TDC2019 Intel Software Day - Tecnicas de Programacao Paralela em Machine Lear...tdc-globalcode
Mais conteúdo relacionado
Semelhante a TDC2017 | POA Trilha Machine Learning - Portfólio Intel para Inteligência Artificial: Usando Python e Tensorflow em Machine e Deep Learning
Semelhante a TDC2017 | POA Trilha Machine Learning - Portfólio Intel para Inteligência Artificial: Usando Python e Tensorflow em Machine e Deep Learning (20)
2. 2
Agenda
Objective
Road to high performance
Optimize Frameworks through libraries
Case: Tensorflow on Keras API
3. 3
• Show how to optimize Python´s AI Frameworks
• Present Intel Math Kernel Library (MKL)
• Show how it is done with Tensorflow
• Aware how important is to configure before running
Objective
4. 4
• Today, we make software like a production line;
• The new wave: Intelligent Systems >> More Software!
• This Software uses Math, a lot of Math;
• We need hardware to execute calculations more faster!
Road to high performance
5. 5
• We have free access to several Frameworks to start developing
• Python dominates! (https://spectrum.ieee.org/computing/software/the-
2017-top-programming-languages)
• Question: How about:
• Vectorization?
• Multithreading?
• Distributed computing?
Road to high performance
10. 10
Case: Tensorflow on Keras
• Running a low workload for neural network
• Running the script unaware of threads used
• Adjustments on thread usage by script
• New results
11. 11
Conclusion
• When you develop for AI, you should to consider hardware
performance
• You do not need to refactor your code, you can signal
hardware before execution
• If you seek for more results, try using libs focused on
performance