The Thirty-third annual conference on Neural Information Processing Systems (NeurIPS) will be held this Sunday 08 December through Saturday 14, 2019 at Vancouver Convention Center. Dr Neill Campbell will present the new paper ‘Fixing Implicit Derivatives: Trust-Region Based Learning of Continuous Energy Functions‘ collaborating with Matteo Toso CVSSP, University of Surrey and Dr Chris Russell CVSSP, University of Surrey and The Alan Turing Institute.

Neural Information Processing Systems is the largest annual conference in the field of Machine Learning looking at all aspects of research from mathematical and theoretical to applied technology including biological systems. In this collaboration with researchers at CVSSP, we have been looking at a new technique for combining established and reliable energy based models with new deep learning approaches to get the best of both worlds. For example, we can combine a number of classical tracking algorithms into a single continuous energy function and train the resulting system in an end-to-end fashion to outperform any of the individual trackers. More formally, we refer to this technique for the learning of continuous energy functions that we refer to as Wibergian Learning due to its parallels with the Wiberg optimisation method. One common approach to inverse problems is to cast them as an energy minimisation problem, where the minimum cost solution found is used as an estimator of hidden parameters. Our new approach formally characterises the dependency between weights that control the shape of the energy function, and the location of minima, by describing minima as fixed points of optimisation methods. This allows for the use of gradient-based end-to- end training to integrate deep-learning and the classical inverse problem methods. We show how our approach can be applied to obtain state-of-the-art results in the diverse applications of tracker fusion and multi-view 3D reconstruction.