Math ANN encoding/decoding data & code

Here we provide analysis scripts for "Artificial neural network modelling of the neural population code underlying mathematical operations".

The preprocessed fMRI data are provided elsewhere (https://doi.org/10.5281/zenodo.6605258).

The analysis scripts require MATLAB (R2019b) and FreeSurfer.

Ridge regression

ID = 'sub-01';
Method = 'Transformer';
Code = 6;
%1-6: Average Q vectors (encoding layer 1 to layer 6)
%7-12: Average K vectors
%13-18: Average V vectors
%19-24: Average R vectors
%25-27: Q vector of each symbol, single operator problems
%28: Q vector of the result value, single operator problems
%29: Average Q vector, single operator problems
%30-34: Q vector of each symbol, double operator problems
%35: Q vector of the intermediate result value, double operator problems
%36: Q vector of the final result value, double operator problems
%37: Average Q vector, double operator problems

Ridge_FormOnly(ID, Method, Code)
FDRcorr_FormOnly(ID, Method, Code)

Ridge regression with non-math regressors:

Ridge_FormOnly_Reg(ID, Method, Code)
FDRcorr_FormOnly_Reg(ID, Method, Code)

Representational similarity analysis 

Method = 'Transformer'; %Method = 'ME', %Method = 'Word2Vec'
Code = 6; %Code = 1, %Code = 1
RSMwithFeat(Method, Code)

Feature brain similarity analysis

FeatBrainSim_ModelCompare
FeatBrainSim_TransformerCompare

Decoding analysis

Ridge_Decoding_FormOnly_New(ID, Method, Code)
Ridge_Decoding_FormOnly_Single2Double(ID, Method, Code)
Ridge_Decoding_FormOnly_Double2Single(ID, Method, Code)
Ridge_Decoding_FormOnly_NonAdd2Add(ID, Method, Code)
Ridge_Decoding_FormOnly_NonSub2Sub(ID, Method, Code)
Ridge_Decoding_FormOnly_NonMul2Mul(ID, Method, Code)
Ridge_Decoding_FormOnly_NonDiv2Div(ID, Method, Code)

If you have any questions, please send an email to nakai.tomoya [at] neuro.mimoza.jp.