Using deep learning to study emotional behavior in rodent models.

Area of Science:

• Neuroscience
• Animal Behavior
• Machine Learning

Background:

• Quantifying animal emotions like anxiety and stress is crucial in neuroscience.
• Manual behavioral scoring is subjective and time-consuming.
• Classical methods use simple metrics, limiting behavioral insight.

Purpose of the Study:

• To review deep learning applications in analyzing animal behavior.
• To explore how different deep learning models and training impact behavioral state representation.
• To bridge the gap between neural activity and complex behaviors.

Main Methods:

• Utilizing pose estimation from videos.
• Applying supervised, unsupervised, and self-supervised deep learning approaches.
• Correlating derived behavioral representations with neural activity recordings.

Main Results:

• Deep learning enables extraction of nuanced behavioral states from pose data.
• Various model architectures and training paradigms yield different behavioral representations.
• These representations enhance the correlation between brain activity and behavior.

Conclusions:

• Deep learning offers powerful tools to objectively quantify complex animal emotions.
• Advanced computational methods are revolutionizing behavioral neuroscience.
• This approach deepens our understanding of brain-behavior relationships.