How this works
Step 1: On Sign-up
After signing up, we create an AWS S3 bucket available to you based on your training data size.
We also send you an email confirming your sign-up, recommending encryption options and providing you with a sample configuration file which shows you how to select your model.
For training data, we accept multiple formats ranging from CSV type text files to floating-point binary files.
Step 2: Encrypt and upload training data
First, we recommend encrypting your data. We provide a free encryption tool that you can request via your sign-up email and will let you manage the keys to your data. You may also choose to encrypt using a provider of your choice.
In the email, you will be required to provide:
The model type through a configuration file. This is a simple text file and a sample will be included in the email.
The encryption key for us to use to train your data.
Optional: One row of values. (This is used to cross-validate that decryption was successfully completed.)
Optional: Two sets of random starting weights (We will use our own if not provided. These will be used to compute training error values for both sets.)
Step 3: Decryption, normalizing and training
With the provided encryption key, we will decrypt the data and normalize it. This makes pre-normalization of your data unnecessary.
Then, we will generate two initial sets of random numbers (or use the ones you provided) and execute two training runs, each with the two initial value numbers.
Once training is complete, we will report the training error values for both sets and send the weights and training error values to you via an encrypted email.
We will train models up to 8 layers and with 4096 nodes.
