y0news
← Feed
Back to feed
🧠 AI NeutralImportance 4/10

Learning Tangent Bundles and Characteristic Classes with Autoencoder Atlases

arXiv – CS AI|Eduardo Paluzo-Hidalgo, Yuichi Ike||6 views
🤖AI Summary

Researchers introduce a theoretical framework connecting multi-chart autoencoders in manifold learning with classical vector bundle theory and characteristic classes. The approach treats collections of locally trained encoder-decoder pairs as learned atlases on manifolds, enabling computation of differential-topological invariants and providing algorithmic criteria for detecting orientability.

Key Takeaways
  • Multi-chart autoencoders can be viewed as learned atlases on manifolds rather than single global Euclidean embeddings.
  • Reconstruction-consistent autoencoder atlases canonically define transition maps that satisfy the cocycle condition.
  • The first Stiefel-Whitney class can be computed from signs of Jacobians of learned transition maps to detect orientability.
  • Non-trivial characteristic classes provide obstructions to single-chart representations in autoencoder architectures.
  • The methodology successfully applies to both low-dimensional manifolds and high-dimensional non-orientable image datasets.
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Connect Wallet to AI →How it works
Related Articles