Dynamic Neural Accelerator featuring EdgeCortix CEO Sakya Dasgupta
In this video, Marcus from BittWare® introduces the topic of machine learning at the edge and the benefits of using FPGAs for edge inference due to their flexibility, power efficiency, and ability to accommodate different neural networks.
Sakya Dasgupta of EdgeCortix then explains their approach of co-design and co-exploration, where they optimally balance neural network requirements with hardware metrics using BittWare card with an Altera® Agilex FPGA.
The advantages of these cards include:
- High-Performance INT8 Operations
- PCIe Gen 4 Support
- DDR4 Memory
- Up to 20 TOPS AI Compute
The MERA software stack & framework, which supports major machine learning frameworks, offers benefits like seamless application deployment, diverse deep neural network operator support, and profiling tools for testing performance and model accuracy.
EdgeCortix highlights the importance of edge AI in both defensive and consumer sectors, enabling data identification and localization through deep neural networks, particularly convolution neural networks, and the necessity for continuous co-designing efforts to balance efficiency and flexibility as the field evolves.