Training terabyte-scale neural networks

We are a neural network training service provider who trains your NN model with your data and generates ready-to-deploy model weights.

Our very high throughput system is designed to take your neural networks to production. As the labeled data is generated from business transactions, attains a terabyte scale volumes, the production models need to be refreshed - and we do that best. For data with low drift, we offer the flexibility to start with previous weights, or, if so desired, generate a new random starting weights each run.

Our model specification software insists that every source data record be encrypted using NIST approved codebooks such as Galois Counter Method GCM, or CBC-CS1 or CBC-CS2 using AES 256 or AES 192 or AES 128 encryption. Our implementation decrypts these records only once these are ready to be processed inside the RAM, thus reducing the attack surface to an absolute minimum.

Organizations across industry domains that need to use terabyte-scale data sets to deploy neural networks-based business processes.

Across all business sizes, the transaction data reaches terabyte-scale volumes fairly easily. While storage and maintenance of such sizes is easy, its ingestion into NN models is still a difficult task.

Our service is designed to offer a low-cost low-latency deployment of the NN model into regular decision making process that a business requires.

Our implementation considers data security as an non-negotiable imperative. We decrypt the data only in-situ, just before its consumed, and each thread decrypts its own data set. The decrypted data is never written back. The threads are run after accounting for physical RAM and core availability, ensuring that the data is also never written out to a swap device.

This is an AVX-512 native code implementation, so by design, is optimized at an instruction level. The actual run is done on a cloud server machine with capacity suited to your budget.

Our pricing model considers the cloud machine cost as a pass-through component, so you are free to leverage your existing cloud contracts to reduce total cost.

We ingest the data in a normalized form, where each value is replaced by (X-a)/b, and the values of (a,b) do not cross your security perimeter, thus making it impossible for us to reconstruct the data - a very critical security characteristic.

Other than the C run time library that interfaces with the operating system, our code does not use any library by any third party. Every line of code, including all security algorithm implementations, all mathematical functions, all thread management, memory management and scheduling, have been developed in-house to guarantee reliability. All the security implementations are now under certification by NIST.