What’s Edge AI?
Much of AI is currently working in the cloud. Soon, however, AI will be able to run on devices such as cars and smartphones, thanks to Edge AI technology. We realize Edge AI with a unique technology that compresses AI models.
Differences between Cloud-processing AI and Edge-processing AI
- Cloud-processing AI
- Delayed performance (taking a long time to send and receive data)
- Unreliable Network (E.g., cannot be used outside the service area)
- High network bandwidth occupancy (heavy data)
- Edge-processing AI
- It’s unnecessary to connect with the cloud as AI can work on the edge device. Better real-time responses can be expected.
Benefits of Edge AI by “compressing” AI models
- AI can be mounted on smaller edge devices
- Costs of hardware for AI inference can be reduced
- Power consumption can be lessened as the number of calculations is decreased
- Not only is the data connection time skipped, but the amount of computation is reduced, enabling higher-speed operation
Our Edge AI offerings
We have developed a patented Edge AI technology, which compresses an original AI model to 1/4-1/16 with comparable
accuracy, and enables you to implement AI on automobiles or smartphones.
We are also developing “Pressai”, an application that automates AI compression.
Advantages of Araya’s compression method
The neural network compression application incorporates Araya’s original compression method (patented).
- The calculation amount can be reduced to 1/4 to 1/16 with comparable accuracy
- Most neural networks can be compressed
- Different neural networks can be implemented on the same dedicated hardware (FPGA).
- You can combine with other compression methods (excluding pruning).
Benefits of the automated compression tool (“Pressai”)
- Optimal compression
- You can compress an AI model with Ayara’s method as well as other compression methods (e.g., quantization, pruning), enabling you to compare results and choose the best model
- Although you normally need several technicians that have different skill for each process (e.g., compression, implementation on CPUs and FPGAs), the app allows you to save that specialized personnel
- Shortening development time
- The development time, which usually takes several months, can be shortened.
- You don’t need to export your data because you can conduct all the processes on your premises.