Towards Sentence-Level Brain Decoding with Distributed Representations

Authors: Jingyuan Sun, Shaonan Wang, Jiajun Zhang, Chengqing Zong7047-7054

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

Reproducibility Variable Result LLM Response
Research Type Experimental We carry out a systematic evaluation, covering both widely-used baselines and state-of-the-art sentence representation models. We demonstrate how well different types of sentence representations decode the brain activation patterns and give empirical explanations of the performance difference.
Researcher Affiliation Academia 1National Laboratory of Pattern Recognition, CASIA, Beijing, China 2University of Chinese Academy of Sciences, Beijing, China 3CAS Center for Excellence in Brain Science and Intelligence Technology, Beijing, China
Pseudocode No The paper does not contain any sections or figures explicitly labeled 'Pseudocode' or 'Algorithm', nor does it present any structured algorithm blocks.
Open Source Code No The paper references third-party code (e.g., Infer Sent, Skip-thought) and experiment setup details online, but does not provide concrete access to the source code for the methodology or analysis developed in this paper.
Open Datasets Yes We experiment with the f MRI neural activation data published by Pereira et al. (2018), acquired on a whole-body 3-Tesla Siemens Trio scanner with a 32-channel head coil.
Dataset Splits Yes The regression model is trained and tested on different subsets of the 176 passages (627 sentences) in a 5-fold cross-validation for each participant. We use 141 passages for training and 35 passages for testing in each fold.
Hardware Specification No The paper mentions the fMRI scanner used for data acquisition ('3-Tesla Siemens Trio scanner with a 32-channel head coil') but does not specify the computational hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions various models and techniques (e.g., Ridge Regression, Infer Sent) but does not provide specific version numbers for any software libraries or dependencies used in their implementation or experiments.
Experiment Setup Yes Formally, regression models are trained on each voxel and its 26 neighbors in 3D to predict each dimension of the sentence representations. The correlation between predicted values and the ground truth sentence representations is then calculated. We take the mean correlation across all dimensions as a voxel s informativeness score. The 5,000 voxels with highest scores are selected. The regression model is trained and tested on different subsets of the 176 passages (627 sentences) in a 5-fold cross-validation for each participant. We use 141 passages for training and 35 passages for testing in each fold. λ is the regularization parameter separately set for each dimension with cross-validation.