kernel-brain-data
This is a repository to leverage kernel brain data to detect laughter.
The source code is available on github as kernel-brain-data.
About the Neural Network Model
This model will take an image of the kernel brain and determine whether the individual is actively laughing.
The Kernel Neural Image model Convolutional Neural Network alone achieves accurate results on predicting laughter vs. non-laughter when an input image of the live kernel brain is used as input to the network. The model uses pre-trained weights from resnet-18 as well as frames from the Lex Fridman podcast.
The metric results of the model performance are below, and the model is publicly available for download and use.
Metrics
| Class | Precision | Recall | F1-score | Support |
|---|---|---|---|---|
| Non-Laughter | 0.89 | 0.66 | 0.76 | 267 |
| Laughter | 0.71 | 0.92 | 0.80 | 251 |
| Accuracy | 0.78 | 518 | ||
| Macro Avg | 0.80 | 0.79 | 0.78 | 518 |
| Weighted Avg | 0.81 | 0.78 | 0.78 | 518 |
| Metric | Value |
|---|---|
| Accuracy | 0.7819 |
| Precision | 0.7143 |
| Recall | 0.9163 |
| F1-Score | 0.8028 |
| ROC AUC | 0.7859 |
Model Availability
The model is publicly available and an example notebook of the models use is also available on the github: kernel-brain-data.
Data Availability
The training and test data is available on huggingface here