Ultrametric Fitting by Gradient Descent

Authors: Giovanni Chierchia, Benjamin Perret

NeurIPS 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed cost functions on synthetic and real datasets, and we show that they perform as good as Ward method and semi-supervised SVM.
Researcher Affiliation Academia Giovanni Chierchia Université Paris-Est, LIGM (UMR 8049) CNRS, ENPC, ESIEE Paris, UPEM F-93162, Noisy-le-Grand, France giovanni.chierchia@esiee.fr
Pseudocode Yes Algorithm 1 Solution to the ultrametric fitting problem defined in (4). Algorithm 2 Subdominant ultrametric operator defined in (5) with (14).
Open Source Code Yes Our code is made publicly available at https://github.com/Perret B/ultrametric-fitting.
Open Datasets Yes We evaluate the proposed optimization framework on five datasets downloaded from the LIBSVM webpage,3 whose size ranges from 270 to 1500 samples. For each dataset, we build a 5-nearest-neighbor graph, to which we add the edges of a minimum spanning tree to ensure the connectivity. https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/
Dataset Splits No The paper mentions a "10-fold scheme" for cross-validation and that "cross-validated performance is reported," but it does not specify explicit training/validation/test dataset splits where a separate validation set is used for hyperparameter tuning or model selection.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions "Higra [44] and Py Torch [45] libraries" as dependencies but does not specify their version numbers.
Experiment Setup Yes The "Dasgupta" method refers to Algorithm 1 with JDasgupta + λJsize and λ = 1. The "Closest+Size" method refers to Algorithm 1 with the cost function Jclosest + λJsize and λ = 10. In both cases, the regularization is only applied to the top-10 dendogram nodes (see supplemental material). The "Closest+Triplet" method refers to Algorithm 1 with Jclosest + λJtriplet, λ = 1 and α = 10.