More accuracy with less precision - from climate change to the human brain
- Speaker: Prof Tim Palmer (Oxford University)
- Host: Steve Furber
- 19th October 2016 at 14:00 in Kilburn L.T. 1.5
Many systems are so complex that they will never be simulated, at least in the foreseeable future, with any degree of completeness. This raises important questions when attempting to simulate a particular system of interest. For example, is it better to solve a more complete computational representation of the system less precisely, or a less complete computational representation of the system more precisely? More generally, for a given physical energy resource, what is the optimal computational mix of energy-intensive precision and determinism on the one hand, and low-energy imprecision and stochasticity on the other, that will produce the most accurate simulations of the system under study? I will spend most of the presentation illustrating these issues by considering the problem of climate simulation, though will talk more speculatively about the human brain towards the end of the talk. A purpose of this talk is to establish some dialogue between scientists who attempt to simulate complex nonlinear systems such as climate, and scientists developing next generation computer technology.