Spherical Inducing Features for Orthogonally-Decoupled Gaussian Processes
This work addresses a practical problem for machine learning practitioners by improving the scalability and flexibility of GPs, though it appears incremental as it builds on existing inter-domain variational GP methods.
The paper tackled the issue of Gaussian processes (GPs) lacking representation learning compared to neural networks by proposing an extension using spherical inter-domain features to enhance flexibility and scalability in GP approximations, with experiments on benchmark datasets showing its effectiveness.
Despite their many desirable properties, Gaussian processes (GPs) are often compared unfavorably to deep neural networks (NNs) for lacking the ability to learn representations. Recent efforts to bridge the gap between GPs and deep NNs have yielded a new class of inter-domain variational GPs in which the inducing variables correspond to hidden units of a feedforward NN. In this work, we examine some practical issues associated with this approach and propose an extension that leverages the orthogonal decomposition of GPs to mitigate these limitations. In particular, we introduce spherical inter-domain features to construct more flexible data-dependent basis functions for both the principal and orthogonal components of the GP approximation and show that incorporating NN activation features under this framework not only alleviates these shortcomings but is more scalable than alternative strategies. Experiments on multiple benchmark datasets demonstrate the effectiveness of our approach.