ATD: Augmenting CP Tensor Decomposition by Self Supervision

Authors: Chaoqi Yang, Cheng Qian, Navjot Singh, Cao (Danica) Xiao, M Westover, Edgar Solomonik, Jimeng Sun

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our proposed ATD on multiple datasets. It can achieve 0.8% 2.5% accuracy gain over tensor-based baselines. Also, our ATD model shows comparable or better performance (e.g., up to 15% in accuracy) over self-supervised and autoencoder baselines while using less than 5% of learnable parameters of these baseline models.
Researcher Affiliation Collaboration 1University of Illinois Urbana-Champaign, 2IQVIA, 3Relativity, 4Massachusetts General Hospital, 5Harvard Medical School 1{chaoqiy2,navjot2,solomon2,jimeng}@illinois.edu, 2alextoqc@gmail.com, 3cao.xiao@relativity.com, 4,5mwestover@mgh.harvard.edu
Pseudocode Yes Algorithm 1: Alternating Least Squares
Open Source Code Yes We have released our data processing and codes in https://github.com/ycq091044/ATD.
Open Datasets Yes We use four real-world datasets: (i) Sleep-EDF (Kemp et al., 2000), (ii) human activity recognition (HAR) (Anguita et al., 2013), (iii) Physikalisch Technische Bundesanstalt large scale cardiology database (PTB-XL) (Alday et al., 2020), (iv) Massachusetts General Hospital (MGH) (Biswal et al., 2018) datasets.
Dataset Splits Yes All datasets are split into three disjoint sets (i.e., unlabeled, training and test) by subjects, while training and test sets have labels. Basic statistics are shown in Table 1.
Hardware Specification Yes The experiments are implemented by Python 3.8.5, Torch 1.8.0+cu111 on a Linux workstation with 256 GB memory, 32 core CPUs (3.70 GHz, 128 MB cache), two RTX 3090 GPUs (24 GB memory each).
Software Dependencies Yes The experiments are implemented by Python 3.8.5, Torch 1.8.0+cu111 on a Linux workstation with 256 GB memory, 32 core CPUs (3.70 GHz, 128 MB cache), two RTX 3090 GPUs (24 GB memory each).
Experiment Setup Yes All models (baselines and our ATD) use the same augmentation techniques: (a) jittering, (b) bandpass filtering, and (c) 3D position rotation. We provide an ablation study on the augmentation methods in Appendix C.5. ... For tensor based models, we use R = 32 and implement the pipeline in CUDA manually, instead of using torch-autograd.