Oscillating RNN Circuit for Digital-to-Analog Fitting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional Recurrent Neural Networks (RNNs) face difficulties in outputting a fitting curve based on discrete values due to the inability to differentiate curves formed by connecting discrete points by line segments, and modifying weighting coefficients between discrete points makes the processing complex for spline curves.
Innovation Solution
A neural network designing method that forms a RNN circuit with multiple oscillating RNN circuits outputting natural oscillations and an adding circuit to compute a fitting curve by summing their outputs, allowing for digital-to-analog fitting using discrete data, which reduces computational load and time by averaging original discrete points.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Shape
If discrete values are connected by line segments to form a curve, then the curve can be visually represented, but the curve cannot be differentiated at the discrete points
Solution Approach 1:
The patent divides the curve representation into multiple segments, each defined by discrete points and their corresponding weights. By segmenting the curve into pieces that can be individually differentiated, the system achieves both visual continuity and mathematical differentiability at the discrete points.
Solution Approach 2:
The patent introduces dynamic weighting coefficients that can be adjusted to control the smoothness and differentiability properties of the curve. By making the weights variable rather than fixed, the system can adapt the curve's mathematical properties while maintaining its visual representation.
2Difficulty of detecting and measuring
If spline curves are used to connect discrete points, then the curve is smooth and differentiable, but modifying weighting coefficients between discrete points makes processing complex
Solution Approach 1:
The patent segments the spline curve into simpler components, each associated with specific discrete points and weights. This segmentation allows the system to maintain smoothness and differentiability while simplifying the modification process, as each segment can be adjusted independently without affecting the entire curve.
Solution Approach 2:
The patent uses parameter changes by introducing weighting coefficients that can be modified to adjust the curve's properties. By changing these parameters rather than the fundamental structure, the system achieves smooth curves while keeping the processing relatively simple.
3Ease of operation
If conventional RNN circuits are used with discrete inputs, then the circuit can process digital data, but it cannot output a fitting curve that satisfies the averaging process
Solution Approach 1:
The patent makes the RNN circuit universal by enabling it to perform both digital data processing and analog curve fitting functions. The circuit is designed to accept discrete inputs and produce continuous output curves that satisfy the averaging process, combining multiple functions in a single system.
Solution Approach 2:
The patent introduces an intermediary mechanism that bridges the gap between discrete digital inputs and continuous analog outputs. This intermediary allows the RNN circuit to process digital data while generating fitting curves that maintain mathematical properties like the averaging process.
Data Source
AI summary
A neural network designing method forms a RNN (Recurrent Neural Network) circuit to include a plurality of oscillating RNN circuits configured to output natural oscillations, and an adding circuit configured to obtain a sum of outputs of the plurality of oscillating RNN circuits, and inputs discrete data to the plurality of oscillating RNN circuits in order to compute a fitting curve with respect to the discrete data output from the adding circuit.