Blog: Why you can’t learn Machine Learning?
It’s evident that the emerging Artificial Intelligence application draws attention from numerous people who’d like to contribute by taking part. But where do you start?
If you’re reading this article, you should already be aware of the impact that AI brought to humanity. You should also be familiar with the fact that we’re not even exploiting it as much as we could. We should be prepared to embrace the exponential growth of advancement of AI technologies. There’s this question I’ve heard multiple times. To be honest, I’ve even asked it myself.
How can I learn Machine Learning?
You can’t. That’s something I’ve learned during my time in one of Microsoft’s Development Centers. Machine Learning is not something you learn and apply. It’s something that’s based on methods that give outstanding results when being used in specific areas. So we should start with a different question.
What is Machine Learning?
Machine Learning is a set of algorithms that are able to, when given a specific amount of data, analyze that data and learn form it. These algorithms have the tendency to automatically learn and improve from experience without being explicitly programmed. That makes these algorithms, based on the patterns that they’ve learned, capable of detecting similarities or anomalies with new input data. Their convenience is found on their ability to be considerably faster and typically more precise than humans are. This brings us to another question.
Where do we use Machine Learning algorithms?
I won’t be afraid to use the biggest cliché statement out there: Sky is the limit! These algorithms are all around you and that makes you their everyday, unaware victim.
If you listen to Geoffrey Hinton, a Turing Prize winner also known as the “Godfather of Deep Learning”, there would be no problem that neural networks can’t solve, you just need the right data to train it. Although it sounds like there are innumerable options, we can easily name the most popular use cases.
Machine Learning found it’s best application in financial services, healthcare, security, marketing, retail, transportation, robotics, communication, sports betting, gaming industry and so forth. In certain areas we use specific Machine Learning methods like image recognition, Natural Language Processing, number crunching, word embedding etc.
Now that you’ve absorbed this, you should contemplate about where do your interests fit the most. In which area can you contribute to process automation. Lets connect some of these methods to areas where they are used so we can get a clearer picture.
This is the area that suffered tremendous improvement so far. We are talking about saving human lives by helping with human decision-making. Image recognition algorithms have been most effective with detecting abnormalities in the fields of X-Ray imaging, Ultrasound, types of Neuroimaging like MRI, CT, EEG along with others. In these cases ML algorithms are fed with thousands of labeled images using supervised learning.
AI driven robots can detect Alzheimer’s disease in less than one minute with more than 80 percent accuracy based on speech patterns and voice. Robotic systems are also used for Robo-assisted surgery. AI-enabled robotics can enhance and guide the precision of minimally-invasive surgical procedures performed through tiny incisions.
My certain guess is that you’ve talked to Siri, Cortana or Alexa at least once. I’m even more certain that you’ve used voice recognition technologies. Proficient Deep Learning neural networks give you that ability. They are used for so-called text-to-speech (TTS) conversion and vice versa, and for language translation as well. If you’ve used Google Translate service, you have noticed that it’s translations are indistinguishable from human-like ones.
Next big thing in communication are chatbots. Natural Language Processing is making continuous progress in understanding context of human sentences. That gives us the ability to automate redundant questions which can be answered by a machine by interpreting conversation between computer and a human.
ML is not just a buzzword when it comes to finances. These algorithms generate, as well as save, a substantial amount of money. Institutions like banks or large investment houses store immense quantity of data that is produced by their users.
Since users generate millions of transactions directly or through third-party services, they are exposed to different kinds of threats. In this specific occurrence we can use ML algorithms for two different cases. First one is for Algorithmic Trading which uses complex AI systems to make extremely fast trading decisions that’s followed by next case, Fraud Detection, systems that can detect unique activities or behaviors and flag them for security teams.
Robo-advisers are algorithms that helps users calibrate a financial portfolio to the goals and risk tolerance that fits them the best without help of a physical adviser. Similar systems are used for identifying a risk score of a customer based on their parameters for banks before they even offer service to customer.
Millions of external and internal data points being out there, ready to transfer a vast amount of data, makes them viable for all forms of cyber attacks. It’s not achievable to process this extent of information with human capabilities. What solves this problem? Yes, you’ve guessed it — Machine Learning.
Algorithms here can detect any suspicious actions by processing immense amount of data almost real-time, therefore protecting users from any unknown malware, malicious files and policy violations even from encrypted traffic. They are not only capable of detecting already known types of attacks based on the patterns that they’ve learned, but also detect new ones never seen before.
Vehicles today already use complex systems like GPS navigation, image recognition for sign detection, parking assistants that are all powered by ML technologies. But what excites me the most are self-driving cars. Major car companies are hyped to become one of pioneers in this field. Progress made so far is rather significant.
This could be considered like one big Internet of Things (IoT) project since we’re confronted with correspondence of both mechanical technology and interaction with online resources. Cars would be able to detect objects real-time and act on them immediately protecting people from car crashes and law disobedience.
You can’t find a solution and then look for a problem. Odds are considerable that you won’t find the right one. Instead, try thinking about what real-world problems exist and how can you automate them or at least improve it’s already automated processes by using Machine Learning algorithms. We are lucky enough to be the pioneers in Machine Learning so don’t waste your chance to engineer the world.