ABR is pleased to announce the release of the Nengo Loihi backend version 0.4.0 today.
This software package for Nengo allows standard spiking Nengo models to run on Intel's Loihi neuromorphic chip.
Nengo Loihi is still under heavy development, but this latest release should improve the speed and accuracy of most models
running on the chip. It also has many new features. Two new examples demonstrate some of the novel uses supported by the backend:
https://www.nengo.ai/nengo-loihi/examples/adaptive_motor_control.html Nonlinear adaptive control example
https://www.nengo.ai/nengo-loihi/examples/mnist_convnet.html MNIST convolutional network example
Version 0.4.0 also includes a chip simulator which allows you to test for functionality and the ability of models to fit on the chip,
purely in software, which can prove useful for initial debugging (warning: the performance might not be identical on the actual hardware).
Finally, we've updated the https://www.nengo.ai/nengo-loihi/ documentation, and made it 'prettier' .
For more details see: https://appliedbrainresearch.com/2018/12/06/nengo-loihi-04-released/https://appliedbrainresearch.com/2018/12/06/nengo-loihi-04-released/
Sincerely,
The ABR Nengo Loihi team
Intel PR just released this News Byte announcing the formal launch of INRC over the past few months, including some early exciting results.
https://newsroom.intel.com/news/intel-announces-neuromorphic-computing-research-collaborators/
Be sure to check out Chris Eliasmith & team's https://arxiv.org/abs/1812.01739 hot-off-the-press arxiv paper showing Loihi giving leading batchsize=1 DNN inference energy efficiency compared to all mainstream architectures. They evaluated a range of DNNs mapped to Loihi with their Nengo DL tool.