The machine-learning based data analytics to support a cloud intelligence (such as Google’s αGo) has already gone beyond the scalability of the present computing technology and architecture. The current training method is mainly based on an iterative backward propagation, which has long latency running on data servers. With the emergence of autonomous vehicles, unmanned aerial vehicle and robotics, there is a huge demand to analyze the real-time sensed data with small latency. In this talk, we will discuss an efficient online machine-learning algorithm, where the training of neuron network is realized by a fast ℓ2-norm direct-solver. An according prototyping accelerator developed in FPGA, CMOS-ASIC and memristor will be discussed for both ℓ2-norm direct-solver and matrix-vector multiplication. Its further applications in IoT sensor network and biomedical imaging will be also addressed.
|Room 501 – Room 502, 5/F, Photonics Centre, 2 Science Park West Avenue, Hong Kong Science Park, Shatin, Hong Kong