Composite acquisition and processing system based on photoelectric neural network
A neural network, acquisition and processing technology, applied in biological neural network models, neural learning methods, etc., can solve the increased time and energy consumption of sensor array serial data transmission, cannot meet the rapid growth of data processing, and data transmission cannot Completely parallel and other problems, to achieve the effect of being conducive to separating excitons, minimizing hardware consumption, and high photoelectric conversion efficiency
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0042] An efficient and low-cost photonic neural network (PNN) structure such as figure 1 Shown: From left to right are input layer, hidden layer and output layer. The system is a two-layer neuromorphic network with 784, 30 and 10 nodes in each layer. 784 × 30 photoelectric synapses were arranged between the input layer and the hidden layer, and 30 × 10 normal synapses were arranged between the hidden layer and the output layer.
[0043] The specific implementation method of embodiment 1 is as follows:
[0044] Two identical devices are grouped into one opto-synapse, that is, 30 opto-synapses sharing the same input neuron are grouped to receive the same light signal from an image pixel; 784 opto-synapses connected to the same hidden node share the same optical signal from the network of the same BP signal. The input signal is set to binary. A certain amount of synaptic weight change occurs only when the device is in light conditions and the BP signal is in the 'on' mode; ...
Embodiment 2
[0046] An efficient and low-cost photonic neural network (PNN) structure such as figure 1 Shown: From left to right are input layer, hidden layer and output layer. The system is a two-layer neuromorphic network with 784, 30 and 10 nodes in each layer. 784 × 30 photoelectric synapses were arranged between the input layer and the hidden layer, and 30 × 10 normal synapses were arranged between the hidden layer and the output layer. Develop an improved Widrow-Hoff algorithm.
[0047] The concrete algorithm of embodiment 2 is as follows:
[0048]
[0049] in, wxya ki is the weight change in the input layer, V is the input signal (only 1 and 0 here), is the BP signal; where, fg (n) is the target output, f (n) is the actual output, h ( i ) is the hidden layer output, W ij is the synaptic weight in the output layer. The threshold was set at 0.004G (G is the device conductivity).
[0050] The specific algorithm flow is as follows:
[0051] The first step is to train...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


