Artificial Neural Networks – The Nervous System of the Next-Generation
Ever since mankind started roaming on earth, evolution has taken place at an accelerated pace. With the countless dangers and obstacles along this journey, ‘survival of the fittest’ has been a concept that chaperoned this march towards the present urbanization. So what exactly kept us safe through all these years? It wouldn’t be wrong to say food and water. But behind all that craziness, lies the brains of the entire concept, I mean literally: the BRAIN.
Who taught you to retreat your palm when it touches something hot? The brain. Who stops you from repeating the same mistake the next time? Your brain. This paradox displays the ability of the brain to adapt to new inputs. In other words, sensory data. Hold on… I don’t intend on rerouting this into an anatomy lesson but it’s worth a shot understanding how you train your brain to respond to future events as that is exactly what an Artificial Neural Network (ANN), or so often referred to as merely the Neural Network, is all about.
The Anatomy of a Biological Neuron
Figure 1: Typical Anatomy of a Biological Neuron
The neuron can be elucidated as the basic building block of the nervous system. This indicates that there are several millions of neurons that are activated when you take a simple step. Since our discussion disregards the in-depth details of how neurons work, I’ll handpick only the Dendrites, cell body, axon and the synaptic terminal (synapses).
Receiving an input is the first event that basically starts up any action. The dendrites are fairly and countlessly branched to receive these incoming triggers. And there are several dendrites spread across the body as these input signals might be more than one at a time. For instance, take the same fire-hazard scenario mentioned in the beginning of this article. First, you feel the heat which can be received at one end and then you see that your hand is on something hot which fires the visual input. So the dendrite is the first-hand receiving agent.
There is little to no use if an event is simply received by the dendrites. If no follow-up action is taken, to say – retrieve your hand, I’m pretty sure you’re not going to like the outcome. Hence, we need to process and control the following actions. This is where the cell body comes into context. The cell body is composed of much more complex constructs which we’ll not unravel right now. Then comes in the axon which is a channel through which the processed information is going to travel through. Considering the biological processing and communication between neurons, inter-cellular junctions called ‘synapses’ bridge the gap. These synaptic terminals or the synapses act like connectors to link one neuron with another as the nervous system is a network of millions of neurons that work together to get a message through to the brain. A neural network requires such a nervous network (not literally!) in order to train systems to respond to digital events that are received through computing terminals.
Enough beating around the bushes. Now let’s get into how this actually formulates an ANN.
Mapping the layout of the Artificial Neuron
Figure 2: An Artificial Neuron
It is quite self-explanatory mapping the biological neuron to the artificial neuron. Inputs are received as x1, x2, x3, and so on. w1, w2, w3, etc. are weights that determine the impact that particular input will have on a suitable response that needs to be delivered for this event. (We’ll look into the weights and their related computations in detail in the ensuing posts). The central unit processes the input and fires it to the next artificial neuron to learn or fire the respective response. Artificial Neural Networks require to make modifications to their “synaptic connectors” as they undergo a process of learning. Even diagrammatically it is convenient to map the various components across the biological and artificial neurons. But in reality, both of these are a little more complicated than depicted while functionally replicating each other.
Realizing the Applications of Neural Networks in the Daily Commute
So, what exactly is the big buzz about ANN?
The potential that neural networks carry was not fully realized before the 1990s. Again, this relates to how human processing takes place, but only a matter of processing it much more efficiently and with minimal human interaction.
Neural networks are avidly used in the following contexts.
To render informative conclusions from complex, unrefined data.
Identify patterns or trends in commonly occurring issues or events.
Orchestrating adaptive learning from classrooms to research center levels.
Predicting outcomes of events with omni-various parameters which may otherwise be tedious and nearly impossible by humans.
Process in real-time by simultaneously learning and training to address similar future occurrences.
Beat the hassle of having to ‘code’ each and every scenario in order to address plausible events as neural networks train themselves.
At this juncture, we might question ourselves if conventional computing cannot suffice the above. They can, but to a very abstract level without rigorous algorithms written to address every viable scenario expected to be encountered. Now this is hard as a human cannot possibly come up with a solution to every expected problem. By now, we would have solved numerous major world crises situations if that was the case. Hence, the approach intended by neural networks is to self-teach itself by encountering similar activities and adapt to new future circumstances...so, like what we do on a daily basis.
Let’s get into a few real life applications that have implemented such neural networks to assist (I strongly believe in rephrasing this as “replace”) human workers in provisioning services, whether be end-user services or academic-innovative contributions.
A. Character Recognition
When you see an object, you’re able to recognize and label it. Likewise training neural networks to recognize characters has been one of the many use-cases addressed. Consider the old-time all-manual labour that went into reading and classifying handwritten texts/manuscripts. With the advent of neural networks, these can be read by the device and the ANN creates classifying rules upon which similar manuscripts are grouped and labeled together . Character recognition has been an area constantly defined using ANN and the precision in recognition has augmented to a level where it is now applied in banking systems to read the characters on cheques and sensitive drafts.
