PyTorch NN Integration (Deep Kernel Learning)

Because QPyTorch is built on top of PyTorch, you can seamlessly integrate existing PyTorch modules into QPyTorch models. This makes it possible to combine neural networks with QEPs, either with exact or approximate inference.

Here we provide some examples of Deep Kernel Learning, which are QEP models that use kernels parameterized by neural networks.

• Exact inference QEP + NN: see the Exact DKL (deep kernel learning) example, based on Wilson et al., 2015.

• Approximate inference QEP + NN: see the CIFAR-10 Classification SVDKL (stochastic variational deep kernel learning) example, based on Wilson et al., 2016.