What is Machine Learning

What is Artificial Intelligence

There are many definitions and explanations of the term “Artificial Intelligence.” (AI) Though many authors have tried to explain it from a different perspective, in the end, they all seem to be saying the same thing.

Artificial Intelligence has to do with that branch of Computer Science which concentrates specifically on creating machines that can react, work and act like human beings. When a computer is designed with AI, it can do some of the things which humans do like learning, problem-solving, speech recognition and planning. These machines/computers can surprise you with their human-like abilities. However, their abilities come from experience and new inputs (Techopedia Inc., 2018).

Maybe you have heard of computers or robots that can play chess or even cars that are self-driven. When a computer is trained on how to process the natural language of humans with the “Natural Language Processing” technology, it becomes easy for them to analyse and recognize patterns in every data available  (SAS Institute Inc, 2018). With these abilities, computers can do some of the tasks which humans do, especially the ones they are programmed to do.

Computers did not start mimicking human characteristics from the beginning. Even though 1940 digital computers proved that with the right programming, they could perform some tasks, the idea of intelligent computers all began in 1950. The first area of exploration in AI research was symbolic methods and problem-solving. However, when the US Dept. of Defense decided that the idea was worth expanding, they moved towards training computers that could reason like humans.

From there, the Defense Advanced Research Projects Agency moved on to create personal assistants that were very intelligent in 2003 (SAS Institute Inc., 2018). These early creations also made way for formal and automation reasoning in today’s computers. One thing you must bear in mind is that, Artificial Intelligence or Machine Intelligence has been evolving till date and right now, many industries are benefiting from it.

Some of the industries using Artificial Intelligence in their operations are:

  1. Healthcare; using AI for X-ray reading and personalized medicine
  2. Retail; these sectors use AI to make personalized recommendations and also discuss shopping options to improve customer’s experience.
  3. Manufacturing; with Artificial Intelligence, companies know their expected load and demand.
  4. Sports; Artificial Intelligence captures images of every gameplay so that they can generate reports that will help coaches to do better in organizing games and optimizing their field strategies and positions (SAS Institute Inc., 2018).

So why is this branch of Computer Science making waves around the globe?

  1. Artificial Intelligence makes repetitive learning, as well as discovery faster through the use of data. AI is not the same as robotic or hardware-driven automation. Instead, it handles those high-volume tasks that entail repetitive steps reliably, without getting tired.
  2. It adds Intelligence to some of the products we already have today. Some of the products we already have and use are now improved with artificial intelligence capabilities to be more performance-oriented than before.
  3. AI helps in analyzing a large volume of data with the use of neural networks with numerous layers. A few years ago, it was not easy to build a system with five layers to detect fraud. However, with adequate data and computer power, it is now easy to achieve even greater feats than these types of systems.
  4. Artificial Intelligence makes use of data in more beneficial ways than before. In cases where you allow training algorithms to learn without labelled data, it turns ordinary data to intellectual property. You may not have noticed the level of answers hidden in your data. However, once you apply AI to it, wait for the best results.
  5. AI makes more accurate predictions since they use neural networks which did not exist before. For instance, you can now use Google Photos, Alexa, and Google Search; all these systems are available as a result of deep learning. The best part is that as you continue to use AI algorithms, the accuracy levels keep rising.

One may wonder if Artificial Intelligence is all good with no downside. The truth is that this field has its unique challenges which should be known. It is true that it makes industrial processes faster and automated, but the truth remains that it has some limitations.

One of the setbacks of Artificial Intelligence is that everything it knows must come from data. The danger in this is that if the data is inaccurate, you will get bad results which may, in turn, lead to wrong decisions.

Secondly, systems with Artificial Intelligence only learn one defined task or the other. It is not easy to create a system that can handle multiple tasks. For instance, any system that can detect fraud will not give legal advice, nor will it drive a car. Also, when a system is taught to play chess, you don’t expect it to play poker as well.

Therefore, no matter how accurate and fast these systems are, they cannot handle multiple tasks like human beings. What they learn is one specific task, and that’s what they can handle for you (SAS Institute Inc., 2018).

What is Machine Learning

Many people use “Artificial Intelligence” and “Machine Learning” interchangeably. However, these two terms are not the same. When you talk about AI, it encompasses a wider scope of programming computers to mimic humans. It focuses on giving computer and systems the ability to perform those tasks which were reserved for human beings in the past. When you see some computers that can solve problems easily and even help you to do your jobs faster, the reason behind them is Machine Learning. Meanwhile, ML is not wide in scope as AI because it is a part of the former which focuses on training machines to learn.

