Disentangling syntax and semantics in the brain with deep networks

Authors: Charlotte Caucheteux, Alexandre Gramfort, Jean-Remi King

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Functional MRI dataset. We analyze the Narratives public dataset (Nastase et al., 2020), which contains the f MRI measurements of 345 unique subjects listening to narratives.
Researcher Affiliation Collaboration 1Inria, Saclay, France 2Facebook AI Research, Paris, France 3 Ecole normale sup erieure, PSL University, CNRS, Paris, France.
Pseudocode No The paper describes methods through text and figures (e.g., Figure 2 for method to isolate syntactic representations) but does not include explicit pseudocode or algorithm blocks.
Open Source Code No The paper does not provide a direct link to a source code repository or an explicit statement about the release of their own source code for the methodology described.
Open Datasets Yes Functional MRI dataset. We analyze the Narratives public dataset (Nastase et al., 2020), which contains the f MRI measurements of 345 unique subjects listening to narratives.
Dataset Splits Yes f g was fitted on Itrain = 99% of the dataset, and evaluated on Itest = 1% of the left out-data (2.5 min of audio). ...We repeat the procedure 100 times with a 100-fold cross-validation, using scikit-learn KFold without shuffling (Pedregosa et al., 2011).
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU models, CPU types, memory) used to run the experiments.
Software Dependencies No The paper mentions several software packages used, such as 'spa Cy', 'Supar', 'Gector', 'scikit-learn', and 'MNE-Python', but it does not specify their version numbers, which is required for reproducibility.
Experiment Setup Yes We use the linear ridge regression from scikit-learn (Pedregosa et al., 2011), with penalization parameters chosen among 10 values log-spaced between 10 1 and 108 and g was a finite impulse response (FIR) model with 5 delays, following (Huth et al., 2016)." and "To isolate the syntactic representations of GPT-2 , we synthesize, for each sentence of each story, k = 10 sentences with the same syntactic structures (Figure 2).