Generalizable and Robust Spectral Method for Multi-view Representation Learning
Authors: Amitai Yacobi, Ofir Lindenbaum, Uri Shaham
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our extensive experiments demonstrate that Spec Ra GE outperforms state-of-the-art methods, particularly in scenarios with data contamination, paving the way for more reliable and efficient multi-view learning. Our extensive experiments demonstrate that Spec Ra GE not only achieves state-of-the-art performance on standard multi-view benchmarks but also significantly outperforms existing methods when dealing with outliers and noisy views. |
| Researcher Affiliation | Academia | Amitai Yacobi EMAIL Department of Computer Science Bar-Ilan University Ofir Lindenbaum EMAIL Faculty of Engineering Bar-Ilan University Uri Shaham EMAIL Department of Computer Science Bar-Ilan University |
| Pseudocode | Yes | A complete algorithm summarizing our method can be found in Alg. 1. Algorithm 1 Spec Ra GE |
| Open Source Code | Yes | Our code is available at: https: //github.com/shaham-lab/Spec Ra GE. |
| Open Datasets | Yes | Datasets. We assess the performance of Spec Ra GE using five well-studied multi-view datasets. Our selection prioritizes datasets that exhibit diversity in the types and number of views, as well as the number of classes, as depicted in Table 3 in Appendix A. The datasets are listed as follows: (1) BDGP (Cai et al., 2012) contains 2500 images of Drosophila embryos divided into five categories with two extracted features... (3) Caltech20 is a subset of 2386 examples derived from the object recognition dataset (Fei-Fei et al., 2004)... (4) Handwritten (Asuncion & Newman, 2007) contains 2,000 digital images of handwritten numerals... (5) Infinite MNIST is a large-scale variant of MNIST (Le Cun et al., 1998)... |
| Dataset Splits | Yes | Data Split. For each dataset, we initially divide it into an 80% training set and a 20% testing set. Subsequently, for training, we further divide the training set into a 90% training subset and a 10% validation subset. |
| Hardware Specification | Yes | OS and Hardware. The training procedures were executed on both mac OS Sequoia (15.0) and Rocky Linux 9.3, utilizing Mac Book M1 processor and Nvidia GPUs including Ge Force GTX 1080 Ti and A100 80GB PCIe. |
| Software Dependencies | No | The paper mentions operating systems (mac OS Sequoia (15.0), Rocky Linux 9.3) and hardware, but does not specify software dependencies like programming languages or libraries with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | Hyper-parameters. In Table 6, we provide a breakdown of the hyperparameters utilized for the various datasets. We ensured that the same hyperparameter tuning procedure was applied to all baselines for a fair comparison... The initial learning rate (LR) was uniformly set to 10 3 for all datasets, with a decay policy in place... Adam optimizer is used for training. |