Nonlinear System Identification via Tensor Completion
Authors: Nikos Kargas, Nicholas D. Sidiropoulos4420-4427
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we demonstrate the effectiveness of the approach using several regression tasks including some standard benchmarks and a challenging student grade prediction task. Our proposed approach is implemented in MATLAB using the Tensor Toolbox (Bader and Kolda 2007) for tensor operations. We then assess the ability of our model to handle missing predictors by hiding 30% of the data as well as its ability to predict vector valued responses. For each experiment we split the dataset into two sets, 80% used for training and 20% for testing, and run 10 Monte-Carlo simulations. Finally, we evaluate the performance of our approach in a challenging student grade prediction task using a real student grade data. |
| Researcher Affiliation | Academia | Nikos Kargas,1 Nicholas D. Sidiropoulos2 1Department of ECE, University of Minnesota, MN, USA 2Department of ECE, University of Virginia, VA, USA karga005@umn.edu, nikos@virginia.edu |
| Pseudocode | No | The paper describes the Alternating Least Squares (ALS) algorithm and its mathematical updates in detail, but it does not include a formal pseudocode block or algorithm box. |
| Open Source Code | No | The paper states: "Our proposed approach is implemented in MATLAB using the Tensor Toolbox (Bader and Kolda 2007) for tensor operations." This indicates the tools used but does not provide open-source code for the authors' specific implementation or methodology. |
| Open Datasets | Yes | We evaluate the proposed approach in single output regression tasks using several datasets obtained from the UCI machine learning repository (Lichman 2013). We used four different machine learning algorithms as baselines, Ridge Regresion (RR), Support Vector Regression (SVR), Decision Tree (DT) and Multilayer Perceptrons (MLPs) using the implementation of scikit-learn (Pedregosa et al. 2011). |
| Dataset Splits | Yes | For each experiment we split the dataset into two sets, 80% used for training and 20% for testing, and run 10 Monte-Carlo simulations. For each method we tune the hyper-parameters using 5-fold cross-validation. |
| Hardware Specification | No | The paper mentions general computational power but does not specify any particular hardware components like CPU models, GPU models, or memory sizes used for the experiments. |
| Software Dependencies | No | Our proposed approach is implemented in MATLAB using the Tensor Toolbox (Bader and Kolda 2007) for tensor operations. We used four different machine learning algorithms as baselines... using the implementation of scikit-learn (Pedregosa et al. 2011). The paper names software components but does not provide specific version numbers for MATLAB, Tensor Toolbox, or scikit-learn. |
| Experiment Setup | Yes | For our method, we fix the alphabet size to be I = 25 and use Lloyd Max scalar quantizer for discretization of continuous predictors. For the MLPs, we set the number of hidden layers to 1,3 or 5 and varied the number of nodes per layer 10, 50, 100 and 250. For RR, SVR and MLP we standardize each ordinal feature such that it has zero mean and unit variance. Categorical features are transformed using one-hot encoding. |