DriPP: Driven Point Processes to Model Stimuli Induced Patterns in M/EEG Signals
Authors: Cédric Allain, Alexandre Gramfort, Thomas Moreau
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Simulations reveal that model parameters can be identified from long enough signals. Results on standard MEG datasets demonstrate that our methodology reveals event-related neural responses both evoked and induced and isolates non-task-specific temporal patterns. (Abstract) and We evaluated our model on several experiments, using both synthetic and empirical MEG data. (Section 4). |
| Researcher Affiliation | Academia | Cédric Allain, Alexandre Gramfort & Thomas Moreau Université Paris-Saclay, Inria, CEA, Palaiseau, 91120, France |
| Pseudocode | Yes | The pseudocode of the algorithm is presented in Algorithm 1. |
| Open Source Code | No | The paper does not explicitly state that the source code for the described methodology (Dri PP) is openly available. It only mentions the use of existing open-source libraries: We used Python (Python Software Foundation, 2019) and its scientific libraries (Virtanen et al., 2020; Hunter, 2007; Harris et al., 2020). We relied on alphacsc for CDL with rank-1 constraints on MEG (Dupré la Tour et al., 2018) and we used MNE (Gramfort et al., 2013) to load and manipulate the MEG datasets. |
| Open Datasets | Yes | Datasets Experiments on MEG data were run on two datasets from MNE Python package (Gramfort et al., 2014; 2013): the sample dataset and the somatosensory (somato) dataset1. These datasets were selected as they elicit two distinct types of event-related neural activations: evoked responses which are time locked to the onset of the driver process, and induced responses which exhibit random jitters. Complementary experiments were performed on the larger Cam-CAN dataset (Shafto et al., 2014)2. 1Both available at https://mne.tools/stable/overview/datasets_index.html 2Available at https://www.cam-can.org/index.php?content=dataset |
| Dataset Splits | No | The paper does not explicitly provide details about training, validation, and test splits for the data used to train its model. It mentions generating synthetic data for evaluation of parameter recovery and preprocessing steps for MEG data but not how data was partitioned for model training and evaluation. |
| Hardware Specification | Yes | Computations were run on CPU Intel(R) Xeon(R) E5-2699, with 44 physical cores. |
| Software Dependencies | Yes | We used Python (Python Software Foundation, 2019) and its scientific libraries (Virtanen et al., 2020; Hunter, 2007; Harris et al., 2020). We relied on alphacsc for CDL with rank-1 constraints on MEG (Dupré la Tour et al., 2018) and we used MNE (Gramfort et al., 2013) to load and manipulate the MEG datasets. |
| Experiment Setup | Yes | For both datasets, only the 204 gradiometer channels are analyzed. The signals are pre-processed using high-pass filtering at 2 Hz to remove slow drifts in the data, and are resampled to 150 Hz to limit the atom size in the CDL. CDL is computed using alphacsc (Dupré la Tour et al., 2018) with the Greedy CDL method. For the sample dataset, 40 atoms of duration 1 s each are extracted, and for the somato dataset, 20 atoms of duration 0.53 s are estimated. The extracted atoms activations are binarized using a threshold of 6 10 11 (resp. 1 10 10) for sample (resp. somato), and the times of the events are shifted to make them correspond to the peak amplitude time of the atom. Then, for every atom, the intensity function is estimated using the EM-based algorithm with 400 iterations and the smart start initialization strategy. Kernels truncation values are hyper-parameters for the EM and thus must be pre-determined. The upper truncation value b is chosen smaller than the minimum ISI. Here, we used in addition some previous domain knowledge to set coherent values for each dataset. Hence, for the sample (resp. somato) dataset, kernel support is fixed at [0.03 s ; 0.5 s] (resp. [0 s ; 2 s]). |