Efficient Riemannian Meta-Optimization by Implicit Differentiation

Authors: Xiaomeng Fan, Yuwei Wu, Zhi Gao, Yunde Jia, Mehrtash Harandi3733-3740

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

Reproducibility Variable Result LLM Response
Research Type Experimental Evaluations of three optimization problems on different Riemannian manifolds show that our method achieves state-of-the-art performance in terms of the convergence speed and the quality of optima. Experiments were conducted on three tasks: principal component analysis (PCA) on the Grassmann manifold, face Recognition on the Stiefel Manifold, and clustering on the SPD manifold...
Researcher Affiliation Collaboration 1 Beijing Laboratory of Intelligent Information Technology School of Computer Science, Beijing Institute of Technology, Beijing, China 2 Department of Electrical and Computer Systems Eng., Monash University, and Data61, Australia
Pseudocode Yes Algorithm 1 Parameter Warmup stage
Open Source Code Yes The code is available at https://github.com/XiaomengFanmcislab/I-RMM.
Open Datasets Yes We used MNIST dataset to evaluate our method on the PCA task. We utilized the Yale B dataset (Lee, Ho, and Kriegman 2005) to conduct this experiment. We also conducted experiments on the clustering task of SPD representations by utilizing the Kylberg texture dataset (Kylberg 2011).
Dataset Splits Yes LV (X(T )) is the loss function of the updated Riemannian parameter X(T ) on validation data.
Hardware Specification No The paper mentions 'GPU memory consumption' but does not provide specific details on the hardware used (e.g., GPU model, CPU, RAM).
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes Require: Initial optimization state S(0) = 0, initial parameters ϕ of our optimizer, maximum iteration T of the inner-loop, maximum iteration Υ of the outer-loop, and hyperparameter B to update the parameter pool. Table 1: Training time (seconds) comparisons on the PCA task (showing specific Inner Loop Steps values).