Improved Learning of One-hidden-layer Convolutional Neural Networks with Overlaps
This work addresses the challenge of provable algorithm design for learning neural networks, potentially inspiring further development in complex models, but it appears incremental as it builds on existing techniques without claiming broad SOTA results.
The authors tackled the problem of learning one-hidden-layer convolutional neural networks with overlapping patches, proposing a new algorithm that works for a general class of structures, drawing on isotonic regression and landscape analysis of non-convex matrix factorization.
We propose a new algorithm to learn a one-hidden-layer convolutional neural network where both the convolutional weights and the outputs weights are parameters to be learned. Our algorithm works for a general class of (potentially overlapping) patches, including commonly used structures for computer vision tasks. Our algorithm draws ideas from (1) isotonic regression for learning neural networks and (2) landscape analysis of non-convex matrix factorization problems. We believe these findings may inspire further development in designing provable algorithms for learning neural networks and other complex models.