Artificial Intelligence

Twister2 has stepped into accelerating the research and development of artificial intelligent technoglogies As the first step we have release an experimental version of Twister2 high performance deep learning library as Twister2:DeepNet. Twister2 is currently working on providing the state of the art support for artificial intelligent application development. Our Objective is to provide seamless APIs for data pre-processing, data processing, AI algorithm training, testing and inference. Our AI framework will be providing support to machine learning, deep learning and reinforcement learning application development.

Machine Learning

Currently we support KMeans and SVM as in-built algorithms in batch and streaming.

Deep Learning

We currently support an experimental version of Twister2:DeepNet (our distributed deep learning runtime). In the current version we support Pytorch distributed training with an MPI backend. We are actively working on providing this support for Tensorflow and MXNet.

Reinforcement Learning

This will be supported in an upcoming release with Pytorch and Tensorflow.