IC Colloquium: Programming Languages for Machine Learning and Perception
Event details
| Date | 21.01.2019 |
| Hour | 14:15 › 15:30 |
| Location | |
| Category | Conferences - Seminars |
By: Andrew Fitzgibbon - Microsoft (Cambridge)
Video of his talk
Abstract:
In 1957, many (most?) computer programs were the expression of mathematical formulae in machine instructions. It was tedious, so Backus and colleagues wrote the “FORmula TRANslator”, perhaps the first modern compiler. FORTRAN revolutionised the writing of numerical code and the richness of tasks the computer could perform. Today, we see another resurgence in numerical programming, in the domain of machine learning, computer vision, speech processing and other areas of “AI”. Such code needs to run fast on large datasets for “training”, or to run very efficiently on small machines for “inference”. To serve this need, a variety of frankly clunky domain-specific tools such as TensorFlow and PyTorch have emerged. Like FORTRAN-1, they are incredibly useful tools, but there is room for improvement. I will talk about several potential improvements, in languages from Julia to C++ to F# to Haskell, covering with some rather surprising benchmarks of algorithmic differentiation tools, to compilation of functional programs to non-garbage-collected runtimes, to a new view of DSLs inspired by Julia’s slogan “it’s just code”. By way of introduction I will develop some of the key ideas in nonlinear optimization as it applies to AI and computer vision, at a level which I plan to be accessible to all.
Bio:
Fitzgibbon is a partner scientist at Microsoft in Cambridge, UK. He has published numerous highly-cited papers, and received many awards for his work, including ten “best paper” prizes at various venues, the Silver medal of the Royal Academy of Engineering, and the BCS Roger Needham award. He is a fellow of the Royal Academy of Engineering, the British Computer Society, and the International Association for Pattern Recognition. He studied at University College, Cork, and then did a Masters at Heriot-Watt University, before taking up an RSE job at the University of Edinburgh, which eventually morphed into a PhD. He moved to Oxford in 1996 and drove large software projects such as the VXL project, and then spent several years as a Royal Society University Research Fellow before joining Microsoft in 2005. He loves programming, particularly in C++, and his recent work has included new numerical algorithms for Eigen, and compilation of F# to a non-garbage-collected runtime.
More information
Video of his talk
Abstract:
In 1957, many (most?) computer programs were the expression of mathematical formulae in machine instructions. It was tedious, so Backus and colleagues wrote the “FORmula TRANslator”, perhaps the first modern compiler. FORTRAN revolutionised the writing of numerical code and the richness of tasks the computer could perform. Today, we see another resurgence in numerical programming, in the domain of machine learning, computer vision, speech processing and other areas of “AI”. Such code needs to run fast on large datasets for “training”, or to run very efficiently on small machines for “inference”. To serve this need, a variety of frankly clunky domain-specific tools such as TensorFlow and PyTorch have emerged. Like FORTRAN-1, they are incredibly useful tools, but there is room for improvement. I will talk about several potential improvements, in languages from Julia to C++ to F# to Haskell, covering with some rather surprising benchmarks of algorithmic differentiation tools, to compilation of functional programs to non-garbage-collected runtimes, to a new view of DSLs inspired by Julia’s slogan “it’s just code”. By way of introduction I will develop some of the key ideas in nonlinear optimization as it applies to AI and computer vision, at a level which I plan to be accessible to all.
Bio:
Fitzgibbon is a partner scientist at Microsoft in Cambridge, UK. He has published numerous highly-cited papers, and received many awards for his work, including ten “best paper” prizes at various venues, the Silver medal of the Royal Academy of Engineering, and the BCS Roger Needham award. He is a fellow of the Royal Academy of Engineering, the British Computer Society, and the International Association for Pattern Recognition. He studied at University College, Cork, and then did a Masters at Heriot-Watt University, before taking up an RSE job at the University of Edinburgh, which eventually morphed into a PhD. He moved to Oxford in 1996 and drove large software projects such as the VXL project, and then spent several years as a Royal Society University Research Fellow before joining Microsoft in 2005. He loves programming, particularly in C++, and his recent work has included new numerical algorithms for Eigen, and compilation of F# to a non-garbage-collected runtime.
More information
Practical information
- General public
- Free
Contact
- Host: Christoph Koch