Adaptive Low-Rank Optimization for Federated Learning

Undergraduate thesis on DyLoRA for communication-efficient federated optimization.

This undergraduate thesis integrated Dynamic Low-Rank Adaptation into a federated learning pipeline to reduce communication cost while preserving model performance.

Highlights

  • Integrated DyLoRA into a modular federated learning benchmark.
  • Reduced communication cost by 80% without loss of accuracy.
  • Improved training stability across clients through reproducible PyTorch experiments.
  • Received 85/100, High Distinction (Niu, 2025).

References

2025

  1. Adaptive Low-Rank Optimization for Federated Learning
    Zerun Niu
    2025
    High Distinction, 85/100