Brain-Mediated Transfer Learning of Convolutional Neural Networks
Authors: Satoshi Nishida, Yusuke Nakano, Antoine Blanc, Naoya Maeda, Masataka Kado, Shinji Nishimoto5281-5288
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate that our brain-mediated TL (BTL) shows higher performance in the label estimation than the standard TL. In addition, we illustrate that the estimations mediated by different brains vary from brain to brain, and the variability reflects the individual variability in perception. |
| Researcher Affiliation | Collaboration | Satoshi Nishida,1 Yusuke Nakano,1 Antoine Blanc,1 Naoya Maeda,2 Masataka Kado,2 Shinji Nishimoto1 1Center for Information and Neural Networks (Ci Net), National Institute of Information and Communications Technology (NICT) and Osaka University, Osaka, Japan 2NTT DATA Corporation, Tokyo, Japan |
| Pseudocode | No | The paper describes the modeling procedure in prose and refers to Figure 1 for a schematic, but does not provide structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide an explicit statement or a link to open-source code for the methodology described. It mentions supplemental material available at https://osf.io/3hkwd, but does not state this includes the code. |
| Open Datasets | No | The paper mentions that two sets of movies were 'provided by NTT DATA Corp.' and describes them, but does not provide a direct link, DOI, or formal citation for public access to these specific movie datasets. While supplemental material is linked (https://osf.io/3hkwd), it's not explicitly stated to contain the full datasets for public use. |
| Dataset Splits | Yes | The optimal regularization parameter for each model is determined by 10fold cross validation of training data and shared across all voxels. |
| Hardware Specification | Yes | f MRI responses to the movies were collected from Japanese participants using a 3T MRI scanner. |
| Software Dependencies | No | The paper mentions using VGG-16 and Sound Net, and word2vec, but does not provide specific version numbers for these software components or any other libraries/frameworks used for implementation. |
| Experiment Setup | Yes | The optimal regularization parameter for each model is determined by 10fold cross validation of training data and shared across all voxels. In this study, the top 300 PCs are used as voxels to predict (i.e., N = 300) for only one estimation task... In addition, the top 10 PCs with the highest prediction accuracy are used for the regressors of the vox2vox model (i.e., M = 10). To extract visual features from movies via VGG-16, which was originally applied to static images with a fixed size of 224 224 pixels, the movies are decomposed into frames and resized to the same size. Then, unit activations of intermediate layers when inputting the movie frames are calculated and pooled for each second. Finally, the maximum activation value of each unit for each second is used as the visual features of the movies. |