Reproduce “FDSML: Formula-Driven Supervised Metric Learning with Parameter-Aware Triplet Loss” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hina Otake, Shinei Arakawa, Keitaro Tanaka, Yoshihiro Fukuhara, Hirokatsu Kataoka, Shigeo Morishima
FDSML uses parameter distance to select positives and negatives among IFS-generated fractals, training an image encoder with triplet loss rather than relying only on class labels. On ImageNet-100, it improves ViT-Tiny Top-1 accuracy from 84.40% to 87.93% on the same FractalDB-1k dataset. For retrieval, Cars Recall@1 reaches 78.4% versus 56.2% for FDSL, below ImageNet pre-training’s 87.7%; tests use one million images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.