Hierarchical Methods of Moments

Authors: Matteo Ruffini, Guillaume Rabusseau, Borja Balle

NeurIPS 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on topic modeling show that our method outperforms previous tensor decomposition methods in terms of speed and model quality.
Researcher Affiliation Collaboration Matteo Ruffini Universitat Politècnica de Catalunya Guillaume Rabusseau Mc Gill University Borja Balle Amazon Research
Pseudocode Yes Algorithm 1 SIDIWO: Simultaneous Diagonalization based on Whitening and Optimization; Algorithm 2 Splitting a corpus into two parts
Open Source Code Yes The implementation of the described algorithms can be found a this link: https://github.com/mruffini/Hierarchical-Methods-of-Moments.
Open Datasets Yes We consider the full set of NIPS papers accepted between 1987 and 2015, containing n = 11, 463 papers [28]. [28] Valerio Perrone, Paul A Jenkins, Dario Spano, and Yee Whye Teh. Poisson random fields for dynamic feature models. ar Xiv preprint ar Xiv:1611.07460, 2016.
Dataset Splits No The paper describes generating synthetic data and processing real-world datasets for unsupervised hierarchical clustering, but it does not specify explicit training, validation, and test splits for model training or evaluation in the traditional sense.
Hardware Specification Yes All the experiments were run on a Mac Book Pro with an Intel Core i5 processor.
Software Dependencies No The experiments were performed in Python 2.7, using numpy library for linear algebra operations, with the exception of the implementation of the method from [22], for which we used the author s Matlab implementation. While software is mentioned, specific version numbers for numpy or Matlab are not provided.
Experiment Setup Yes We generate 400 samples according to this model and we iteratively run Algorithm 2 to create a hierarchical binary tree with 8 leafs. ...we keep the d = 3000 most frequent words as vocabulary and we iteratively run Algorithm 2 to create a binary tree of depth 4. ...relevance [29] of a word w 2 Cnode C is defined by r(w, Cnode) = λ log P[w|Cnode] + (1 λ) log P[w|C]) where the weight parameter is set to λ = 0.7