Skip to main content

What are Emotional Neural Networks? - I

We know that Neural Networks were built trying to copy the structure of the human brain. 

The brain is composed of neurons, all interconnected and shooting/receiving impulses to/from multiple edge connections. That's how a Neural Network is too. Multiple nodes, all interconnected with each other and propagating information and errors to and from, trying to correct itself in every try.

But have you ever wondered if we can ever emulate human emotions? Or if we can replicate human emotions in a Neural Network?

Well, part of this has been implemented already, and although it sounds somewhere along the lines of Affective Computing, it's not completely the same. 

What is an Emotional Neural Network?

Emotional Neural Networks are a new form of Neural Networks that seem to emulate a certain part of human emotions while learning. There are a couple more models and definitions for such a Network, but we'll focus on this aspect first.

Picture a scenario where you're learning something completely new. For instance, learning an instrument. How do you feel at the beginning? 

Sure, you enjoy it. But it's a really new task and you don't know anything about it! You're anxious, maybe not dramatically anxious as shown in movies but there is a doubt, a certain level of uncertainty about the future that will you be able to learn this? Will you be able to play as you had imagined?

As time progresses, your skill grows and with that, your confidence. The anxiety reduces. You're happy with the small achievements and still looking forward to accomplish more. Maybe you still are anxious if you're ambitious about your goals, but the newly acquired confidence gives you an edge to practice better, and enjoy it.

This principal has been employed in a new form of Neural Network. The algorithm has been called Emotional Backpropagation Algorithm. Why?

You see, the Neural Network is given 2 parameters using which it will now make decisions, anxiety and confidence. As the Neural Network iterates over each iteration and computes the gradient, the change is influenced by the anxiety and confidence coefficients for every neuron. 

To put it simply, your neural network is not just punishing itself by the margin of its wrong answers, it is also keeping a record of how many times it was wrong and influencing its learning capabilities by it.

Here, YI1, YI2, ...YIi are simply the input neurons. Consider them as input features only. 

So for each neuron in the second layer, you have the product of all input features with their respective weights, and also the product of the Emotional Neuron, with its own weight. The latter part is what is new here.

It is the backpropagation algorithm that makes a difference, and finally takes into account the anxiety and confidence parameters that we discussed before. Normally, the gradient matrix that is computed depends only on the delta terms (change in target and real output) for that particular neuron, which in turn is influenced by other delta terms. But in the EmBP algorithm, we compute the gradient matrix using the product of the confidence coefficient and the previous change in gradient, the product of the anxiety conefficient and delta terms and the normal conventional way of computing the gradient regularized using certain parameters.

This is as if the amount of change is being influenced with the anxiety and confidence coefficients.

This was the Emotional Backpropagation Algorithm, a model of Neural Networks that has been tested and found to be more efficient than conventional neural networks in many cases such as Face Recognition, Credit Risk Evaluations, Rainfall-Runoff Modelling predictors, etc. If you want to read more about this algorithm and research, please read Dr. Adnan Khashman's journal paper on EmBP.

But, this is not all about Emotional Neural Networks.
Since the original question was, "Can we ever emulate human emotions?" there have been a lot of approaches to this possibility. Some models, such as this algorithm above, have simply emulated the 'behaviour' of emotions in a certain aspect of human life. But there are other cases, where people have actually dived deeper and tried to actually implement the structure of the brain that is responsible for producing and sensing emotions. 
If we can achieve that, we might just be able to produce any or all kinds of human emotions, right? How, and what is it? 
The answer to these questions is not so simple, and I will be presenting those in my next part of this series.

Stay safe and stay tuned.

Comments

Post a Comment

Popular posts from this blog

Namaste JavaScript Quick Notes

Note:  Akshay Saini's Namaste JavaScript is probably the best course for JavaScript developers out there. These are my personal notes that I made while watching the course; they serve more of as an online quick reference for my understanding and revision, and I hope it benefits anyone reading it too! Everything in JS happens inside an Execution Context. Before a JS code is run, memory is allocated and variables are set as undefined   , and functions are set as their exact code in the scope within the Execution Context. The global execution context hosts all the global variables and function definitions. An Execution Context has 2 components: Memory, that stores variables and functions; and Code, that reads and executes the code. Call Stack maintains the order of execution contexts. Since JS is single threaded and asynchronous, at one point of time, only one function is executed which is at the top of the call stack. For each function, an execution context is created before executi

An introduction to APIs

API is an acronym for Application Programming Interface. Let's start with first defining some basic terms: Browser: These are browsers. To visit any website on the internet, you need a browser. Server: Hmm, this is tough. In simple words, server is a computer. Yes, just like the laptop, or PC at your home. The only difference is that it does not have a screen. Of course, there are other differences in technical specifications, but at its core, the server is just, simply, a computer. That's it. So why is it called a server? Because it serves . When you go to a website like google.com , your computer connects to the internet and gets you your search result. But your computer's internet connection has to get that result from somewhere, right? If the google search result is giving you some answers, the answers have to come from somewhere. What is that place? The answer to that some place is: a server. When you click on the search button on google, or hit enter after typing, &q

Review: Nestjs - Finally a scalable way to build APIs

I have been thinking about this for a long time. There HAS to be a defined way to build APIs in a scalable way.  If you have used Node, Express, etc in your side projects, you might have felt that after a point in development, debugging truly becomes a pain. Sure, enterprise-level API codes are great. But a lot of times, these configurations are too much, and probably not even needed in other projects. To be honest, I haven't seen a lot of Open-Source API codes either to make a judgement on how experienced developers build their APIs. Anyway, I came across an amazing framework recently, and I think if you are coding a complex API, this should be your way to go. Nest.js Nest.js is a framework for building efficient, reliable and scalable server-side applications.  You essentially break your APIs into controllers, services, and modules, which allow you to modularize the smallest of functionalities in your endpoints or the API as a whole. Why is modularizing important? As I have talk