MicroAlgo Inc. has announced the launch of their latest classifier auto-optimisation technology based on Variational Quantum Algorithms (VQA). This technology reduces the complexity of parameter updates during training through deep optimisation of the core circuit, markedly improving computational efficiency. Compared to other quantum classifiers, this optimised model has lower complexity and incorporates advanced regularisation techniques, effectively preventing model overfitting and enhancing the classifier’s generalisation capability. The introduction of this technology marks a step forward in the practical application of quantum machine learning.
Traditional quantum classifiers can theoretically use the advantages of quantum computing to accelerate machine learning tasks, but they still face numerous challenges in practical applications. Firstly, current mainstream quantum classifiers often require deep quantum circuits to achieve efficient feature mapping, which results in high optimisation complexity for quantum parameters during training. Additionally, as the volume of training data increases, the computational load for parameter updates grows rapidly, leading to prolonged training times and impacting the model’s practicality.
MicroAlgo’s classifier auto-optimisation technology reduces computational complexity through deep optimisation of the core circuit. This approach improves upon two key aspects: circuit design and optimisation algorithms. In terms of circuit design, the technology adopts a streamlined quantum circuit structure, reducing the number of quantum gates and thereby lowering the consumption of computational resources. On the optimisation algorithm front, this classifier auto-optimisation model employs a parameter update strategy, making parameter adjustments more efficient and substantially accelerating training speed.
In the training process of classifiers based on variational quantum algorithms (VQA), parameter optimisation is one of the most critical steps. Generally, VQA classifiers rely on Parameterised Quantum Circuits (PQC), where updating each parameter requires computing gradients to adjust the circuit structure and minimise the loss function. However, the deeper the quantum circuit, the more complex the parameter space becomes, requiring optimisation algorithms to perform more iterations to achieve convergence. Furthermore, uncertainties and noise in quantum measurements can also affect the training process, making it difficult for the model to optimise stably.
Traditional optimisation methods often employ strategies such as Stochastic Gradient Descent (SGD) or Variational Quantum Natural Gradient (VQNG) to find optimal parameters. However, these methods still face challenges such as high computational complexity, slow convergence rates and a tendency to get trapped in local optima. Therefore, reducing the computational burden of parameter updates and improving training stability have become key factors in enhancing the performance of VQA classifiers.
MicroAlgo’s classifier auto-optimisation technology, based on variational quantum algorithms, reduces the computational complexity of parameter updates through deep optimisation of the core circuit. It also incorporates regularisation techniques to enhance the stability and generalisation capability of the training process. The core breakthroughs of this technology include the following aspects:
Depth optimisation of quantum circuits to reduce computational complexity: In traditional VQA classifier designs, the number of layers in the quantum circuit directly impacts computational complexity. To lower computational costs, MicroAlgo employs an Adaptive Circuit Pruning (ACP) method during optimisation. This approach dynamically adjusts the circuit structure, eliminating redundant parameters while preserving the classifier’s expressive power. As a result, the number of parameters required during training is reduced, leading to a substantial decrease in computational complexity.
Hamiltonian Transformation Optimisation (HTO): Additionally, MicroAlgo introduces an optimisation method based on Hamiltonian transformations. By altering the Hamiltonian representation of the variational quantum circuit, this technique shortens the search path within the parameter space, thereby improving optimisation efficiency. Experimental results demonstrate that this method can reduce computational complexity by at least an order of magnitude while maintaining classification accuracy.
Novel regularisation strategy to enhance training stability and generalisation capability: In classical machine learning, regularisation methods are widely used to prevent model overfitting. In the realm of quantum machine learning, MicroAlgo introduces a novel quantum regularisation strategy called Quantum Entanglement Regularisation (QER). This method dynamically adjusts the strength of quantum entanglement during training, preventing the model from overfitting the training data and thereby improving the classifier’s generalisation ability on unseen data.
Additionally, an optimisation strategy based on the Energy Landscape is incorporated, which adjusts the shape of the loss function during training. This enables the optimisation algorithm to more quickly identify the global optimum, reducing the impact of local optima.
Enhanced noise robustness for real quantum computing environments: Given that current Noisy Intermediate-Scale Quantum (NISQ) devices still exhibit significant noise levels, a model’s noise resilience is critical. To improve the classifier’s robustness, MicroAlgo proposes a technique based on Variational Quantum Error Correction (VQEC). This method actively learns noise patterns during training and adjusts circuit parameters to mitigate noise effects. This strategy markedly enhances the classifier’s stability in noisy environments, making its performance on real quantum devices more reliable.
MicroAlgo’s classifier auto-optimisation technology, based on variational quantum algorithms, reduces the computational complexity of parameter updates through deep optimisation of the core circuit and the introduction of novel regularisation methods. This approach boosts training speed and generalisation capability. This breakthrough technology not only demonstrates its effectiveness in theory but also exhibits superior performance in simulation experiments, laying a crucial foundation for the advancement of quantum machine learning.
As quantum computing hardware continues to advance, this technology will further expand its application domains in the future, accelerating the practical implementation of quantum intelligent computing and propelling quantum computing into a new stage of real-world utility. In an era where quantum computing and artificial intelligence (AI) converge, this innovation will undoubtedly serve as a significant step in advancing the frontiers of technology.
Comment on this article via X: @IoTNow_ and visit our homepage IoT Now
