Relay Oscillator Gradient Updates in Thermodynamic Neural Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning algorithms using classical computing devices face challenges with increased execution time and energy consumption due to the complexity of calculating statistical probabilities, and existing thermodynamic computing systems require conversions that reduce their efficiency.
Innovation Solution
Implementing a thermodynamic computing system using oscillators to determine gradient terms and update weights and biases in a fully analogue way.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If classical computing devices are used to calculate statistical probabilities for machine learning algorithms, then calculation accuracy is maintained, but execution time and energy consumption increase significantly
Solution Approach 1:
The patent replaces classical digital computing systems with a thermodynamic computing system that uses physical oscillators to naturally simulate statistical probability calculations. The oscillators' thermal dynamics inherently perform the statistical computations that would otherwise require complex digital algorithms, thereby reducing execution time while maintaining calculation accuracy through physical law-based computation.
Solution Approach 2:
The invention changes the fundamental computation parameter from digital bit manipulation to continuous physical oscillator states governed by thermodynamic laws. By using oscillators with continuous amplitude and frequency parameters that naturally encode probability distributions, the system achieves efficient statistical calculations without the discrete step-by-step processing required by classical computers.
2Productivity
If thermodynamic computing systems are used to perform machine learning algorithms, then execution speed and energy efficiency improve, but communication overhead increases due to conversion requirements
Solution Approach 1:
The patent introduces relay oscillators as intermediary elements that bridge the thermodynamic computing system and classical computing devices. These relay oscillators receive gradient information from the thermodynamic system and translate it into a form suitable for classical processing, enabling efficient hybrid operation without requiring full conversion between systems and reducing communication overhead.
Solution Approach 2:
The invention segments the computing system into distinct functional modules: thermodynamic oscillators for statistical probability calculations, relay oscillators for gradient computation and information transfer, and classical computing components for weight and bias updates. This segmentation allows each component to operate in its optimal domain while minimizing inter-component conversion requirements.
3Use of energy by moving object
If thermodynamic computing systems are used to update weights and biases, then energy efficiency improves through analogue computation, but system complexity increases due to oscillator coupling requirements
Solution Approach 1:
The patent designs the oscillator system with universal coupling mechanisms that serve multiple functions: the same oscillator connections used for statistical probability calculations also facilitate gradient computation and weight updates. This multi-functionality reduces the need for separate dedicated components, thereby lowering overall system complexity while maintaining energy-efficient analogue computation throughout the learning process.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The thermodynamic computing system enables efficient and fast learning by updating weights and biases in a fully analogue way.
Implementation Method 1
a first set of oscillators of the thermodynamic chip may be configured to determine, in a fully analogue way, gradient terms of an engineered Hamiltonian of the energy-based model
Data Source
AI summary
Systems, methods and computer readable media relating to a neuro-thermodynamic computers configured to train a learning model based on values representing gradient terms stored in position degrees of freedom of relay oscillators are described. An energy-based model comprising oscillators representing neurons and oscillators representing synapse values may be trained using gradient terms obtained in an analogue way. The gradient terms may be stored on respective relay oscillators and gradient terms may be combined with other gradient terms. Oscillators representing synapse parameters may be updated based on one or more gradient terms. In some embodiments, the training protocol is implemented in a fully analogue way. In some embodiments, measurements of relay oscillators are performed and stored in a classical computing device for post-processing.