PROWLER.io is proud to announce that the following four papers have been accepted for presentation at
NIPS 2017: The Thirty-first Annual Conference on Neural Information Processing Systems
This year NIPS received a record-breaking 3240 submissions, of which 678 were selected for acceptance (among them 40 as 20-minute orals and 112 as 2-minute spotlights).
Identification of Gaussian Process State Space Models
Stefanos Eleftheriadis, Thomas F.W. Nicholson, Marc Peter Deisenroth, James Hensman (all PROWLER.io)
The Gaussian process state space model (GPSSM) is a powerful model of nonlinear dynamical systems. We present a practical approach to inference in these systems using a recurrent neural network as a recognition model.
Active Exploration for Learning Symbolic Representations
Garrett Andersen (PROWLER.io), George Konidaris (Brown)
We introduce an online active exploration algorithm for data-efficiently learning an abstract symbolic model of an environment. Our agent uses a Bayesian symbolic model to guide its exploration towards regions of the state space that the model is uncertain about.
Convolutional Gaussian Processes
Mark van der Wilk (Cambridge), Carl E Rasmussen (PROWLER.io and Cambridge), James Hensman (PROWLER.io)
Convolution is a key component of machine learning tools for image recognition. Here we construct a Gaussian process model with convolutional properties and derive an effective inference procedure.
Doubly Stochastic Variational Inference for Deep Gaussian Processes
Hugh Salimbeni (PROWLER.io and Imperial College), Marc Deisenroth (PROWLER.io and Imperial College)
We present an algorithm for inference in deep Gaussian processes. The method requires fewer factorisation assumptions than previous methods and works very well in practice.