It is a method which computer programmers use to achieve Artificial Intelligence faster and easier. With the Machine Learning technology, a computer can easily make predictions that are accurate when you feed data into it. Secondly, computers need not be programmed, but   will be thought on how to access data and use it. As such, learning and further improvements will be based on experience and not programming.

Some examples of predictions which machine learning enables computers to make may be answering a question about whether a fruit is an apple or an orange (Heath, 2018). It could also be to spot the number of people crossing a major road or to answer if an email can be classified as spam or to recognize speech for a YouTube caption.

Initially, olden day computer software didn’t have codes that could teach a system how it could differentiate between an apple and an orange. However, with the field of Machine Learning, a model today has been given adequate training on how it can identify the differences in fruits using a large quantity of data (Heath, 2018).

One of the factors that have contributed to the popularity of ML as a branch of AI is the growing volume of data and the varieties of data available for machines to process. Therefore, it is safe to say that the success of Machine Learning relies heavily on lots of data.

Big Data & Data mining

When the available set of data becomes larger than the conventional data processing software to handle, they are referred to as big data. In simpler terms, big data describes data in large volumes, be it structured or unstructured.

According to an industry analyst, Doug Laney, 2000, big data can be defined using three important terms namely, volume, velocity and variety.

Volume: This explains that many data sources are available for organizations to utilize in their data gathering. These sources can be through the social media, from their business transactions, from machine-machine data or sensor. As such, big data doesn’t make use of sampling in the collection of data.

Velocity: The speed with which data streams into organizations is usually unprecedented. As such, they should be handled without delay since they flow in real-time.

Variety: Data flows into the organization in different types of formats such as numeric, structured, video, unstructured text docs, financial transactions, stock sticker, audio or email (SAS Institute Inc., 2018).

Data Mining

Someone may ask, “What happens when organizations collect big data? Do they use them like that or what?” Now the answer to these questions comes through explaining the term “data mining”. Organizations don’t start using the large volume of data that comes in regularly without extracting the pertinent or relevant information to their operations from the large pool of data. When all the big data comes in, data mining becomes the next step that will sieve the data and give decision-makers access to specific and small data from the pool. Without data mining, leaders may not have the information they need to plan for profitable courses of action for their businesses (Techopedia Inc., 2018).

Sometimes, data mining require some analytic tools, it may also be labor intensive, or the organization may decide to automate the whole process. Also, data mining can also involve the use of processes and techniques within business intelligence to ensure that patterns and relationships existing between data and systems become evident and accessible to key players.

Probability and Statistics

Many people’s knowledge of probability and statistics began and ended in their mathematics courses. Even though the two are all branches of mathematics, they are also used in data science. While probability tries to predict what will happen, statistics try to explain why something happens.

Probability measures the chance of occurrence of an event. It is that chance that something will happen. It predicts the likelihood that an event will occur or not in the future. The number 0 and 1 are used to quantify probability, where 1 indicates that there is a certainty of that event occurring while 0 indicates that the event will not occur.

Statistics, on the other hand, centres on analyzing the frequency at which past events occurred. This is applied mathematics and it tries to explain real-world situations from what occurred in the past. Therefore, Statistics makes use of data to explain events. The two types of statistics that apply to data are descriptive and inferential statistics. With the former, you can change raw data into information while the latter enables you to start from small samples to the whole domains of data (Skiena, 2001).

The question one may ask about Statistics is its relationship with Machine Learning. These two fields are closely related, and even, statisticians normally see ML as “Statistical Learning” or “Applied Statistics.” When presenting the field of Machine Learning to beginners, it is always assumed that they already have some background knowledge in Statistics. Therefore, what are the reasons for advising someone interested in Machine Learning to learn statistics?

Data is a raw observation, and in that state, you can’t use them as knowledge or as information. Therefore, when you have data, you must be prepared to answer the following questions:

  1. What observation is common or expected?
  2. What limits do these observations have?
  3. What is the nature of the data?

These questions may seem simple, but one needs to answer them so that those mere observations can be transformed into information which will be usable and beneficial. Apart from the raw data you gather, sometimes, you may need to design some experiments so that you can collect observations as well. If you conduct experiments, other questions may arise such as:

  1. What are the most relevant variables in these experiments?
  2. What difference is there in the outcome from two experiments?
  3. Can I trust these differences to be real or are they as a result of noise present in the experimental data?

To answer these questions which of course matter a lot, you need to employ statistical methods. Now you see why knowledge of statistics matters in Machine Learning. You need to understand the ML training dataset and also be able to interpret each result you get from testing different models used during the Machine Learning process. When you start executing the steps of predictive modelling, you will come to realize the need for statistical methods.

