Accelerating Machine Learning

Machine Learning in form of Deep Convolutional Neural Networks has demonstrated significant advantages over classical algorithms, however at the cost of a significant compute burdon. MLE has started putting together a set of accelerated platforms, solutions and focused services within the Xilinx FPGA ecosystem.

Currently, these accelerated platforms, solutions and services focus on the Inference phase of Deep-Learning. MLE’s acceleration techniques for Deep Convolutional Neural Network Inference combine “unconventional” dataflow-oriented architectures with modern design flows using Xilinx High-Level Synthesis, and the Xilinx SDx tool chain.

Close collaboration with re-knowned experts from the Bavarian Multi-Media Lab at Augsburg University, Germany, facilitates rapid adoption of recent research results, for example, in the field of Reduced Precision Neural Networks.

MLE is a licensee of Xilinx and offers sub-licensing, technology support and complementary design services for integrating Accelerated Deep-Learning Inference into your application. When applied to the Deep-Learning Inference phase, Xilinx FPGA technology can provide a unique combination of low-latency response times, high compute performance in the Tera-OPS range, at very low Wattage.