Large-Scale Computing Platforms
- Target existing machine customers currently deployed on GPUs, as well as high-end government compute facilities.
- Reconfigure single-layer usage to potentially permit cascaded multi-layer model for deeper level operation.
- Internal layer training likely requires support GPU for back propagation, so employ common tools like CAFFE with additional mapping for neurons.
- Co-develop and deploy emerging research from University of California, Berkeley engineering and neuroscience.
- Exploring future partnerships with Department of Energy national labs (LLBL, ORNL).
- In-house projects for extreme-pixel camera platforms (100+ megapixels)
Additionally, government entities have extremely demanding requirements for analysis, encryption, and unsupervised learning.
Toward that end, our partners have deployed hyperspectral partially-masked target recognition in battlefield tank scenarios, very fast secure communication channel decryption, and natural language monitoring in real time for social networks.
Key characteristics of the technology suitable for these tasks include very low power (essential for airborne environments), in-situ learning (potential battlefield adaptability), and rapid response (decision in under 500ns in some deployments).
Toward that end, our partners have deployed hyperspectral partially-masked target recognition in battlefield tank scenarios, very fast secure communication channel decryption, and natural language monitoring in real time for social networks.
Key characteristics of the technology suitable for these tasks include very low power (essential for airborne environments), in-situ learning (potential battlefield adaptability), and rapid response (decision in under 500ns in some deployments).