Unsupervised Manifold Alignment with Joint Multidimensional Scaling
Authors: Dexiong Chen, Bowen Fan, Carlos Oliver, Karsten Borgwardt
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate the effectiveness of our approach in several applications, including joint visualization of two datasets, unsupervised heterogeneous domain adaptation, graph matching, and protein structure alignment. The paper presents empirical evaluations in Section 5 with tables and figures showing performance metrics. |
| Researcher Affiliation | Academia | Dexiong Chen, Bowen Fan, Carlos Oliver & Karsten Borgwardt Department of Biosystems Science and Engineering, ETH Z urich, Switzerland SIB, Swiss Institute of Bioinformatics, Switzerland |
| Pseudocode | Yes | Algorithm 1 Joint Multidimensional Scaling |
| Open Source Code | Yes | The implementation of our work is available at https://github.com/Borgwardt Lab/Joint MDS. |
| Open Datasets | Yes | The synthetic datasets respectively consist of a bifurcation, a Swiss roll and a circular frustum from (Liu et al., 2019). Each synthetic dataset has 300 samples... The datasets can be downloaded from https://noble.gs.washington.edu/proj/mmd-ma/. The digit dataset MNIST-USPS contain two subsets of MNIST and USPS respectively. (Deng, 2012; Hull, 1994). For the graph matching task, the datasets can be downloaded at https://github.com/Hongteng Xu/s-gwl and https://github.com/Hongteng Xu/gwl respectively. |
| Dataset Splits | No | While a 75% training and 25% testing split is mentioned for one specific task (MIMIC-III graph matching), the paper does not consistently provide detailed training, validation, and test splits (e.g., percentages, sample counts, or explicit validation sets) for all datasets used in its experiments to ensure full reproducibility of data partitioning. |
| Hardware Specification | Yes | All experiments are run on a single Macbook Pro 2020 laptop with a 2 GHz Quad-core Intel core i5 CPU. |
| Software Dependencies | No | The paper mentions various algorithms and software components (e.g., SMACOF, Sinkhorn's algorithm, k-NN classifier, linear SVM classifier) but does not provide specific version numbers for any software dependencies required to replicate the experiments. |
| Experiment Setup | Yes | We fix the number of components d to 2 to visualize datasets in R2. We fix the matching penalization parameter λ to 0.1 for all the datasets. Then, we train a k-NN classifier (k is fixed to 5) on the source domain... For the more complex MNIST-USPS dataset, we use a linear SVM classifier with the regularization parameter set to 1 instead, which results in better prediction accuracy. The entropic regularization parameter ε for OT is fixed to 1.0 with an annealing decay equal to 0.95. |