Neural Network Control Neurons for Hardware Error Verification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional approaches for providing redundancy in deep neural networks are unsatisfactory as they often interfere with the architecture or require dedicated hardware, making it difficult to effectively recognize and correct hardware errors in neurons.
Innovation Solution
A method involving separate control neurons to verify the functions of multiple neurons by calculating a first comparison value from neuron values and a second comparison value through a control calculation, using the associative law of mathematics to ensure proper neuron function, which can be implemented in standard modules without altering the neural network architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional approaches for providing redundancy are used in deep neural networks, then error recognition capability is improved, but the architecture is interfered with or dedicated hardware is required
Solution Approach 1:
The control neurons are integrated into the existing neural network architecture and serve dual purposes: they participate in normal computation while simultaneously providing error verification functionality. This allows the same hardware resources to perform both computation and verification without requiring separate dedicated hardware modules.
Solution Approach 2:
Control neurons act as intermediary elements that verify the functionality of computational neurons without directly interfering with the primary computation flow. These control neurons compute comparison values that are used to detect errors, serving as a mediating verification mechanism within the existing architecture.
2Reliability
If conventional redundancy approaches are used, then error recognition is improved, but training methods must be changed or dedicated hardware is required
Solution Approach 1:
The neural network performs self-verification using its own computational resources. The control neurons utilize the same computation resources and training frameworks as the computational neurons, allowing the system to verify itself without requiring external verification systems or specialized training procedures.
3Reliability
If control neurons are added to verify neuron functions, then error recognition is improved, but calculation time increases
Solution Approach 1:
The control neurons perform verification calculations concurrently with the computational neurons during the same time step. By preparing and comparing comparison values in parallel with the main computation rather than sequentially after, the verification process does not add significant time delay to the overall calculation cycle.
Data Source
AI summary
A method for verifying a calculation of a neuron value of multiple neurons of a neural network, including: carrying out or triggering a calculation of neuron functions of the multiple neurons, in each case to obtain a neuron value, the neuron functions being determined by individual weightings for each neuron input; calculating a first comparison value as the sum of the neuron values of the multiple neurons; carrying out or triggering a control calculation with one or multiple control neuron functions and with all neuron inputs of the multiple neurons, to obtain a second comparison value as a function of the neuron inputs of the multiple neurons and of the sum of the weightings of the multiple neurons assigned to the respective neuron input; and recognizing an error as a function of the first comparison value and of the second comparison value.