Generative Synthesis
DarwinAI's Generative Synthesis technology uses AI itself to observe and understand a neural network and then generates a number of new and highly optimized versions of said network. The result are new and entirely unique networks that are not only significantly smaller than the original, but infer faster without sacrificing functional accuracy. The understanding garnered by the engine also enables explainable deep learning, whereby the platform can illuminate how a network is reaching its conclusions. The Generative Synthesis platform uses TensorFlow* and provides a web-based interface for orchestration. The platform is typically made available by way of Docker* images and is hosted in the client’s environment, which is important for enterprise As evidenced by the Solution Brief from the AI Builders group, the process can yield order of magnitude improvements for performance on Intel devices.
*Please note that member solutions are often customizable to meet the needs of individual enterprise end users.
CONTACT COMPANY►SOLUTION FEATURES
- Accelerate deep learning development by collaborating with AI during the design process.
- Reduce computational requirements by generating highly optimized models (particularly useful in deploying deep learning at the edge).
- Understand how a neural network is reaching its decisions – important in both diagnostic and regulatory contexts.
RESOURCES
- Generative Synthesis
- AI Builder Micorsite
- Solution Brief - DarwinAI Generative Synthesis* Platform and Intel® Optimizations for TensorFlow* Accelerate Neural Networks
- Using AI to Build Explainable AI with Intel Optimizations and DarwinAI - Intel on AI Episode 25
- Case Study - DarwinAI Delivers Explainable AI Using OpenVINO™ Toolkit