Boolan machine learning summit
Download as PPTX, PDF1 like1,828 views
This document summarizes GPUs as complementing rather than replacing the Hadoop big data stack. It notes that wholesale replacement of the big data stack would be cost-prohibitive for many clients. The best approach is to sell GPUs for accelerated computation and a few other use cases to gain a foothold in the ecosystem. The functionality of Volta GPUs may change this ecosystem over time.
Boolan machine learning summit
- 2. Who are we? This slide shows that GPUs should complement the big data stack on the Hadoop ecosystem, rather than trying to replace Hadoop etc. outright. Wholesale replacement of the big data stack will be cost-prohibitive to many clients. We believe the right approach is to sell GPUs for accelerated computation and a few other use cases. That’s our beach head. (Obviously, the widening functionality of the Volta will change the GPU ecosystem.) Founded 2014 Distributed worldwide Lots of activity in china
- 4. Most deep learning frameworks ● Python based ● Symbolic variables with automatic differentiation in a computation graph ● Low level, focus on flexibility ● Supplemented by the ecosystem in python ● Run on cpu and gpu. Usually roll own distributed compute ● Big emphasis on research
- 5. Dl4j
- 7. Connectors via datavec (data types) Supported Languages: Mandarin English Japanese Korean
- 11. UI
- 12. Nd4j Parameter Server Aeron: More stable latency than GRPC and way faster (25x!) than TF
- 13. Python support PyJNIus Jumpy (no copy numpy nd4j integration!)
- 14. Pretrained models/Transfer Learning Model Zoo Keras model import Explicit TL API
- 15. Upcoming Nd4j Autodiff (yes it looks like what you’d expect) Nd4j sparse arrays Pipeline api