Researchers have introduced XFactors, a novel weakly-supervised variational auto-encoder framework designed for disentangled representation learning. This method decomposes representations into specific factor subspaces and a residual subspace, utilizing contrastive supervision with an InfoNCE loss to align target factors. KL regularization organizes the geometry of non-targeted factors without additional supervision, avoiding adversarial objectives and auxiliary classifiers. XFactors has demonstrated state-of-the-art disentanglement scores across various datasets, including CelebA, and enables controlled factor swapping through latent replacement. AI
IMPACT This research could lead to more interpretable and controllable AI models by improving how they learn and represent underlying data factors.
RANK_REASON This is a research paper detailing a new method for disentangled representation learning. [lever_c_demoted from research: ic=1 ai=1.0]
- Alexandre Myara
- Celeba
- Disentangled Information Bottleneck
- Gaussian function
- InfoNCE loss
- KL regularization
- variational auto-encoder
- XFactors
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
