Reduced-Rank Linear Dynamical Systems

Authors: Qi She, Yuan Gao, Kai Xu, Rosa Chan

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Results on both simulated and experimental data demonstrate our model can robustly learn latent space from short-length, noisy, count-valued data and significantly improve the prediction performance over the state-of-the-art methods.
Researcher Affiliation Collaboration 1Princeton Neuroscience Institute, Princeton University 2Tencent AI Lab 3Department of Electronic Engineering, City University of Hong Kong 4Department of Computer Science, Princeton University 5School of Computer Science, National University of Defense Technology
Pseudocode Yes Algorithm 1 Framework of inference and learning (VBEM)
Open Source Code Yes We implement RRLDS in Matlab(2017a), and our code is available at https://github.com/sheqi/RRLDS
Open Datasets Yes We also evaluated our method on two experimental hippocampus datasets (Mizuseki et al. 2009).
Dataset Splits Yes β1,β2 are selected (in all experiments) by the internal cross validation while optimizing model s predictive performance. ... The length is 500 for training data and 100 for testing data.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies Yes We implement RRLDS in Matlab(2017a)
Experiment Setup Yes β1,β2 are selected (in all experiments) by the internal cross validation while optimizing model s predictive performance. ... We select the step size to assure fast convergence rate based on Theorem 1 and proof is in the supplementary material. ... In practice, we initialize our parameters using Laplace-EM algorithm (Buesing et al. 2014), which empirically gives runtime advantages, and produces a sensible optimum.