine-learning-assisted correction of correlated qubit errors in a topological code | 2018 | 141 |
7
|