CASE STUDY
Quantum Utility Unlocked for the Automotive Industry
Access world-class algorithms, build, tune and run hybrid quantum models and unlock business value today
The Problem: Enhanced Image Recognition for Car Classification
Image recognition is critical in the automotive industry for tasks like defect detection, part sorting, and self-driving perception. Automotive manufacturers need early and accurate fault detection during the manufacturing process to prevent unnecessary waste and costs. While improving the accuracy of image recognition for fault detection in manufacturing has the potential to create cost savings and operational efficiencies, training deep neural networks is computationally very demanding.
The Objective
The goal is to minimize costs and computational effort required while improving image classification accuracy despite the growing size of data sets.
Aiming for enhanced machine learning models, a global leading car manufacturer leveraged our proprietary approach to classify images of their cars into makes and models. The aim was to optimize the architecture of a hybrid quantum-classical neural network, maximizing accuracy.
Aiming for enhanced machine learning models, a global leading car manufacturer leveraged our proprietary approach to classify images of their cars into makes and models. The aim was to optimize the architecture of a hybrid quantum-classical neural network, maximizing accuracy.
Our Solution
Using a proprietary hybrid approach, Terra Quantum combined classical and quantum layers into an end-to-end quantum-classical residual neural network. For a car classification test, this hybrid model achieved 98.9% accuracy, significantly outperforming a classical model. The tensor train optimization further accelerated finding the optimal circuit architecture versus standard grid search.
Performance Metrics
- Maximize accuracy of car classification tasks
- Increase quality delivered as the problem scale increases
- Minimize optimization time by accelerating the optimal circuit architecture, thus increasing the efficiency of the solution
98.9%
Accuracy of the cars image classification tasks, leveraging the hybrid quantum-enhanced approach
6.9%
Improvement in image classification accuracy with a Hybrid Quantum Neural Network (HQNN)
3X
Improvement in computational effort required, using Tensor Train Optimization methods