Analog Neural Learning Circuit for Fast Weight Tuning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital machine learning is cumbersome due to the large number of neurons in neural networks, making the calculation of derivatives and error contours time-consuming and requiring significant mathematical computing power.
Innovation Solution
An analog neural network error contour generation mechanism that perturbs weights and biases associated with analog neurons to measure errors, using a backpropagation mechanism with mini-batches of training samples and a weight tuning circuit employing ΣΔ modulators and charge domain multipliers to adjust weights and biases efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If digital machine learning is used to calculate derivatives and error contours, then accurate weight adjustments can be achieved, but the calculation time and computing power required become enormous
Solution Approach 1:
The patent replaces digital computational systems with an analog neural network system that performs weight adjustments through physical analog computations. The analog system uses continuous voltage or current signals to represent weights and performs derivative calculations through analog circuit operations, eliminating the need for digital computation of error contours and significantly reducing calculation time while maintaining adjustment accuracy.
Solution Approach 2:
The patent changes the fundamental parameter domain from digital discrete values to analog continuous values. By representing neural network weights as analog signals (voltages or currents) and performing computations in the analog domain, the system achieves real-time weight adjustments without the computational overhead of digital derivative calculations, thus resolving the time-accuracy contradiction.
2Measurement precision
If digital machine learning calculates chain of derivatives for error contour generation, then precise error measurement is achieved, but significant mathematical computing power is consumed
Solution Approach 1:
The patent substitutes digital mathematical computation with analog circuit operations. The analog neural network uses operational amplifiers, resistors, and capacitors to perform differentiation and integration operations physically, generating error contours through analog signal processing rather than digital calculation. This eliminates the need for significant mathematical computing power while maintaining precise error measurement through the physical properties of the analog circuits.
3Adaptability or versatility
If the number of neurons in neural network is large, then better learning capability is achieved, but the calculation of derivatives becomes more cumbersome
Solution Approach 1:
The patent replaces the digital computation architecture with an analog implementation where each neuron and connection is represented by analog circuit elements. The complexity of derivative calculations across large numbers of neurons is handled by the parallel analog circuit operations, where each analog neuron independently performs its local derivative calculations through physical circuit operations, avoiding the combinatorial complexity explosion that plagues digital systems with large neuron counts.
Data Source
AI summary
A neural network learning mechanism has a device which perturbs analog neurons to measure an error which results from perturbations at different points within the neural network and modifies weights and biases to converge to a target.