Insights on Galaxy Evolution from Interpretable Sparse Feature Networks

TL;DR

This paper addresses the interpretability challenge of deep networks applied to galaxy morphology by introducing Sparse Feature Network (SFNet), which enforces a $k$-sparse constraint so predictions are a linear combination of a small set of interpretable, pixel-localized features. The method is evaluated on ~250k SDSS galaxies, predicting optical emission line fluxes and gas-phase metallicity while preserving competitive accuracy. The learned sparse features align with physical patterns in the BPT diagram, enabling direct mapping from morphology-derived features to ionization and metallicity regimes. This interpretable approach facilitates physical pattern discovery in large imaging datasets and provides astronomers with a scalable tool for linking galaxy appearance to evolutionary physics.

Abstract

Galaxy appearances reveal the physics of how they formed and evolved. Machine learning models can now exploit galaxies' information-rich morphologies to predict physical properties directly from image cutouts. Learning the relationship between pixel-level features and galaxy properties is essential for building a physical understanding of galaxy evolution, but we are still unable to explicate the details of how deep neural networks represent image features. To address this lack of interpretability, we present a novel neural network architecture called a Sparse Feature Network (SFNet). SFNets produce interpretable features that can be linearly combined in order to estimate galaxy properties like optical emission line ratios or gas-phase metallicity. We find that SFNets do not sacrifice accuracy in order to gain interpretability, and that they perform comparably well to cutting-edge models on astronomical machine learning tasks. Our novel approach is valuable for finding physical patterns in large datasets and helping astronomers interpret machine learning results.

Figures (2)

• Figure 1: Our SFNet architecture, which resembles a resnet18 modified with a penultimate "Top-k" layer to ensure interpretable, sparse image features. A 3x3 conv, 128, /2 block denotes a $3\times 3$ 2D convolution layer with 128 channels that downsamples the image size by a factor of 2. Each convolution layer is followed by batch normalization and then a ReLU activation function. Arrows show how inputs flow through the network, and when two inputs are sent as inputs to the same layer, they are concatenated together (i.e. a residual connection).
• Figure 2: SFNet learned features when trained to predict optical emission lines. We show how each feature correlates with optical line ratios (BPT diagrams; left) and examples of the top nine image activations (right). For the BPT diagrams, colors denote scaled activation strength, ranging from black (zero) to dark magenta (low) to bright yellow (high).