MetaTF is BrainChip's toolset designed to facilitate the development and deployment of neural networks on their Akida platform. It simplifies the process of leveraging BrainChip’s AI capabilities by allowing the conversion of existing TensorFlow models to the Akida platform. Using Python and associated tools like Jupyter notebooks, MetaTF offers a seamless environment for training and deploying AI models optimized for event-based computations.
The framework consists of multiple Python packages: the Akida package which provides an interface to BrainChip's neuromorphic chip, and CNN2SNN to convert convolutional neural networks for event domain processing. MetaTF's core value lies in streamlining the creation of low-latency, low-power networks that are inherently suited for BrainChip's portfolio of AI processors.
With a built-in model zoo and performance simulation features, MetaTF enables users to evaluate and optimize models efficiently. This toolset ensures smooth integration with existing AI workflows by removing the necessity to adopt new machine learning paradigms completely, making it a vital component in BrainChip’s AI enablement strategy.
