helloTC / SemanticRelation

Code for paper: Semantic relatedness emerges in deep convolutional neural networks designed for object recognition

SemanticRelation

Code for paper: Semantic relatedness emerges in deep convolutional neural networks designed for object recognition

Code

1. Prepare WordNet semantic similarity: WordNet_preparation.py
   
2. Extraction of DCNN activation: extract_imagenet_activation.py
   
3. Stability of representational similarity for different implementations: DCNN_consistency.py
   
4. Model training: train_model.py
   
5. Developmental trajectory: corr_traincatesim.py, merge_mat.py and corr_traincatesim_singleobj.py
   
6. Examine no parent-child relationship among the 1000 categories: examine_no_conceptual_relationship.py
   
7. Effect of task demand on the emergence of semantic relatedness: taskdemand_comparecorr.py
   

Files

Data was provided in folder: data.

• cate_pathsim_wup.npy: WordNet semantic similarity
  
• hypernyms_idx.npy: rearrange categories according to the WordNet hierarchy (For detail please refer to WordNet_preparation.py)
  
• DCNNsim: representational similarity of different DCNNs.
  
• AlexNetsim_layers: Representational similarity of AlexNet of each layers. Noted that validation_corr_alexnet_fc3.npy is same to validation_corr_alexnet_fc.npy from DCNNsim
  
• DevelopTraj: developmental trajectory for the original AlexNet and the single-object AlexNet. Values from them are correspondences to the WordNet semantic similarity in different training stages.
• TaskDemand: Averaged responses of AlexNet-Cate2, AlexNet-Cate19 and AlexNet-Cate1000 in different layers. Responses are formatted as [Category x Units].

This paper has been published in Frontiers in Computational Neuroscience: Huang T, Zhen Z and Liu J (2021) Semantic Relatedness Emerges in Deep Convolutional Neural Networks Designed for Object Recognition. Front. Comput. Neurosci. 15:625804. doi: 10.3389/fncom.2021.625804

Feel free to contact me if you have any questions :)

This paper has finally published in Frontiers in Computational Neuroscience doi: 10.3389/fncom.2021.625804