Multi-Mode Tensor Space Clustering Based on Low-Tensor-Rank Representation

Authors: Yicong He, George K. Atia6893-6901

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that the proposed MMTSC algorithm can in many cases outperform existing clustering algorithms. In this section, we carry out experiments to verify the performance of the proposed method. We compare the performance with several existing clustering methods, including matrix-based methods LRR, SSC, SSC-PZF (Yang, Robinson, and Vidal 2015), the sequential OSC method (Tierney, Gao, and Guo 2014), and the tensor-based method t SVD-TLRR (Zhou et al. 2019). We use the clustering accuracy (percentage of correctly classified subtensors) for evaluation with synthetic data. For real data, the performance is evaluated using the accuracy and the normalized mutual information (NMI) between the clustering results and the ground truth labels.
Researcher Affiliation Academia Yicong He1, George K. Atia1,2 1Department of Electrical and Computer Engineering, University of Central Florida 2Department of Computer Science, University of Central Florida {yicong.he, george.atia}@ucf.edu
Pseudocode Yes Algorithm 1: Multi-mode tensor space clustering (MMTSC)
Open Source Code Yes The code is available at https://github.com/he1c/LTRR-Tensor SC.
Open Datasets Yes YUV dataset1 contains a collection of color videos from different scenes. For each video, a sequence consisting of the first 50 frames is selected. Each frame is converted to grayscale and down-sampled to 36 44, yielding a subtensor of size 36 44 50. 1http://trace.eas.asu.edu/yuv/ UCSD dataset2 provides 18 short videos... 2http://www.svcl.ucsd.edu/projects/background subtraction/ Coil20 dataset3 is a collection of grayscale images... 3https://cs.columbia.edu/CAVE/software/softlib/coil-20.php
Dataset Splits No The paper uses Monte Carlo (MC) runs with different random selections and observation rates but does not specify explicit train/validation/test dataset splits (e.g., fixed percentages or sample counts for each split).
Hardware Specification Yes All experiments were performed using MATLAB R2018b on a desktop PC with a 2.6-GHz processor and 16GB of RAM.
Software Dependencies Yes All experiments were performed using MATLAB R2018b on a desktop PC with a 2.6-GHz processor and 16GB of RAM.
Experiment Setup Yes The parameters for MMTSC are set to λ = 0.1M, µ = 0.01, β = 0.1, ρ = 1.1, ϵ = 10 3.