Every single encounter from waving to someone on the street to a one-on-one interview calls upon a person’s emotional sense; however this sense is significantly dulled by certain disorders, such as Autism or Asperger’s syndrome. This year, even normal people have trouble identifying emotions caused by lack of physical interactions during the pandemic. Emotional misinterpretation like this can often lead to spoiled relationships, opportunities, and lives. Additionally, current treatments to these disorders do not guarantee concrete results, which we will talk in more detail later.
As of 2015, 15%, or 31 million people are affected by some sort of emotional disorders in the US - ALONE. For autism only, from 1975 to 2009, the probability of autism increased from 1 in 5000 to 1 in 110. This is a massive growth, and doesn’t seem to be slowing down, which means that this is a serious problem that needs to be adequately addressed.
Introducing our solution: decrypt, an extremely versatile program that can be used in a multitude of places for daily use, helping those affected notice and interpret emotions. We used a combination of open cv, python, keras, and tensorflow libraries to build this project. Every neural network model needs to have weights and a configuration. In simple terms each of those connections between layers of the model requires weights to determine what gets passed on and to what extent. These weights paired with other layers, such as pooling layers and optimizers, are what determines the accuracy or overall effectiveness of a model.
Optimized weights and other layers will determine the overall accuracy and effectiveness of the model. In order to have full control over our model in terms of tuning, we designed our own configurations and trained our own weights.
The model is made up of 4 key components. We chose to use a CNN model because of its spatial coherence, resistance to image noise or distortions, as well as other beneficial features. We added three max pooling layers in different parts of the model to downscale the data points to prevent the model from overfitting to the training data, which leads to lower accuracy with the testing data. The adam optimizer helps us to efficiently and effectively optimize the model weights, and the softmax activation function allows us to provide probabilistic outputs for each emotion: happy, sad, fear, surprised, neutral.
As for now, current solutions can not guarantee concrete results. These solutions may take years to show actual results, or even turn out to not work at all. Medications also have a multitude of side effects, such as excessive sleep, irritation, and more. Dietary approaches are another solution, but alterations to one’s diet can cause a downfall of overall functioning.
Decrypt can simplify daily interactions by allowing those with mental disorders to respond correctly and effectively to the other’s emotions. This also improves social skills and social confidence as well, and therefore opening up more pathways and opportunities to them later in life. By fostering stronger and closer relationships and improved confidence, their overall productivity is also boosted along with their creativity. As a society, with an overall more capable and efficient workforce, Decrypt can further cultivate innovation.
In summary, millions in the world population experience emotional recognition problems. Decrypt can significantly lessen the burden of broken relationships and connections for people with Autism or Asperger’s syndrome. By allowing them to accurately gauge emotions while helping them interpret each emotion, Decrypt can simplify a multitude of person to person interactions every single day.