Noisy Dual Principal Component Pursuit

Authors: Tianyu Ding, Zhihui Zhu, Tianjiao Ding, Yunchen Yang, Rene Vidal, Manolis Tsakiris, Daniel Robinson

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This paper extends the global optimality and convergence theory of DPCP to the case of data corrupted by noise, and further demonstrates its robustness using synthetic and real data.
Researcher Affiliation Academia 1Department of Applied Mathematics & Statistics, Johns Hopkins University, USA 2Mathematical Institute for Data Science, Johns Hopkins University, USA 3School of Information Science and Technology, Shanghai Tech University, China
Pseudocode Yes Algorithm 1 DPCP-PSGM for (3)
Open Source Code No The paper does not provide an explicit statement or link to open-source code for the described methodology.
Open Datasets Yes We use the 3D point clouds from the KITTI dataset (Geiger et al., 2013). In addition to the 7 frames annotated in Zhu et al. 2018a, we further annotate 131 frames. Each point cloud contains around 105 points with approximately 50% outliers.
Dataset Splits No The paper mentions tuning parameters on a 'randomly selected training set of 13 frames' and using 'the rest of the frames for evaluation', but does not explicitly define a separate validation set for hyperparameter tuning.
Hardware Specification Yes Experiments done on a laptop with Intel i7-6700HQ @ 2.6GHz CPU, 16GB 2133MHz DDR4 Memory.
Software Dependencies No The paper mentions parameters for algorithms (e.g., 'The λ of ℓ2,1RPCA is set to 1.92 M', 'µmin for DPCP-PSGD is set to 10 9'), but it does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes We tune the parameters of the algorithms on a randomly selected training set of 13 frames and use the rest of the frames for evaluation. Each method is tuned to achieve an optimal error and then re-tuned to be as fast as possible without exceeding 5% of that error. The λ of ℓ2,1RPCA is set to 1.92 M , the τ of DPCP-d is set to 2.76 N+M , µmin for DPCP-PSGD is set to 10 9, and the relative convergence accuracy, wherever applicable, is set to 10 6.