The paper introduces a novel SO(3)-equivariant convolutional neural network designed for data on the sphere, specifically using the Hierarchical Equal Area isoLatitude Pixelization (HEALPix). By employing steerable filters, parallel transport, and spin-weighted features, the architecture guarantees exact equivariance and avoids the need for the network to learn spherical symmetry from data. Benchmarks show the model achieves percent-level SO(3)-invariance error and outperforms existing methods in dense gradient regression, cosmic microwave background foreground removal, and denoising tasks.
arXiv:2505. 21736v2 Announce Type: replace-cross Abstract: Translation equivariance is a central reason convolutional neural networks have been successful in computer vision.
arXiv:2607. 27320v1 Announce Type: cross Abstract: Field-level inference of cosmological initial conditions from galaxy surveys requires a forward model that is simultaneously accurate in the non-linear regime, computationally efficient, and fully differentiable.
arXiv:2603. 22006v2 Announce Type: replace-cross Abstract: Upcoming stage-IV surveys such as Euclid and Rubin will deliver vast amounts of high-precision data, opening new opportunities to constrain cosmological models with unprecedented accuracy.
arXiv:2608.27521v1 Announce Type: new Abstract: Many dynamical processes unfold on the sphere but the default scientific machine learning architectures are Euclidean. Applying these architectures on...
Graph Neural Networks (GNNs) are presented as a natural framework for modeling the relational structure inherent in astrophysical and cosmological data, such as galaxies in their environments or photons in detectors. The review argues that GNNs align with the native geometry of these domains, incorporate existing symmetries and prior knowledge, and remain flexible enough to capture unknown phenomena, offering greater interpretability than other deep‑learning models. It surveys applications across cosmology, galaxy formation, dynamics, and instrumentation, noting that only about one in six non‑instrumentation papers apply GNNs to real data rather than simulations, and provides best‑practice guidelines for when to use or avoid graph approaches.
arXiv:2606.23346v2 Announce Type: replace-cross Abstract: We perform a realistic KiDS-Legacy mock analysis with field-level neural compression and simulation-based inference using just 60 $N$-body si...
The paper introduces a new compression pipeline for scientific simulation data that combines Residual Vector Quantization (RVQ) with a U‑Net post‑processing network to correct pixel‑space residuals, followed by a Guaranteed Autoencoder (GAE) that enforces block‑wise error bounds. The U‑Net is trained to predict spatially structured residuals, addressing limitations of latent‑space only approaches. Experiments on S3D, JHTDB, and E3SM datasets show improved NRMSE and compression ratios compared to RVQ alone while maintaining strict error guarantees.
arXiv:2505. 15441v5 Announce Type: replace-cross Abstract: Natural images exhibit strong geometric regularities: local structures, such as edges, corners, and textures, appear in many orientations and mirror configurations.
The paper demonstrates that a self‑supervised Vision Transformer (ViT) pretrained on a fast, low‑cost semi‑numerical simulator can produce data summaries that transfer across different simulators without retraining. In 21cm cosmology, the ViT—named SKATR—pretrained on 67,000 21cmFAST lightcones is applied unchanged to hydrodynamical Loreli II lightcones, enabling accurate inference of five astrophysical parameters with fewer radiative‑transfer simulations than a fully‑supervised baseline. SKATR remains accurate, informative, and calibrated even under realistic SKA antenna array noise, outperforming supervised models retrained on noisy data.
arXiv:2106. 06998v5 Announce Type: replace Abstract: Training convolutional neural networks at scale demands substantial memory, largely because intermediate activations must be stored for backpropagation.
arXiv:2606. 15760v1 Announce Type: new Abstract: A significant gap exists between theory and practice in deep learning.
The paper introduces SparK-SAR, a sparse convolutional masked pretraining framework that leverages ConvNeXt for high‑resolution Sentinel‑1 GRD SAR data. It presents three key contributions: a domain‑adapted ConvNeXt‑RS encoder with a reduced stem stride, a lightweight ConvNeXt‑512 decoder that cuts VRAM usage while matching UNet accuracy, and an analysis of FP16 instability and layer‑wise learning‑rate decay benefits. Experiments on four Copernicus‑Bench S1GRD tasks show that the 27.8 M‑parameter ConvNeXt‑RS encoder outperforms a larger ViT baseline on most tasks and demonstrates strong data efficiency, with classification saturating at 10–25 % of the SSL4EO‑S12 corpus.