Federated Learning in Chemical Engineering: A Tutorial on a Framework for Privacy-Preserving Collaboration Across Distributed Data Sources

Overview

Welcome to the Review on Federated Learning (FL) for Chemical Engineering repository. Federated Learning is where multiple decentralized devices collaboratively train a model without sharing their local data. Each device trains the model on its own data and only shares the model updates.

A Review paper is coming soon for the Industrial & Engineering Chemistry Research Journal! - Link - {WIP}

The comprehensive results section can be seen:

Experiment Results

Experiment	Dataset	Train Accuracy	Test Accuracy	Train Loss	Test Loss	Time (sec)
Centralized Learning Experiments (Num Epochs: 25)
PILL - Standard_CentralNN_PILL.ipynb	PILL: Pharmaceutical Dataset	93.75%	82.39%	0.2600	0.5700	226.75
DNA - Standard_CentralNN_DNA.ipynb	Human DNA Sequence	100.00%	94.50%	0.0000	0.5080	97.49
DNA (MMoE) - Standard_CentralNN_DNA+MRI.ipynb	Brain MRI Scan + Human DNA Sequence	99.24%	84.64%	0.0274	0.7513	350.32
MRI (MMoE) - Standard_CentralNN_DNA+MRI.ipynb	Brain MRI Scan + Human DNA Sequence	99.72%	90.63%	0.0115	0.8555
HIV - Standard_CentralNN_HIV.ipynb	Human DNA Sequence	96.77%	95.45%	0.1640	0.1770	100.79
Federated Learning Experiments (Num Rounds: 20, Num Clients: 10, Num Epochs: 10)
PILL - Standard_FedNN_PILL.ipynb	PILL: Pharmaceutical Dataset	93.54%	94.79%	0.2800	0.2070	2215.40
DNA - Standard_FedNN_DNA.ipynb	Human DNA Sequence	100.00%	94.09%	0.0000	1.2030	3921.43
DNA (MMoE) - Standard_FedNN_DNA+MRI.ipynb	Brain MRI Scan + Human DNA Sequence	99.00%	94.75%	0.0776	0.4167	5543.29
MRI (MMoE) - Standard_FedNN_DNA+MRI.ipynb	Brain MRI Scan + Human DNA Sequence	99.38%	85.56%	0.1997	1.0720
HIV - Standard_FedNN_HIV.ipynb	Human DNA Sequence	96.31%	95.34%	0.1790	0.1870	1042.82

Results of Central and FL Experiments. In FL, both train loss and accuracy refer to a single client.

Repository Structure

.
├── LICENSE
├── README.md
├── run-cpu.sh
├── run-gpu.sh
├── run-tff-cpu.sh
├── run-tff-gpu.sh
└── src
    ├── README.md
    ├── Standard_CentralNN_DNA.ipynb
    ├── Standard_CentralNN_DNA+MRI.ipynb
    ├── Standard_CentralNN_HIV.ipynb
    ├── Standard_CentralNN_PILL.ipynb
    ├── Standard_CentralNN_Wafer.ipynb
    ├── Standard_FedNN_DNA.ipynb
    ├── Standard_FedNN_DNA+MRI.ipynb
    ├── Standard_FedNN_HIV.ipynb
    ├── Standard_FedNN_PILL.ipynb
    ├── Standard_FedNN_Wafer.ipynb
    ├── Standard_FedQNN_DNA.ipynb
    ├── Standard_FedQNN_DNA+MRI.ipynb
    ├── Standard_FedQNN_HIV.ipynb
    ├── Standard_FedQNN_PILL.ipynb
    ├── Standard_FedQNN_Wafer.ipynb
    ├── TFF_FedNN_PILL.ipynb
    └── utils
        ├── common.py
        ├── data_setup.py
        ├── engine.py
        └── __init__.py

Installation

Clone the Repository

git clone https://github.com/elucidator8918/Fed-ML-Chem.git
cd Fed-ML-Chem

Install Dependencies

For CPU

conda create -n fed python=3.10.12 anaconda
conda init
conda activate fed
bash run-cpu.sh

For GPU

conda create -n fed python=3.10.12 anaconda
conda init
conda activate fed
bash run-gpu.sh

Running Experiments

Choose the appropriate notebook based on your dataset and encryption preference:

Standard Federated Learning

• DNA Sequence Dataset:

  • Notebook: src/Standard_FedNN_DNA.ipynb
  • Description: This dataset includes DNA sequences used for various biological and genetic studies, focusing on sequence classification and pattern recognition.

• DNA+MRI Multimodal Dataset:

  • Notebook: src/Standard_FedNN_DNA+MRI.ipynb
  • Description: It is used as a MoE with Multimodaility leveraging both DNA Sequence and MRI scans data to develop and evaluate models for detecting and interpreting tumors and dna classes.

• PILL Dataset:

  • Notebook: src/Standard_FedNN_PILL.ipynb
  • Description: This dataset includes images of pharmaceutical pills from the Pill Dataset used for various pharmaceutical studies, focusing on classification and pattern recognition.

Standard Quantum Federated Learning

• DNA Sequence Dataset:

  • Notebook: src/Standard_FedQNN_DNA.ipynb
  • Description: This dataset includes DNA sequences used for various biological and genetic studies, focusing on sequence classification and pattern recognition.

• DNA+MRI Multimodal Dataset:

  • Notebook: src/Standard_FedQNN_DNA+MRI.ipynb
  • Description: It is used as a MoE with Multimodaility leveraging both DNA Sequence and MRI scans data to develop and evaluate models for detecting and interpreting tumors and dna classes.

• PILL Dataset:

  • Notebook: src/Standard_FedQNN_PILL.ipynb
  • Description: This dataset includes images of pharmaceutical pills from the Pill Dataset used for various pharmaceutical studies, focusing on classification and pattern recognition.

License

This project is licensed under the MIT License. See the LICENSE file for details.