Deep learning power for low-end devices

Deep learning is one the fastest growing segments in technology today. It is increasingly being used for tasks like speech recognition, face identification, even game play. Powered by artificial neural networks, deep learning models employ machine learning algorithms that use multiple layers of non-linear processing units for feature extraction and transformation, with each layer using the output from the previous layer as input.

Avnet’s Xilinx 201808 empowers low-end device XC7Z007s with this deep learning capability. Combining Multilayer Perceptron (MLP) network topology with an extreme quantized network using Binarized Neural Networks*, it boasts a 95.8% accuracy on MNIST with a resource utilization of 91%/66% (LUT/FF). It is suitable for applications like smart cameras for surveillance and smart cameras for industrial automation. 

Features

• Deep learning capability on low-end device XC7Z007s
• Multilayer Perceptron (MLP) network topology
• Extreme quantized network using Binarized Neural Networks*
• 95.8% accuracy on MNIST
• Small Fully Connected Layer: 256 nodes
• Jupyter notebook over built-in Wifi
• 91%/66% (LUT/FF) Resource Utilization 

*Yaman, “FINN: A Framework for Fast, Scalable Binarized Neural Network Inference”