Coupling Artificial Neurons in BERT and Biological Neurons in the Human Brain

Authors: Xu Liu, Mengyue Zhou, Gaosheng Shi, Yu Du, Lin Zhao, Zihao Wu, David Liu, Tianming Liu, Xintao Hu

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results demonstrate 1) The activations of ANs and BNs are significantly synchronized; 2) the ANs carry meaningful linguistic/semantic information and anchor to their BN signatures; 3) the anchored BNs are interpretable in a neurolinguistic context. Overall, our study introduces a novel, general, and effective framework to link transformer-based NLP models and neural activities in response to language and may provide novel insights for future studies such as brain-inspired evaluation and development of NLP models.
Researcher Affiliation Academia Xu Liu1*, Mengyue Zhou1*, Gaosheng Shi1*, Yu Du1, Lin Zhao2, Zihao Wu2, David Liu3, Tianming Liu2, Xintao Hu1 1 School of Automation, Northwestern Polytechnical University 2 School of Computing, University of Georgia 3 Athens Academy {liu xu, zhou my, 2021202420, dddyyy}@mail.nwpu.edu.cn, {lin.zhao, zw63397}@uga.edu, {david.weizhong.liu, tianming.liu}@gmail.com, xhu@nwpu.edu.cn
Pseudocode No The paper describes its methods textually and with diagrams (e.g., Fig. 1, 2, 3), but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper cites and links to several third-party tools used in the research (e.g., SPM, VS-DBN, Hugging Face's BERT, spaCy, Text-Attention-Heatmap Visualization tool) but does not provide a link to the authors' own source code implementation for the coupling framework described in the paper.
Open Datasets Yes We use the Narratives f MRI dataset (Nastase et al. 2021) in this study.
Dataset Splits No The paper describes data preprocessing and parameters for training the VS-DBN (e.g., hidden units, learning rate, batch size, epochs), but it does not explicitly mention train/validation/test splits for the primary task of coupling ANs and BNs or for the fMRI data beyond the VS-DBN training.
Hardware Specification Yes Model training is performed on a workstation with 10 Ge Force 1080Ti GPUs.
Software Dependencies No The paper mentions several software tools and libraries (e.g., SPM, VS-DBN, Hugging Face's BERT, spaCy, text attention heatmap visualization tool) but does not provide specific version numbers for these software components, which is necessary for reproducible software dependencies.
Experiment Setup Yes All the volumes in the two sessions are aggregated to train the VS-DBN with the following parameters: 512/256/128 hidden units in the 1st/2nd/3rd RBM layer, Gaussian (zeromean and a standard deviation of 0.01) initialization, learning rate 0.001/0.0005/0.0005, batch-size 20, L1 weightdecay rate 0.001/0.00005/0.00005, 100 training epochs, batch normalization.