index.qmd

---
title: "Automatic Classification of Social Behaviors among Mice Using Deep Learning"
author: "Elías Aburto Camacllanqui"
date: last-modified
format: 
  revealjs:
    theme: white
    chalkboard: true
    logo: Images/eliaslogo.png
    transition: slide
    lang: Es-es
    auto-stretch: true
---

# Introduction {background-image="Images/Resident_Intruder_Paradigm.png"}

---

## Resident-Intruder Paradigm


:::: {layout="[ 60, 40 ]"}

::: {#first-column}

<div style="font-size: 30px;">  <!-- Adjust the font size as needed -->

- Social behaviors which provoke territorial agression (Thurmond, 1975).

- Experiment:
  - A resident mouse is put in a cage for 24 hours.
  - Then, an intruder mouse enters the cage.
  - A camera records the secuence of behaviors that provoce agression.
  - At the end, researchers manually label the observed behaviors in the video to identify patterns of behavior.
</div>

:::

::: {#second-column}

![](Images/Resident_Intruder_Paradigm.png)
:::

::::


---

## Problems

:::: {layout="[ 60, 40 ]"}

::: {#first-column}

<div style="font-size: 35px;">  <!-- Adjust the font size as needed -->

- **Replicability Crisis**: The observations can vary between researchers, which affect the replicability of studies.

- **Fatigue and Error**: The process demands time and effort, which can provoke errors.


</div>

:::

::: {#second-column}

![](Images/Replicability.png)
:::

::::


---

# Justification {background-image="Images/Resident_Intruder_Paradigm.png"}

---

## Theoretical Justification

:::: {layout="[ 50, 50 ]"}

::: {#first-column}


::: {#text-block style="background-color: #f8d7da; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 100px;"}

<div style="font-size: 35px; color: #a71c24; font-weight: bold; background: linear-gradient(to right, black, red); -webkit-background-clip: text; -webkit-text-fill-color: transparent;">  <!-- Adjust the font size and color as needed -->

**Problem**: The observations can vary between researchers, which affect the replicability of studies.

</div>

:::

:::

::: {#second-column }

::: {#text-block style="background-color: #8B0000; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 70px;"}

<div style="font-size: 35px; color: white;">  <!-- Adjust the font size as needed -->

The classification model drives more **objective**, **precise**, and **scalable** measurements compare to manual labeling. Therefore, it increase the quality of social research.

</div>
:::

:::

::::

---

## Practical Justification



:::: {layout="[ 50, 50 ]"}

::: {#first-column}


::: {#text-block style="background-color: #f8d7da; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 100px;"}

<div style="font-size: 35px; color: #a71c24; font-weight: bold; background: linear-gradient(to right, black, red); -webkit-background-clip: text; -webkit-text-fill-color: transparent;">  <!-- Adjust the font size and color as needed -->

**Problem**: The process demands time and effort, which can provoke errors.

</div>

:::

:::

::: {#second-column }

::: {#text-block style="background-color: #8B0000; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 70px;"}

<div style="font-size: 35px; color: white;">  <!-- Adjust the font size as needed -->

The automatic classification model streamlines the research process in research centers making researchers more productive.

</div>
:::

:::

::::


---

# Objective {background-image="Images/Resident_Intruder_Paradigm.png"}

---

## Main Objective


<div style="font-size: 60px; margin-top: 80px;">  <!-- Adjust the font size as needed -->

Develop an automatic classification model for the monitoring and analysis of animal behavior in research environments.

</div>


---

# Methodology {background-image="Images/Resident_Intruder_Paradigm.png"}

---

## Data

::: {.panel-tabset}

### Tool

- **MARS**: A tool for capturing the positions of mice using deep learning (Segalin et al, 2021).

::: {layout="[40,60]"}

![](Images/MARS.png)

![](Images/mars_demo.gif)

:::


### Challenge

::: {layout="[50,50]"}

- **Caltech Data** published the <a href="https://data.caltech.edu/records/s0vdx-0k302" target="_blank">dataset</a> which contains positions and labels.

![](Images/Caltech.png){width=800px height=300px}
:::

### Task


![](Images/Task1.png)

:::


---

## Exploratory Data Analysis

::: {.panel-tabset}

### Sequence

- Each sequence corresponds to a particular experiment involving a resident mouse (0) and an intruder mouse (1).

![](Images/sequence.png)


### Visualitation

::: {layout="[40,60]"}
- Each sequence has a different number of frames.

![](Images/pose_sequence.gif)

::: 


### F / sequence

![](Images/Frame1.png)

### F / Dataset

Percentaje of frames per class in the whole dataset.

![](Images/Frame2.png)



:::


---

## Preprocessing

::: {.panel-tabset}

### Loading data

![](Images/LoadData.png)


### Padding data

![](Images/PaddingData.png)


::: 


---

## Training

::: {.panel-tabset}


### Loss function

<div style="font-size: 25px;">  <!-- Adjust the font size as needed -->

Sparse Categorical Cross Entropy is appropriate for multi-class classification problems.

</div>

![](Images/LossFunction.png)




### Model Structure

![](Images/ModelStructure.png)



### Training

![](Images/Training.png)

::: 


---

## Evaluation

::: {.panel-tabset}

### Model Accuracy

![](Images/Model_Accuracy.png)

### Model Loss

![](Images/Model_Loss.png)

### Test Accuracy

![](Images/Test_Accuracy.png)

:::

---

# Expected Impact {background-image="Images/Resident_Intruder_Paradigm.png"}


---

## Expected Impact


:::: {layout="[ 50, 50 ]"}

::: {#first-column}

::: {#text-block style="background-color: #8B0000; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 70px;"}

<div style="font-size: 25px; color: white;">  <!-- Adjust the font size as needed -->

**Advances in Scientific Research**: 

- Significantly reduces time and errors in the process.
- Improves the precision and consistency of the results.
- Optimizes the replicability of studies by minimizing variance between observers.

</div>

:::

:::

::: {#second-column }

::: {#text-block style="background-color: #8B0000; padding: 20px; border-radius: 10px; border: 1px solid #f5c6cb; margin-top: 70px;"}

<div style="font-size: 25px; color: white;">  <!-- Adjust the font size as needed -->

**Practical Application**: 

- Can be expanded to identify social human behaviors, such as harassment and bullying.
- Could help address security issues in schools and businesses by identifying dangerous behaviors automatically.
- Could aid in the care of animals during their development.

</div>
:::

:::

::::



---


# References{background-image="Images/Resident_Intruder_Paradigm.png"}


---


<div style="font-size: 30px; text-indent: -1.27cm; padding-left: 1.27cm;">

Segalin, C., Williams, J., Karigo, T., Hui, M., Zelikowsky, M., Sun, J. J., Perona, P., Anderson, D. J., & Kennedy, A. (2021). The Mouse Action Recognition System (MARS) software pipeline for automated analysis of social behaviors in mice. *eLife, 10*, e63720. https://doi.org/10.7554/eLife.63720 

<br>

Sun, J. J., Karigo, T., Chakraborty, D., Mohanty, S. P., Wild, B., Sun, Q., Chen, C., Anderson, D. J., Perona, P., Yue, Y., & Kennedy, A. (2021). The Multi-Agent Behavior Dataset: Mouse Dyadic Social Interactions. *Adv Neural Inf Process Syst*(Db1), 1-15.

<br>

Thurmond, J. B. (1975). Technique for producing and measuring territorial aggression using laboratory mice. *Physiology & Behavior, 14*(6), 879-881. https://doi.org/https://doi.org/10.1016/0031-9384(75)90086-4

</div>

---

# Thanks

---