Get Started
Research questionCan autoencoder parameter spectra serve as usable representations of their training data?The statistical properties of training data may be reflected in the singular values of an autoencoder’s parameter matrices. The difficulty is determining whether this information remains sufficiently distinctive and usable as a data representation.
Machine Learning
Neural and Evolutionary Computing
Statistical Machine Learning
Latest papersRecent research connected to this question, newest first.Spectral characteristics of autoencoder parameters as a vector representation of dataThe evidence concerns encoder-decoder autoencoders trained on CIFAR-10 and FashionMNIST. The proposed representations use spectral characteristics of parameter matrices; theoretical analysis relates their singular values to training-data covariance eigenvalues, while experiments test discrimination among models trained on different data subsets.research paper · Sep 3, 2026
Related questions
How can sparse autoencoder features be shared across language models without per-model retraining?How can implicit neural representations capture high-frequency detail without memorizing noise or requiring fragile tuning?How can spectral compressive imaging models adapt to unseen optical configurations without heavy computation?How can protein design models jointly model amino acid sequences and 3D structures without separate autoencoder pretraining?
Home
Topics
Search
Library