STEMDL

A Python package for distributed deep learning with a special focus on inverse problems in materials imaging.
stemdl was used in the following (applied and fundamental) deep learning research projects:

1. 3-D reconstruction of Structural Distortions from Electron Microscopy (Link to Paper)
2. 27,600 V100 GPUs and 7MW(h) of Power to solve an age-old scientific inverse problem (Link to Paper and Medium story )
3. YNet: a Physics-Constrainted and Semi-Supervised Learning Approach to Inverse Problems

---

Getting Started

See scripts folder for the following:

1. stemdl_run.py:
   Python script. Runs from the CLI to setup Neural Nets and start training/evaluation operations.
2. generate_json.py:
   Python script. Generates .json files needed as input for stemdl_run.py

---

Brief description of Modules:

1. inputs.py:
   Classes to read training/evaluation data, create training batches, and image transformations.
   Can handle I/O ops on TFRecords, numpy arrays, and lmdb files
2. network.py:
   Classes to setup various kinds of Neural Nets (ConvNets, ResNets, etc...)
3. runtime.py:
   Functions and Classes to perform (low-level) network training/evaluation
4. io_utils.py:
   Functions to generate .json files for model architectures input files, hyperparameters, and training runs configurations
5. losses.py:
   Functions to generate and manipulate loss functions
6. optimizers.py:
   Optimizer setup and gradients pre-processing and reduction
7. automatic_loss_scaler.py:
   Python module for dynamic loss scaling during fp16 training (taken as is from OpenSeq2Seq)

---

Software Requirements:

1. numpy >= 1.13
2. tensorflow >=1.2
3. python 3.6
4. horovod >=0.16

Hardware Requirements:

1. CUDA compatible GPU >=1

Install ####:

Project not yet on Pypi. For now:

git clone
cd stemdl
pip install .