Write a Blog >>
ICFP 2019
Sun 18 - Fri 23 August 2019 Berlin, Germany
Tue 20 Aug 2019 13:30 - 13:52 at Aurora Borealis - The Real World Chair(s): Robert Atkey

Deep learning has seen tremendous success over the past decade in computer vision, machine translation, and gameplay. This success rests crucially on gradient-descent optimization and the ability to “learn” parameters of a neural network by backpropagating observed errors. However, neural network architectures are growing increasingly sophisticated and diverse, which motivates an emerging quest for even more general forms of differentiable programming, where arbitrary parameterized computations can be trained by gradient descent. In this paper, we take a fresh look at automatic differentiation (AD) techniques, and especially aim to demystify the reverse-mode form of AD that generalizes backpropagation in neural networks. We uncover a tight connection between reverse-mode AD and delimited continuations, which permits implementing reverse-mode AD purely via operator overloading and without managing any auxiliary data structures. We further show how this formulation of AD can be fruitfully combined with multi-stage programming (staging), leading to an efficient implementation that combines the performance benefits of deep learning frameworks based on explicit reified computation graphs (e.g., TensorFlow) with the expressiveness of pure library approaches (e.g., PyTorch).

Tue 20 Aug

icfp-2019-papers
13:30 - 15:00: Research Papers - The Real World at Aurora Borealis
Chair(s): Robert AtkeyUniversity of Strathclyde
icfp-2019-papers13:30 - 13:52
Talk
Fei Wang, Dan ZhengPurdue University, Google Brain, James Decker, Xilun WuPurdue University, Gregory EssertelPurdue University, Tiark RompfPurdue University
Pre-print
icfp-2019-papers13:52 - 14:15
Talk
Amir ShaikhhaUniversity of Oxford, Andrew FitzgibbonMicrosoft Research, Cambridge, Dimitrios VytiniotisDeepMind, Simon Peyton JonesMicrosoft, UK
icfp-2019-papers14:15 - 14:37
Talk
Rajan WaliaIndiana University, Praveen NarayananIndiana University, USA, Jacques CaretteMcMaster University, Sam Tobin-HochstadtIndiana University, Chung-chieh ShanIndiana University, USA
Pre-print
icfp-2019-papers14:37 - 15:00
Talk
Benjamin ShermanMassachusetts Institute of Technology, USA, Jesse MichelMassachusetts Institute of Technology, Michael CarbinMassachusetts Institute of Technology
DOI Pre-print Media Attached