B. Navigation Systems
Another application of neural networks can be mentioned as navigation systems. With urbanization, travel and traffic has become a major concern. This is where navigation systems have managed to inject new avenues in suggesting shortest distances, calculating travel time via various plausible routes, warning about traffic lights and pedestrians...it’s like having your mom as the backseat driver, only less annoying. There is no possible way to write an algorithm for each and every route or predict traffic hours. Hence, GPS and DR (Dead Reckoning) navigation systems have resolved to neural networks to self-train and analyze these calculations .
C. Projecting Stock Market Conditions
Buyers and sellers of stocks are always on the edge, observing the steady up-curve or downfall of stock prices and the stock index. It’s a market that has gained attention and since millions of shares are bought and sold every second, neural networks have been brought up as a solution to keep track of these changes that take place in almost real-time .
D. Intelligent Chatbots
As we march into a world where actual human conversations seldom take place, virtual conversationalists, appropriately called “Chatbots”, have taken over. Simply put, chatbots are messaging programs that understand natural human conversational inputs and reply ingeniously with suitable responses. The underlying core of research within chatbots had been the fact that users come from a vast pool of varying demographics. And these user inputs can be interpreted in multiple ways with multiple corresponding responses .
So how ANN intends on servicing chatbots is to understand natural language, identify the intent of the conversation and provide precise user-expected results via smart replies. This has called for chatbots to go through multiple training procedures that mold the chatbot to assume the role of a person specialized in the context of the conversation and carry out unscathing dialogues with users. Such chatbot applications have been implemented in Customer support, Air-ticket Reservations, Online shopping, Delivery tracking, etc.
E. Gaming Applications
The gaming industry has also seen light with neural networks. It might seem like a waste of such tech resources to simply go through a game of Dungeons & Dragons, but I did a little bit of research and the market for this is alarmingly massive (obviously, am not a gaming fanatic). Well this maybe because that games have now taken a turn towards transcending players into a parallel universe with Virtual Reality . Mapping the existent surrounding points to the gaming environment, recalculating paths based on players’ choices, tracking multiple players across varying platforms simultaneously...wow, I just realized that gaming applications are no joke anymore.
Wrapping up this edition, we’ve discussed the nitigrities of what is a neural network, how its functions relate to that of a neuron and the importance it dictates towards multitudinous real-life applications.
Stay in the loop for future posts where we’ll be discussing the technical details of how neural networks perform character recognition, firing rules for incoming signals, underlying concepts such as LSTM(Long-Short Term Memory) and more.
 Breuel, T. M. (n.d.). Handwritten character recognition using neural networks. Handbook of Neural Computation. doi:10.1887/0750303123/b365c93
 Zhang, Y., & Chong, K. T. (2014). An GPS/DR navigation system using neural network for mobile robot. International Journal of Precision Engineering and Manufacturing, 15(12), 2513-2519. doi:10.1007/s12541-014-0622-4
 Wang, L., & Wang, Q. (2011). Stock Market Prediction Using Artificial Neural Networks Based on HLP. 2011 Third International Conference on Intelligent Human-Machine Systems and Cybernetics. doi:10.1109/ihmsc.2011.34
 Neuralconvo – Chat with a Deep learning brain. (n.d.). Retrieved March 21, 2018, from http://neuralconvo.huggingface.co/
 Larsson, J., & Mänttäri, J. (2011). Applications of Artificial Neural Networks in Games; An Overview.
A Neuron's Anatomy [Digital image]. (n.d.). Retrieved March 21, 2018, from https://www.memorangapp.com/flashcards/104120/The Nervous System I: Neuron Structure/
Character Recognition on iPad [Digital image]. (n.d.). Retrieved March 22, 2018, from https://www.pastemagazine.com/articles/2015/04/jot-script-2-review-the-great-ipad-stylus-you-neve.html
GPS on Car Navigation Systems [Digital image]. (n.d.). Retrieved March 22, 2018, from https://www.viva.co.id/otomotif/mobil/864339-gps-mobil-yang-bisa-sadap-telepon-mulai-diburu
Stock Market Predictions using AI Gold Mine [Digital image]. (n.d.). Retrieved March 22, 2018, from https://aboveintelligent.com/the-a-i-gold-mine-predicting-stock-market-success-19082ec87ef5
Hair Appointment via Chatbot [Digital image]. (n.d.). Retrieved March 22, 2018, from https://www.wordstream.com/blog/ws/2017/10/04/chatbots
Role Playing Games [Digital image]. (n.d.). Retrieved March 22, 2018, from http://ubuntunews.ru/news/legends-of-aethereus-new-3d-action-rpg.html