Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces an unsupervised framework for recovering latent domains and signals by leveraging symmetry discovery in data distributions, addressing the limitations of traditional blind deconvolution methods that fail with general linear transformations. By modeling observations as linear measurements from a latent random field, the authors optimize a shallow group-convolutional network with stationarity and locality regularization, enabling the extraction of latent signals from unstructured observations. Experimental results across various domains, including stochastic processes and neural recordings, demonstrate the effectiveness of this approach, highlighting symmetry discovery as a promising avenue for unsupervised structure learning in blind inverse problems.
Discovering symmetries in data can unlock hidden structures in unstructured observations, revolutionizing how we tackle blind inverse problems.
Primary motivation in blind inverse problems is to recover signals of interest from corrupted observations without knowing the obfuscating mechanism. Blind deconvolution is a prominent approach when the corruption is convolutional, but it is not applicable when general linear transformations obfuscate the domain structure. In this work, we propose an unsupervised framework for recovering latent domains and signals by discovering symmetries of the data distribution. Our framework models observations as linear measurements of signals sampled from a latent random field, and optimizes a shallow group-convolutional network by imposing stationarity and locality regularization at the model output. The model learns a latent symmetry action and an appropriate filter, thereby mapping unstructured observations to a symmetry-based representation that reveals latent signals. Experiments on stochastic processes, Ising models, shuffled and bit-scrambled images, and neural recordings show that the method recovers latent domains and signals from unstructured observations, suggesting symmetry discovery as a new direction for unsupervised structure learning and blind inverse problems.