Laken

This repository contains code of Laken presented in the paper "Enhancing Active Learning through Latent Space Exploration: A K-Nearest Neighbors Approach"

How It Works

1. VAE Training:

   • Run vae_train.py to train the VAE model. This script will save the trained VAE for later use;
   • Important: Configure the VAE training parameters directly within the vae_train.py script, in the section designated for parameter settings.

2. Active Learning Loop:

   • Execute main_active_learner.py. This script loads the trained VAE and begins the AL process;
   • Important: Set the AL loop parameters (e.g., AL technique, number of iterations) directly within the main_active_learner.py script, in the parameter settings section.

Requirements

• Python (3.10.6)
• PyTorch (2.2.1)
• Torchvision (0.17.1)
• NumPy (1.26.4)
• SciPy (1.12.0)
• Scikit-learn (1.4.1.post1)
• Pandas (2.2.1)
• IPython (8.22.2)
• matplotlib (3.8.3)
• plotly (5.20.0)
• tqdm (4.66.2)

Citation

@article{FLESCA2025100584,
  title = {Enhancing Active Learning through Latent Space Exploration: A k-nearest Neighbors Approach},
  journal = {Array},
  pages = {100584},
  year = {2025},
  issn = {2590-0056},
  doi = {https://doi.org/10.1016/j.array.2025.100584},
  url = {https://www.sciencedirect.com/science/article/pii/S2590005625002115},
  author = {Sergio Flesca and Domenico Mandaglio and Francesco Scala},
  keywords = {Active learning, Latent space, Pseudo-labeling}
}