Event-Driven Data Processing Nodes With Variable-State Outputs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial neural networks face limitations in processing event-based data from neuromorphic sensors due to reliance on constant output values and complex networks, leading to inefficiencies in computing speed, power consumption, and network traffic.
Innovation Solution
A data processing node with variable output values and internal states, capable of performing simple mathematical operations, allowing flexible configuration for various functionalities, and utilizing potential and current values to encode information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If traditional artificial neural networks use constant output values and complex network structures, then they can process event-based data from neuromorphic sensors, but computing speed is slow and power consumption is high
Solution Approach 1:
The patent segments the neural network into simplified processing nodes that handle specific operations (accumulation, comparison, spike generation) rather than using complex interconnected structures. Each node processes events independently with simple internal state variables, dividing the computational task into manageable units that execute quickly with minimal inter-node communication.
Solution Approach 2:
The patent changes the fundamental parameters of neural network operation by using variable output values that depend on internal state (accumulated event values) rather than constant outputs. This allows nodes to encode information in their output magnitude, reducing the need for complex network structures to represent diverse information states.
2Use of energy by moving object
If traditional artificial neural networks process event-based data, then they can recognize patterns, but power consumption is high due to redundant data processing
Solution Approach 1:
The patent extracts and processes only the essential features of event-based data (temporal patterns, event sequences) rather than processing all raw sensor data. By focusing computation on salient temporal characteristics rather than complete frame data, the system reduces power consumption while maintaining pattern recognition capability.
Solution Approach 2:
The patent implements continuous processing of event streams without the need to wait for complete data frames or perform redundant processing of unchanged regions. The accumulation and comparison operations continue seamlessly as events arrive, eliminating idle processing cycles and reducing overall power consumption while maintaining high processing efficiency.
3Loss of information
If frame-based acquisition is used to capture scene information, then static images can be obtained, but temporal dynamics are lost and large amounts of redundant data are acquired
Solution Approach 1:
The patent inverts the traditional approach by processing events as they occur in temporal sequence rather than organizing data into spatial frames. Instead of capturing complete images at fixed intervals and then analyzing temporal changes, the system processes individual events chronologically, preserving temporal dynamics while generating minimal data output.
4Speed
If conventional cameras with fixed frame rates are used, then illumination can be measured at regular intervals, but acquisition speed and power consumption are limited
Solution Approach 1:
The patent implements dynamic event-driven processing where computation is triggered by actual changes in the scene rather than fixed temporal intervals. The system adapts its processing rate to the dynamics of the input data, processing events only when they occur and remaining idle otherwise, thereby achieving high effective acquisition speed with minimal power consumption.
Data Source
Figure 1
Figure 2~3
Figure 4~5
AI summary
A processing node for use in a data processing engine, the node having an input connection and an output connection, wherein: the node has an associated first internal value; the node is configured to receive an input signal from the input connection, the input signal containing an input value; the node is configured to vary the first internal value in response to the input signal; and when the a triggering condition is met, the node is configured to emit an output signal containing an output value, to the output connection, the output value being based on the first internal value.