Does the Definition of Difficulty Matter? Scoring Functions and their Role for Curriculum Learning

This repository contains the configurations and code to reproduce the experiments and analyses of the paper "Does the Definition of Difficulty Matter? Scoring Functions and their Role for Curriculum Learning" published in IJCNN 2026 (arXiv). The implementation is based on aucurriculum v0.1.0 and autrainer v0.4.0.

Implemented by Simon David Noel Rampp with contributions by Manuel Milling and Andreas Triantafyllopoulos.

Experiment Reproduction

All experiments were conducted using Python 3.10 and PyTorch 2.1.0, and training was performed across a variety of GPU architectures, including NVIDIA GeForce GTX TITAN X, GeForce RTX 3090, and A40 GPUs.

After cloning the repository and navigating to the root directory, create a virtual environment and install the dependencies:

python -m venv .venv && source .venv/bin/activate
pip install aucurriculum==0.1.0
pip install efficientnet-pytorch==0.7.1

The experiments are organized in the cifar (for the CIFAR-10 experiments) and dcase (for the DCASE2020 experiments) directories (the following steps should be executed for each dataset separately). To reproduce the experiments, navigate to the respective directory:

cd <dataset> # replace <dataset> with cifar or dcase

Fetch the data and models and preprocess the dataset:

aucurriculum fetch -cn train
aucurriculum preprocess -cn train # only for DCASE2020

Launch the initial grid search and seed baselines training:

aucurriculum train -cn train
aucurriculum train -cn train_baselines

Launch the curriculum scoring function calculation:

aucurriculum score -cn curriculum
aucurriculum score -cn curriculum_agg_seed

Launch the curriculum training:

aucurriculum train -cn cl
aucurriculum train -cn cl_agg_seed

Postprocess the results:

aucurriculum postprocess results <dataset> -a seed # replace <dataset> with cifar or dcase

To reproduce the analyses, tables, and figures, execute the respective notebooks in the root directory of the repository.

Citation

If you use this code or aucurriculum in your research, please cite the following paper:

@inproceedings{rampp2026difficulty,
  title         = {Does the Definition of Difficulty Matter? Scoring Functions and their Role for Curriculum Learning},
  author        = {Rampp, Simon and Milling, Manuel and Triantafyllopoulos, Andreas and Schuller, Bj{\"o}rn W.},
  booktitle     = {2026 International Joint Conference on Neural Networks (IJCNN)},
  year          = {2026},
  pages         = {1--8},
  doi           = {10.1109/IJCNN69854.2026.11729720},
  eprint        = {2411.00973},
  archivePrefix = {arXiv}
}