What domain-adaptation strategy would map dense collision centers to sparse outer-detector signatures?

This is a specialized question about high-energy physics detector analysis, and the preprints provided don't directly address mapping dense inner-detector collision signatures to sparse outer-detector observables.

That said, the transfer learning and domain-adaptation strategies in these papers offer relevant principles. The key insight across several works is that point cloud representations paired with geometry-agnostic neural architectures (transformers, flow-matching models) can learn to compress and translate between different detector configurations and data modalities.

For instance, [AG-2025.11-1571] uses point clouds and bias-only parameter-efficient fine-tuning to adapt shower simulations across detector geometries with minimal retraining—a similar sparse-to-dense or vice-versa mapping problem. The paper achieves 44% improvement over training from scratch with only 100 target-domain samples, suggesting that a pretraining phase on dense inner-detector data could transfer efficiently to sparse outer signatures.

More directly, [AG-2025.02-1472] demonstrates cross-detector transfer in particle-flow reconstruction, where a model trained on one collider geometry fine-tunes successfully on a different design with an order of magnitude fewer samples. This hints that a shared latent representation learned from dense inner collisions could be adapted downstream to outer-detector sparsity.

[AG-2024.05-1320] and [AG-2025.09-1622] both use conditional flow matching and vision transformers to handle sparse high-dimensional detector data—architectures that naturally compress and denoise, which could bridge inner-to-outer mappings.

However, none of these papers explicitly tackles the inverse problem of reconstructing or predicting outer-detector sparse signatures from inner-detector dense collision centers, so the archive does not provide a direct blueprint for your specific domain-adaptation challenge.