Dendritic Artificial Neural Networks Benchmark Package

This repository contains the minimal public package for a controlled benchmark study of dendritic artificial neural networks (DANNs).

Repository URL: https://github.com/canay/dendritic-ann-benchmarks

Contact

• Özkan Canay
• Department of Information Systems and Technologies
• Faculty of Computer and Information Sciences
• Sakarya University
• Email: canay@sakarya.edu.tr
• ORCID: 0000-0001-7539-6001
• Web: https://canay.sakarya.edu.tr/

Public contents

• dann_benchmark/ Benchmark code, dataset-access notes, and archived accuracy and timing outputs.
• stats_outputs/ Supporting statistical summary derived from the archived benchmark histories.
• REPRODUCIBILITY.md Short guide to the canonical result folders and reproduction scope.

What is intentionally excluded

• Manuscript source files and submission materials
• Internal planning or archive materials
• Local dataset cache files under dann_benchmark/data/
• Legacy helper scripts that are not needed to inspect or rerun the archived benchmark

Benchmark scope

The public package preserves a controlled comparison among:

• DANN variants: DANN_LRF, DANN_RANDOM, DANN_GRF
• Controls: NAIVE_BRANCH, MLP_PARAM, VANN_SAME
• Full-data tasks: FashionMNIST, KMNIST, CIFAR-10
• Archived reduced-dataset diagnostics
• Separate CPU-only timing runs

The focus is controlled architectural comparison, not state-of-the-art image classification.

Quick start

cd dann_benchmark
python benchmark.py --dataset fashionmnist --epochs 3 --seeds 0 --models dann_lrf naive_branch mlp_param vann_same

If the local cache is missing, the benchmark code will download the required dataset files into dann_benchmark/data/.

Start here

1. dann_benchmark/README.md
2. dann_benchmark/DATASETS.md
3. REPRODUCIBILITY.md
4. stats_outputs/README.md