Datacollection.gif

emotIOn 1.0

20181029_125735.jpg

emotIOn

2018

Creating machines with empathy…

"Emotional programs" promise to promote Human - Computer Interactions (HCIs) to a new level in which humans and technology can start to function as a well orchestrated and natural symbiosis. emotIOn is an exploratory project that seeks to understand human emotions using machine learning to predict a user’s emotional state depending on the environmental state in which the user is in. A framework describing different emotional sources or weights was developed. These emotional weights represent the main variables that have a big impact on a human’s emotions. The models explored by emotIOn are models that are based only on the “Environmental State Weights”.

emotIOn Neural Network Architecture.jpeg

Collecting Data

A training data set was built by mounting an MIT terMITe (wireless environmental sensor that captures data to help us better understand our environments and human behavior) to the user and capturing values for X Acceleration, Y Acceleration, Z Acceleration, Temperature, Light and Humidity, those values were concatenated to the values obtained from the analog emotion interface which are adjusted by the user to reflect his/her emotional state.

Anger

Anger

Joy

Joy

Sadness

Sadness

Disgust

Disgust

Environmental State Monitoring (MIT terMITe)

Environmental State Monitoring (MIT terMITe)

Emotional State Adjustment

Emotional State Adjustment

Collecting Environmental & Emotional Data

Collecting Environmental & Emotional Data

Building & Training NN Model

Model was created to be used with individual users.

Model was created to be used with individual users.

Each individual user model is trained with more than half a million data points.

Each individual user model is trained with more than half a million data points.

Analog emotion interface & terMITe mount design and construction

Wearable Case Design

Wearable Case Design

3D Printing Y Modulator & Wearable

3D Printing Y Modulator & Wearable

Soldering and construction of Y Modulator

Soldering and construction of Y Modulator

Y Modulator Final Design

Y Modulator Final Design

Model Results

The trained model is capable of predicting completely new data points with an accuracy of up to 87%.

ModelAccuracy.png
Model Loss.png
PredicitonSpace.png
Humidity vs Z Acceleration vs Emotional State

Humidity vs Z Acceleration vs Emotional State

Light vs Temperature vs Emotional State

Light vs Temperature vs Emotional State

Light vs X Acceleration vs Emotional State

Light vs X Acceleration vs Emotional State


Credits and Acknowledgements

Carson Smuts (terMITe development) | Jason Nawyn (Project’s Vision) | Leonardo Garrido (Machine Intelligence Advisor)