SciCoNet (Scientific Computing Neural Networks) is a deep learning library designed for scientific computing on top of TensorFlow.
Use SciCoNet if you need a deep learning library that
- approximates functions from a dataset with/without constraints,
- approximates functions from multi-fidelity data,
- solves partial differential equations (PDEs),
- solves integro-differential equations (IDEs),
- solves fractional partial differential equations (fPDEs).
SciCoNet is extensible to solve other problems in scientific computing.
SciCoNet supports
- uncertainty quantification using dropout;
- four domain geometries: interval, disk, hyercube and hypersphere;
- two types of neural networks: fully connected neural network, and residual neural network;
- many different losses, metrics, optimizers, learning rate schedules, initializations, regularizations, etc.;
- useful techniques, such as dropout and batch normalization;
- callbacks to monitor the internal states and statistics of the model during training.
SciCoNet is built with four main modules, including
- domain geometry,
- data, i.e., the type of problems and constraints,
- map, i.e., the function space,
- model, which trains the map to match the data and constraints,
and thus is highly-configurable. It is easy to add new functions to each modules to satisfy new requirements.
The art of scientific computing with deep learning is to design Loss ℒ.