Optimization Encoders: Rethinking Second-Order Meta-Learning for Neural Fields
MetaLF frames latent optimization as an encoder for second-order meta-learning of conditional neural fields.
The paper formalizes latent optimization in meta-learned neural fields as an optimization encoder, linking second-order differentiation, latent parameterization, and task supervision. Attentive Latent Fields (MetaLF) is an equivariant transformer that contextualizes a latent point cloud with self-attention so local observations shape non-local structure. It unifies reconstruction, classification, and segmentation, using reconstruction-only adaptation at test time, and improves image and 3D fidelity within three to five gradient updates.
- Treats latent gradient updates as an optimization encoder
- MetaLF uses an equivariant transformer over a latent point cloud
- Unifies reconstruction, classification, and segmentation
- Improves fidelity within three to five gradient updates
- AI models
- MetaLF
From the original article
1 min read at arxiv.orgConditional neural fields represent signals continuously, but their effectiveness depends on how the conditional latent representations are inferred from observed data. In meta-learning, this encoding occurs through gradient updates induced by the decoder, tying representation learning directly to decoder design.
By Rudolf L. M. van Herten · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.