From Boltzmann Machines to Neural Networks and Back Again
arXiv:2007.12815
Abstract
Graphical models are powerful tools for modeling high-dimensional data, but learning graphical models in the presence of latent variables is well-known to be difficult. In this work we give new results for learning Restricted Boltzmann Machines, probably the most well-studied class of latent variable models. Our results are based on new connections to learning two-layer neural networks under bounded input; for both problems, we give nearly optimal results under the conjectured hardness of sparse parity with noise. Using the connection between RBMs and feedforward networks, we also initiate the theoretical study of [Hinton, 2012], a version of neural-network learning that couples distributional assumptions induced from the underlying graphical model with the architecture of the unknown function class. We then give an algorithm for learning a natural class of supervised RBMs with better runtime than what is possible for its related class of networks without distributional assumptions.
References in corpus (12)
- Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
- Statistical mechanics for natural flocks of birds
- Latent variable graphical model selection via convex optimization
- Inapproximability of the Partition Function for the Antiferromagnetic Ising and Hard-Core Models
- Globally Optimal Gradient Descent for a ConvNet with Gaussian Inputs
- Kernel and Rich Regimes in Overparametrized Models
- Reliably Learning the ReLU in Polynomial Time
- Interaction Screening: Efficient and Sample-Optimal Learning of Ising Models
- A Provably Correct Algorithm for Deep Learning that Actually Works
- Training Restricted Boltzmann Machines via the Thouless-Anderson-Palmer Free Energy
- Deep Learning and Hierarchal Generative Models
- Learning Two-layer Neural Networks with Symmetric Inputs