By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.
GLAI is a new Artificial Intelligence system designed to efficiently replace feed-forward neural networks. It significantly improves their performance and efficiency while retaining all their functionality.
The key element of GLAI is its foundation on a linear system that can emulate highly non-linear behaviour by selecting model subsets for each sample.
GLAI stands out in scenarios where:
1
The AI model needs frequent or real-time retraining to improve its accuracy
2
Retraining is costly due to its high frequency or large data volume
3
Retraining must be performed directly on edge devices, keeping data secure without leaving the device.
4
AI/ML model explainability is necessary to bring it into production and gain stakeholders trust
Technology
Key Benefits of GLAI
Powering the next generation of AI/ML applications
Rapid Model Combination
GLAI allows for nearly instantaneous combination of different models, provided they share the same structure and subset selector for each sample. This feature has multiple applications, such as incremental retraining, federated (re)training, and dynamic creation of new models from existing ones.
Faster Training and Retraining
GLAI achieves 100-1000 times faster retraining due to model combination (depending on the dataset size) compared to conventional neural networks, representing significant improvements in cost and efficiency (reducing energy consumption by the same magnitude).
Accuracy
GLAI maintains the same level of accuracy (+-1%) as traditional neural networks, ensuring reliable performance without sacrificing model quality. In some cases, it can achieve notable improvements in accuracy, thanks to the independence of the effect of different trainable parameters on the outputs.
Explainability
Unlike traditional neural networks, GLAI provides model explainability by clearly showing how inputs affect outputs (model results). This is crucial for sectors that require transparency and greater oversight.
Quantum-Ready
Fully equipped to leverage the power of quantum computing for training. This readiness positions our model to take full advantage of quantum computing advancements, ensuring faster convergence and improved performance during training.
