Embedded Files
Acknowledgment: "This material is based upon work supported by the National Science Foundation under Grant No. 1617107 and No. 1617497."
Acknowledgment: We acknowledge NVIDIA's Titan X Pascal GPU donation for this research.
More information about this project can be accessed via here.
1 Complete Visualization and Sonification For CNN Neurons
1 Complete Visualization and Sonification For CNN Neurons
1.1 Recording tower visualization: pretraining only, w/o pretraining, and pretraining + finetuning
1.1 Recording tower visualization: pretraining only, w/o pretraining, and pretraining + finetuning
Right click and download the pictures to see details.
(a) Recording tower Conv1 visualization: pretraining only
(b) Recording tower Conv2 visualization: pretraining only
(c) Recording tower Conv3 visualization: pretraining only
(a) Recording tower Conv1 visualization: w/o pretraining
(b) Recording tower Conv2 visualization: w/o pretraining
(c) Recording tower Conv3 visualization: w/o pretraining
(a) Recording tower Conv1 visualization: pretraining + finetuning
(b) Recording tower Conv2 visualization: pretraining + finetuning
(c) Recording tower Conv3 visualization: pretraining + finetuning
Figure 1. Visualizations of input patterns that maximize neurons in Conv1, Conv2, and Conv3 of recording tower
1.2 Recording tower sonification: pretraining + finetuning
1.2 Recording tower sonification: pretraining + finetuning
1.3 Imitation tower: pretraining only, w/o pretraining, and pretraining + finetuning
1.3 Imitation tower: pretraining only, w/o pretraining, and pretraining + finetuning
Right click and download the pictures to see details.
(a) Imitation tower Conv1 visualization: pretraining only
(b) Imitation tower Conv2 visualization: pretraining only
(c) Imitation tower Conv3 visualization: pretraining only
(a) Imitation tower Conv1 visualization: w/o pretraining
(b) Imitation tower Conv2 visualization: w/o pretraining
(c) Imitation tower Conv3 visualization: w/o pretraining
(a) Imitation tower Conv1 visualization: pretraining + finetuning
(b) Imitation tower Conv2 visualization: pretraining + finetuning
(c) Imitation tower Conv3 visualization: pretraining + finetuning
Figure 2. Visualizations of input patterns that maximize neurons in Conv1, Conv2, and Conv3 of imitation tower
1.4 Imitation tower sonification: pretraining + finetuning
1.4 Imitation tower sonification: pretraining + finetuning
2 Complete Visualization and Sonification for FC neurons
2 Complete Visualization and Sonification for FC neurons
2.1 Visualization for Imitation-Recording Pairs
2.1 Visualization for Imitation-Recording Pairs
Right click and download the pictures to see details.
(a) Imitation tower input patterns: without pretraining
(b) Recording tower input patterns: w/o pretraining
(c) Imitation tower input patterns: pretraining + finetuning
(d) Recording tower input patterns: pretraining + finetuing
Figure 3. Visualizations of recording-imitation pattern pairs that maximize neurons in FC layer, without pretraining (The up-left 1st pattern in (a) is paired with the up-left 1st pattern in (b), etc.)
2.2 Sonification for Imitation-Recording Pairs
2.2 Sonification for Imitation-Recording Pairs
Google Sites
Report abuse