Two of the types of statistics that are relevant to Machine Learning are Descriptive Statistics and Inferential Statistics.

Descriptive Statistics: These are methods through which reviewed raw observations is summerized into usable information. People usually think that descriptive statistics ends in calculating values on available data samples in a bid to summarize its properties such as standard deviation, variance, median or mean. However, this statistics encompasses of those graphical methods which we can use to visualize data samples.

Inferential Statistics: These are those methods used to quantify properties of a sample. Sometimes, the tools of inferential statistics can be used to quantify observing data samples when there is an assumption (Brownie, 2018).

Statistics is very important in Machine Learning. If you must succeed in the ML projects you want to embark on,   knowledge of statistics is a good and compulsory idea.

From the brief explanation, we can understand that both disciplines work together because statistical analysis relies on probability distributions to conclude its findings.

Due to the uncertainties that fraught our existence on a daily basis, it is advisable to understand probability and statistics. With this knowledge, we can understand uncertainties and also make the right judgment when they occur based on past data or predictions.

Unsupervised Learning

This term refers to one of the approaches used in training machines how to learn. In unsupervised learning, the algorithms are left to explore data and uncover the structure within. The system doesn’t have labelled or classified information which it can use to give an accurate output. It solely depends on the available data to identify patterns or spot similarities that can divide the data it has into categories (Heath, 2018).

This method of learning works better when there is transactional data. For instance, the system can easily segment customers based on those attributes that are similar to them or those attributes that differentiate them in order to facilitate the marketing campaign program of an organization. Some of the techniques that enable algorithms to accomplish such are k-means clustering, self-organizing maps, singular value decomposition and nearest-neighbor mapping.

Supervised Learning

In supervised learning, algorithms make use of labelled data to come up with the right output. The system already has what is called a training data that contains training examples made up of the input object and the output value which is desired.

With this training data, the algorithm’s work is simplified. It will simply analyze the data and come up with the inferred function for mapping out new examples. In other words, the algorithm applies past knowledge to a new set of data to predict events that can occur in the future.

Also, since the system has the intended output, it can compare between the new output and the intended output to ensure that there is an agreement. However, if the two outputs are different, find out where the errors occurred for model modification (Expert System, 2018).

The truth of the matter is that supervised training of systems will not be successful without the availability of labelled data in large quantities. Sometimes, the systems you aim to train may need millions of likely examples which must be fed into the system before they can master that particular task. Due to these requirements, you need a vast dataset to succeed in supervised training.

The good news here is that with sources such as Google’s Open Images Dataset, which contains about nine million images, and YouTube-8M, which have links to seven million labelled videos and finally, ImageNet with 14 million images in different categories, you can find excess datasets. Also, Facebook is another source to get 3.5 billion images  from Instagram alone. At least, if you use one billion photos from this pool to train a system on image recognition, you are sure of at least 85.4% accuracy.

The next thing you need to know about supervised learning is that it is not easy to label the dataset you will use for the session. However, if you use crowd working services like that of Amazon Mechanical Turk, you can handle the process easily. Another way to this is by following Facebook’s measure of training systems with publicly-available data. So, instead of manual labelling, you can use billion-strong datasets (Heath, 2018).

Deep Learning – DL

When you read topics like Artificial Intelligence and Machine Learning, you will come across a subset of ML known as Deep Learning (DL). Without a simple explanation, this aspect of Computer Science may leave you perplexed. However, if you have been gradually following our simple explanations of Artificial Intelligence to Machine Learning, it will be easy to grasp what Deep Learning is all about and how it relates to ML.

DL is a part of Machine Learning which utilizes neural networks to learn under supervised, unsupervised or semi-supervised conditions. As humans learn from their experience, the DL algorithms learn by performing a particular task repeatedly. As it performs this taskover and over, it will continue to improve the process until it arrives at an acceptable outcome (Marr, What is deep learning, 2018).

This Machine Learning subset teaches a computer how to perform some tasks from sound, images or texts. The best part is that sometimes, the algorithm can be surprisingly more accurate than human performance.

Since the system makes use of architectures like deep neural networks, recurrent neural networks and deep belief networks to learn, it can solve any complex problem no matter the type or format of the available data.

Some of the applications of Deep Learning in the real world are virtual assistants, translations, driverless trucks, drones, autonomous cars, service bots, chatbots, image colorization, facial recognition, personalized shopping, pharmaceuticals, entertainment, etc (Marr, What is deep learning, 2018).

Leave a Reply

Your email address will not be published. Required fields are marked *