Federated Learning in IoT–Fog–Cloud Hierarchies Using Progressively Deepening Split Neural Networks

Status: Architecture-Driven Reference Implementation (End-to-End Simulation)

---

Submitted by

• Pranav Gupta (22JE0710)
• Saransh Shivhare (22JE0867)
• Ayush Shaw (22JE0221)

Under the guidance of:
Prof. Prasanta K. Jana
Department of Computer Science and Engineering
Indian Institute of Technology (ISM) Dhanbad

---

Overview

This repository contains a reproducible research prototype for Progressive Split-Federated Learning (HSFL) across an IoT–Fog–Cloud hierarchy.

The goal is to provide:

• A clear system architecture
• A reference implementation (prototype)
• A small-scale simulation validating dataflow and communication overhead

⚠️ This is a research repository, not a production deployment.
It is designed for reproducibility, experimentation, and extension.

---

System Architecture

Overall IoT–Fog–Cloud Hierarchy

Training & Communication Workflow

---

Key Highlights

• Architecture-first design: Explicit separation of IoT, Fog, and Cloud layers
• Reproducible prototype: Run locally using CIFAR-10 or random tensors
• Experiment artifacts: Logs and plots generated automatically
• Honest scope: Pipeline validation, not full-scale convergence

---

Development & Setup

Tested on Windows (PowerShell) and Linux/macOS with Python 3.10+.

1. Clone the repository

git clone https://github.com/saranshs17/hybrid-SplitNN-FL.git
cd hybrid-SplitNN-FL

2. Create & activate a virtual environment

Windows (PowerShell)

python -m venv .venv
.\.venv\Scripts\Activate.ps1

Linux / macOS

python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies

python -m pip install --upgrade pip
pip install -r requirements.txt

requirements.txt includes PyTorch, matplotlib, pytest, PyYAML, and torchvision.

4. Run the prototype simulation

python -m src.sim.run_simulation --config configs/experiment.yaml

• If dataset: cifar10 → CIFAR-10 will be downloaded (requires torchvision)
• For fast offline runs, set dataset: random

Outputs (saved to results/):

• results/log.txt
• results/loss_curve.png
• results/bytes_per_batch.png

5. Run unit tests

pytest -q

Run a specific test:

pytest tests/test_protocol.py::test_run_round_smoke -q -s

---

Prototype Experiment Results (Example)

The following results are from a CIFAR-10 experiment (one batch per epoch) to validate architectural correctness.

Loss per step

Approximate communication per batch

Loss fluctuations are expected due to the small dataset subset and limited steps.

---

High-Level Workflow

IoT Clients

• Perform lightweight feature extraction
• Generate smashed activations

Fog Nodes

• Receive activations
• Perform regional feature computation

Cloud Server

• Final inference and parameter updates
• Full cross-device SplitNN backpropagation is planned as future work

---

Reference Documents

• docs/Project_Report.pdf — thesis (architecture & analysis)
• docs/Architecture_Slides.pdf — Presentation slides

---

Future Extensions

• Implement full SplitNN backpropagation
• Add secure aggregation & compression
• Scale experiments using Flower / FedML
• Multi-host IoT–Fog–Cloud deployment

---

License

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

---

Contact

Saransh Shivhare
saranshmain17@gmail.com