The human brain is exceptionally efficient. Machines are nowhere near the energy consumption of our own organic computers. As computer or data scientists we are complicit in this waste. We build systems regularly underutilising the GPU capacity and very few of us care about CPU utilisation. Using experience from computing in the 1980s and 1990s, we’ll look at how a resource-centric approach can give faster, smaller, and more efficient classifiers, leading to easier productionising of AI.
Required audience experience
Developer experience in pipeline or machine learning will allow greatest benefit
Objective of the talk
Understanding of inefficiencies in AI and development and deployment, ideas for changing approach to AI R&D to improve efficiency and speed. Interfacing with non data scientists to bridge the skills shortage and add pipeline resilience.
