Source: Deep Learning on Medium
AI, ML, Deep_Learning, Supervised/Unsupervised Learning, Computer_Vision, Self-Driving Car, iPhone X portrait mode, Pixel 3-night sight, Ok_GOOGLE, Hey_SIRI, Hey_CORTANA, Auto-correct and what not..all of these are sets and subsets of one another. A girl with many Faces and no name (GOT), what if all this started with Terminator 1 maybe not, actually it didn’t. In my lethargic search for truth and enlightenment, I found out some very brainy prof. belonging to a very reputed University gave a totally bizarre assignment to one of his students which was to develop a simple computer code for computer vision, which he did or did not is not the question here.
Almost everything around an avg. person has a silicon chip in it which was very very obedient in following orders that to very specific ORDERS, like setting oven timer to 3 min, and making a beep sound when you forgot to switch off headlights in your car. But apparently, some group of geniuses decided to put a brain/ command center alongside that silicon chip and now your oven is sending you text messages about what to eat, how to eat and when to eat. And your car can now pick you from your veranda drive by its own to a mall, drop you and park itself. I agree there have been development in Medical-Science and this brain/ silicon chip combination is saving a life no doubts there.
AI, ML, DL all of these are a father, son, brother (not in order). I am not getting technical as there are billions of definition out there for each one of them and they all are changing every minute. Bottom line is the above-mentioned group of geniuses was actually trying to figure out how a human brain works and why it looks so ugly. Fortunately, they didn’t get any answer but in there quest they isolated one of the neuron among 100 billion others(neuron the basic working unit of the brain) and tortured it at best and made mathematical equation (with weights, biases, activation function) to imitate the working of a single neuron. Single Mathematical Neuron on its own is nothing but just a traffic signal with only 2 options, red or green depending on the amount of traffic it receives. Then similar to human brain numerous neurons were stacked in layers interconnected and they were assigned to do a very specific task until they are perfect in doing so. Example: the working of Convolutional Neural network for image recognition has many layers of filters i.e one layer with many filters then another layer with another number of filters. and after training and testing the model outputs 85% accurately on the images it has never seen before. And it is still unclear how this filter works and how the network as a whole does what it does with ever so increasing accuracy. there is definite mathematics behind it but at a certain point where the data is so huge and gibberish it is impossible for the human brain to comprehend the process just like the human brain, we know it works but still don’t know why.
All the heavy duty terms like weights, biases, activation function, convolutional, feedforward, backpropagation, gradient descent will be giving their proper introduction when required. I am a complete noob with no past experience in Computer Science, coding, coding language. I have started this journey into the unknown as I am fascinated by the application of ML. I have no degree in the field and I want to learn, I have sorted out certain open source guides to gain some respectable background knowledge in ML, I have handwritten notes and Python codes which I will be uploading as soon as I have tried and tested them my self. I will also be providing links to all my references.
Now back to sKyNet, though I have not made any piece of code myself that may end world hunger or solved poverty. I am absolutely sure that day is not far from today, the growth in the field of ML is exponential and I mean why not Machine is Learning by itself and therefore no human interference or silly mistakes are made during the task. Warning: Automatic Machines are not ML or AI, automatic machines are just a very efficient piece of code that has been programmed on a very broad spectrum and will definitely fail if faced a situation it was never programmed for. Machine Learning, on the other hand, is like human kid 2–3 yrs old, we teach it several times what is alphabet looks like what digits look like and then ask it to recognize it, there is always room for error and that’s to be expected as we never hard-coded our program, we just tried to teach it over and over again. The downside being our machine kid is only good for very specific things. All will be very clear when we work on data and run an actual program. Google, Amazon, apple, facebook these companies are using ’n’ number of machine kids which are good at ’n’ number of tasks, now these machine kids need to learn on something right or else how would they be better at there job this something is DATA, user data the more the data the more efficient learning can be done by these machine kids and the better they perform when tested. Hence the recent scam where Facebook was hit by private companies for DATA theft, Google is also facing many allegations regarding same.
So how these machine kids are helping you in your real life?? Lets take an example of GooGle Maps: lets say in a very general sense you have to drive from Michigan to Miami, we know source and destination and we input these in maps and our app draws a blue line over the best route possible, there is also a grey line somewhere in between showing a different approach with 10–20 mins delay. how does it show us the best possible route, why not show us all the possible roads leading to Miami from Michigan and let the user decide what’s best for them. There is a caveat in this approach as the user has very limited data that is a number of roads he can take and how many miles will he travel in total, that’s all that he knows on his end. But our machine kids working in GooGle Maps have previous data from previous journeys many people had make, machine kid has even divided the whole route into different segments so there is now more data as there are journeys made for only a patch of road that we might need as a whole. M kid has their travel velocities , there pit stop time and some very polite person also report highway construction to the M kid. This way M kid knows a lot more than the end user , and hence predict a best way possible. If the user is in agreement with the M kid, the user will start the journey and reach Miami, M kid will now match his recommend time before the journey started and comapre it to the real time user took to travel. If the time matches M kid is rewarded and if the real time taken is much longer than anticipated the M kid will recount all the pit stops an droad conddtions and then may be suggest an altrenative road to the next user opting for the same trip.
This is my 1st written article ever and my next article will have a guide for NOOBS by the NOOB, to start there hobby in ML from scratch installing software setting up libraries etc etc…