12/23/20 PR Draft Update:
Includes updates from [PR #14], compatible with NVIDIA v3.1 Docker image

Updates

• Added NVIDIA-specific installation instructions: docker/README.md
• Client contributions are now passed as a list in the aggregation headers and is no longer hardcoded
• Docker run scripts added in docker

Pending Issues

• Client drops out of training after training for 2 rounds
• When more training/validation datasets are added, strange behavior with feature shapes that breaks client training
• Training with multiple clients results in clients randomly dropping out of the training
• Client runs into segfaults (doesn't happen for hardcoded contributions)
• Directory references to kvah need to be changed to secure-aggregation to match with installation instructions

Some issues are potentially a result of changes made in host.cpp and ecalls.cpp that include client contributions in the function headers (I might not be accessing the values in the enclave appropriately)