Neocortex Early User Program
Applications will be accepted in Summer 2020
The Neocortex Early User Program (EUP) provides a unique opportunity for the earliest possible access to the remarkable technology of the Cerebras CS-1 and HPE Superdome Flex integrated system. Participating research groups will receive one-on-one assistance from PSC and Cerebras experts to port, scale, and optimize their workflows. The EUP is expected to run for 30 days. Areas of focus in the EUP will be on generating scientific outcomes, achieving high performance, and maximizing ease of use. The EUP includes providing feedback to PSC and Cerebras to document the results of the EUP and inform the evolution of the system.
There is no cost to participate in the Neocortex EUP or to use Neocortex for open research following the EUP. The Neocortex team is seeking applicants for the Early User Program who meet the following criteria:
- Applicants must have deep learning training tasks that would benefit from acceleration. All research domains (biology, physics, economics, robotics, etc.) are welcome.
- Applicants must be able to articulate a goal for the 30-day EUP. Some examples are as follows:
- [Scientific Outcome] “In our drug repurposing study, screen 100,000 candidate molecules and come up with a ranked list of the top 10 candidates likely to have therapeutic value.”
- [Performance] “Decrease the training time for our natural language understanding task from 1 week to 1 hour.”
- [Performance] “We’ve been unable to train our model on our full dataset because it would take too long on the resources to which we have access. We’d like to train on our full data in less than 8 hours.”
- [Model Development] “Existing networks are not well suited to identifying gene signaling pathways from our new single-cell sequencing data. We need to develop a new network architecture from around 8 candidates, optimize its hyperparameters, and demonstrate its improved accuracy.”
- Applicants must commit to collaborating with the Neocortex team to maximize the chance of success. Ideally, there will be one point of contact for each research project. A research group can pursue more than one research project.