MultiObjectiveSurrogateOptimization

Efficient multi-objective optimization of high-dimensional neural dynamical systems by joint learning of objectives, constraints, and sensitivities

Setting up the environment

Install uv and uv sync.

This project uses machinable; check out the interface module to learn about available CLI options.

Experiments and results

All experiments and plots are in the notebooks at the top-level of this repository.

Data availability

Data is available on globus.org and can be retrieved automatically using the machinable globus interface.

Authenticating the Globus SDK client

The globus client must be authenticated once using the following steps:

1. Make sure the globus SDK is available

pip install globus_sdk

2. Create the client configuration

Create ~/.globus-config.json

{
    "client_id": "af2d8f73-c999-43d9-8cae-ec841055f6ca",   # machinable client
    "local_endpoint_id": "<your-globus-client-id-here>",   # UPDATE with your globus client ID that is running on your machine
    "local_endpoint_directory": "$STORAGE",
    "remote_endpoint_id": "<remote storage globus id>",
    "remote_endpoint_directory": "<remote storage location>"
}

3. Initiate the authentication

$ python -c "from machinable import get; get('interface.storage.globus').authorizer"

This will generate a URL to obtain a client token that you need to copy back into the terminal.