Neural Network Training via Quantum Simulation to Escape Local Minima
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning models, such as neural networks, are reaching their limitations and require new learning approaches due to the complexity of quantum machine learning, which is hindered by the need for expensive and difficult-to-maintain physical quantum systems, and classical training methods like gradient descent often get stuck in local minima.
Innovation Solution
A method is proposed that simulates quantum systems using digital computers to train neural networks, leveraging quantum tunnelling effects through approximated density functional theory simulations, avoiding the need for physical quantum systems and enabling efficient training of neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If physical quantum systems are used for quantum machine learning, then quantum computational advantages can be achieved, but the hardware becomes expensive and difficult to maintain
Solution Approach 1:
The patent applies the copying principle by creating a simulated quantum system that replicates the essential quantum behaviors needed for machine learning without requiring physical quantum hardware. The simulation uses classical computing to model quantum phenomena, thereby copying the functional advantages of quantum systems while avoiding their hardware complexities and maintenance requirements.
Solution Approach 2:
The patent substitutes the mechanical/physical quantum system with a computational simulation. Instead of using actual quantum hardware with all its physical constraints and maintenance needs, the invention replaces it with a software-based quantum system simulation that runs on classical computers, thereby eliminating the hardware complexity while preserving the quantum computational advantages.
2Productivity
If classical training methods like gradient descent are used, then training is computationally efficient, but the methods get stuck in local minima
Solution Approach 1:
The patent applies parameter changes by transitioning from classical training parameters to quantum-inspired parameters in the simulation. The quantum system simulation uses different optimization dynamics that allow the training process to escape local minima while maintaining computational efficiency. The quantum simulation introduces new parameter spaces and optimization mechanisms that differ fundamentally from classical gradient descent.
Solution Approach 2:
The quantum system simulation acts as an intermediary between classical training methods and the neural network. Instead of directly applying gradient descent to the neural network, the patent introduces a quantum simulation layer that mediates the training process, allowing the system to benefit from quantum-inspired optimization while still working with classical neural network architectures.
3Adaptability or versatility
If quantum machine learning is implemented with physical systems, then novel learning approaches can be achieved, but the implementation becomes prohibitively expensive
Solution Approach 1:
The patent uses copying to create a simulated quantum environment that replicates the novel learning capabilities of physical quantum systems without the associated costs. The simulation copies the essential quantum behaviors and learning mechanisms, enabling researchers to explore quantum machine learning approaches using standard computational resources rather than expensive physical quantum hardware.
Solution Approach 2:
The patent employs cheap computational simulations instead of expensive physical quantum systems. The simulated quantum system can be created, used, and modified at minimal cost compared to physical quantum hardware, allowing for rapid experimentation and iteration in quantum machine learning research without the prohibitive costs of maintaining actual quantum devices.
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
This approach allows for qualitatively different neural network behaviors, efficiently training networks that are difficult with classical methods, and provides real-life quantum machine learning capabilities without requiring actual quantum computing devices.
Implementation Method 1
leveraging quantum tunnelling effects through approximated density functional theory simulations
Implementation Method 2
through approximated density functional theory simulations
Data Source
AI summary
The disclosure relates to a computer implemented method, system, apparatus and non-transitory computer readable media for training an artificial neural network (ANN). The method comprises defining an energy function, for the ANN and a dataset, in terms of quantum objects and simulating a quantum system, using the quantum objects, to reduce the energy function and obtain a trained ANN. The method may further comprise using the reduced energy function as an input to a genetic algorithm for refining the reduced energy function.