Granger Components Analysis: Unsupervised learning of latent temporal dependencies
Authors: Jacek Dmochowski
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The technique is tested on scalp electroencephalography (EEG) data from a motor imagery experiment where the resulting components lateralize with the side of the cued hand, and also on functional magnetic resonance imaging (f MRI) data, where the recovered components express previously reported resting-state networks. |
| Researcher Affiliation | Academia | Jacek P. Dmochowski Department of Biomedical Engineering City College of New York New York, NY 10031 jdmochowski@ccny.cuny.edu |
| Pseudocode | Yes | The proposed algorithm is described in Algorithm 1, where X is a D-by-T matrix storing the observed data, Yp is an L-by-T convolution matrix allowing the regression of yp(t) onto X, convmatrix is an operation that produces a convolution matrix, # denotes the Moore-Penrose pseudoinverse, and σ is a small positive constant that is used to randomly initialize the weights of the coefficient vectors. Algorithm 1 Grouped coordinate descent algorithm for GCA. |
| Open Source Code | Yes | The code that was employed to generate the experimental results is available at https://github. com/dmochow/gca. |
| Open Datasets | Yes | The technique was first applied to publicly available EEG data recorded from n = 52 human participants performing a motor imagery task [27]. A previously collected f MRI dataset [38] from n = 20 healthy adults in the resting state (eyes open) was employed here... |
| Dataset Splits | No | The paper does not specify distinct training, validation, or test dataset splits for the real-world EEG or fMRI data. It mentions aggregating covariance matrices across subjects for group analysis. |
| Hardware Specification | No | The paper does not explicitly mention any specific hardware used for running the experiments, such as GPU or CPU models, or cloud computing specifications. It only states that numerical differentiation 'reduced computational time'. |
| Software Dependencies | No | The paper mentions 'MATLAB s fmincon and Python s scipy.optimize were both tested' as software used during development, but it does not provide specific version numbers for these tools or any other software dependencies. |
| Experiment Setup | Yes | The temporal aperture was set to L = 16 samples (0.5 seconds). ...the temporal aperture was set to L = 4 samples ( 11 s) and the number of components to P = 6. wp N(0, σ2I) vp N(0, σ2I) |