Dharmendra S. Modha

Guest Blog by Arnon Amir, Brian Taba, and Timothy Melano

The Conference on Computer Vision and Pattern Recognition (CVPR) is widely considered as the preeminent conference for computer vision. This year the IBM Brain Inspired Computing team had the pleasure of demonstrating our latest technology at the CVPR 2016 Industry Expo, held in the air-conditioned conference halls of Caesars Palace, Las Vegas. The expo was co-located with academic poster presentations, which created an excellent opportunity for us to not only meet very interesting academics, but also to see the latest demos from other amazing companies, both large and small.

We too were excited to demonstrate our new Runtime API for TrueNorth. To showcase it, we connected an event-based vision sensor, the DVS128 (made by iniLabs), over USB to our NS1e board.

We used our Eedn framework to train a convolutional neural network on hand and arm gestures collected from our team, including air-drums and air-guitar! This Eedn network was used to configure the TrueNorth chip on the NS1e board. Overall, the system received asynchronous pixel events from the DVS128 sensor and passed them to TrueNorth. A new classification was produced every one millisecond, or at 1000 classifications per second.

The reaction to the real-time gesture classifications was very positive and drew large crowds (and other hardware vendors ;). People were blown away by that fact that we were running a convnet in real-time at 1000 classifications per second while consuming only milliwatts of power. We invited anyone who was interested to come behind our table to play with the gesture recognition. With a little bit of adjustment, people were able to interact with TrueNorth and have their gestures recognized. To many in the audience, the entire concept of neuromorphic engineering was new. Their visit to our booth was a great opportunity to introduce them to the DVS128, a spiking sensor inspired by the human retina, and TrueNorth, a spiking neural network chip inspired by the human brain!

A video can be seen here.

Previously, we have demonstrated that TrueNorth can perform greater than 1000 classifications per second on benchmark datasets. Therefore, the new Runtime API opens the interface to the NS1e board and the TrueNorth chip for many exciting real-time applications, processing complex data at very fast rates, yet consuming very low power.

We give special thanks to our teammates David Berg, Carmelo di Nolfo and Michael Debole for leading efforts to develop the Runtime API, to Jeff Mckinstry for performing the Eedn training, to Guillaume Garreau for his help with data preparation, and to the entire Brain Inspired Computing team for volunteering to create the training data set!

Guest Blog by Rathinakumar Appuswamy

To seek feedback from fellow scientists, my colleagues and I are very excited to share a preprint with the community.

Title: Structured Convolution Matrices for Energy-efficient Deep learning

Authors: Rathinakumar Appuswamy, Tapan Nayak, John Arthur, Steven Esser, Paul Merolla, Jeffrey Mckinstry, Timothy Melano, Myron Flickner, Dharmendra S. Modha 

Extended Abstract: We derive a relationship between network representation in energy-efficient neuromorphic architectures and block Toplitz convolutional matrices. Inspired by this connection, we develop deep convolutional networks using a family of structured convolutional matrices and achieve state-of-the-art trade-off between energy efficiency and classification accuracy for well-known image recognition tasks. We also put forward a novel method to train binary convolutional networks by utilising an existing connection between noisy-rectified linear units and binary activations. We report a novel approach to train deep convolutional networks with structured kernels. Specifically, all the convolution kernels are generated by the commutative pairs of elements from the Symmetric group S4. This particular structure is inspired by the TrueNorth architecture and we use it to achieve an improved accuracy vs energy tradeoff than we had previously reported. Our work builds on the growing body of literature devoted to developing convolutional networks for low-precision hardware toward energy-efficient deep learning.

Link: http://arxiv.org/abs/1606.02407

Guest Blog by Paul A. Merolla

To seek feedback from fellow scientists, my colleagues and I are very excited to share a preprint with the community.

Title: Deep neural networks are robust to weight binarization and other non-linear distortions

Authors: Paul A. Merolla, Rathinakumar Appuswamy, John V. Arthur, Steve K. Esser, Dharmendra S. Modha 

Abstract: Recent results show that deep neural networks achieve excellent performance even when, during training, weights are quantized and projected to a binary representation. Here, we show that this is just the tip of the iceberg: these same networks, during testing, also exhibit a remarkable robustness to distortions beyond quantization, including additive and multiplicative noise, and a class of non-linear projections where binarization is just a special case. To quantify this robustness, we show that one such network achieves 11% test error on CIFAR-10 even with 0.68 effective bits per weight. Furthermore, we find that a common training heuristic–namely, projecting quantized weights during backpropagation–can be altered (or even removed) and networks still achieve a base level of robustness during testing. Specifically, training with weight projections other than quantization also works, as does simply clipping the weights, both of which have never been reported before. We confirm our results for CIFAR-10 and ImageNet datasets. Finally, drawing from these ideas, we propose a stochastic projection rule that leads to a new state of the art network with 7.64% test error on CIFAR-10 using no data augmentation.

Link: http://arxiv.org/abs/1606.01981