🧠 S3LiteNet: Lightweight Sparse Neural Models for Real-Time Edge-Based Anomaly Detection

S3LiteNet is a multi-phase research and development initiative aimed at advancing lightweight neural network models for anomaly detection at the edge, particularly in Information-Centric Networking (ICN) and IoT contexts. This repository documents the full pipeline — from systematic literature review to benchmarking, and ultimately the design of a custom sparse model optimized for edge deployment.

---

🔍 Project Overview

Phase	Title	Status	Focus
1	Systematic Literature Review	90% Complete	Analyze trends, gaps, and deployment barriers in existing edge-based lightweight models
2	Benchmarking Study	Complete	Compare CNN, LSTM, GRU, Transformer, and hybrids across accuracy, latency, memory, and model size
3	S3LiteNet Development	In Progress	Design a custom sparse neural network optimized for edge inference with explainability

---

📁 Repository Structure

S3LiteNet/
│
├── Phase_1_Literature_Review/     # PRISMA protocol, extracted data, and analysis scripts
├── Phase_2_Comparative_Study/       # Source code for training, evaluating, and profiling models
├── Phase_3_S3LiteNet_Model/       # Work-in-progress implementation of the S3LiteNet architecture
├── paper/                         # IEEE manuscript, pre-print, and related figures
├── results/                       # Aggregated benchmarking results and visualizations
└── README.md                      # You're here!

---

📌 Research Goals

• ⚡ Efficiency: Reduce computational and memory overhead for real-time inference.
• 🔍 Explainability: Enhance transparency through integrated XAI techniques (e.g., SHAP, attention).
• 🛰️ Deployment: Support deployment in ICN and IoT environments with limited connectivity.
• 🧠 Adaptability: Address concept drift, data imbalance, and edge-device heterogeneity.

---

📝 How to Cite

If you use this repository, please cite our preprint:

Nti et al. “Evaluating Lightweight Neural Models for Edge-Based Anomaly Detection: Performance and Efficiency Trade-offs.” Research Square Preprint (2025). https://doi.org/10.21203/rs.3.rs-7138288/v1

BibTeX:

@article{asiedu2025s3litenet,
  title={Evaluating Lightweight Neural Models for Edge-Based Anomaly Detection: Performance and Efficiency Trade-offs},
  author={Nti and Others},
  journal={Research Square},
  year={2025},
  doi={10.21203/rs.3.rs-7138288/v1}
}

---

🧪 Publications and Manuscripts

• 📄 IEEE Manuscript (Under Review): paper/IEEE Manuscript_draft_R2.pdf
• 📄 Preprint (Published): paper_pre_print.pdf