When we think:
Can you imagine before Machine Learning came into your life?
The changes made so far are huge!
Google and the Machine Learning
A huge people majority, as you and us, use search engine like Google to obtain a wide range of information and answers. What does this search engine do?
It collect all the information based on your search query and it presents you with the most relevant results.
Without Google of Bing, the task will be hard. Because you would have to go through dozens or hundreds of books and articles, wouldn’t you agree?
It’s all the work of machine learning that simplifies this previously tedious task.
Speech and vocal recognition
A long time ago, in a faraway galaxy…
Speech and visual recognition were a concept shows in science fictions films. Machine Learning made it possible today.
By the way, did you know that you may be using it to your advantage much more often than you think?
For example, Facebook automatically recognizes the people in a photo and tags them for you, saving you a lot of time….
Without Machine Learning, Siri, Cortana or Iris would not be able to answer your questions in real time.
What is Machine Learning?
In one sentence?
Machine Learning is a generic term means for a set of techniques and tools that help computers learn and adapt on their own. Machine Learning algorithms help artificial intelligence (AI) learn without being explicitly programmed to perform the desired action. They can predict and execute tasks based on the learned model.
Thanks to Machine Learning, our world has seen systems that facilitate human work.
Machine Learning: How does it work?
1. Data Collection:
As you know, machines initially learn from the data you provide them.
You must collect reliable data so that your machine learning model can be the right patterns. The quality of the data that you provide to the machine will determine the accuracy of your model.
If you have incorrect or outdated data, you will get incorrect results or predictions that are not adequate.
What is good data? It is relevant, contains very few missing and repeated values, and has a good representation of different subcategory/classes present.
2. Data Preparation:
Once you have your data, you need to prepare it. You can go this by:
- Gathering all the data you have and randomizing it. This ensure that the data is evenly distributed and that the order does not affect the learning process.
- Clean up the data to remove unwanted data, missing values, rows and columns, duplicated values, data type conversion, etc. You may even need to restructure the dataset and change the row and columns or the row and columns index.
- Visualize the data to understand how it is structured and to understand the relationship between the different variables and classes present.
- Divide the cleaned data into two sets: a training set and a test set. The training set is the set from which your model learns. A test set is used to check the accuracy of your model after training.
3. Choosing a model:
A Machine Learning model determines the results you get after running an algorithm on the collected data.
It is important to choose a model that is relevant to the task at hand.
Over the years, scientists and engineers developed various models suitable for different tasks, such as speech recognition, image recognition, predictions, etc.
One last detail: You should also determine whether your model is suitable for numerical or categorical data and choose accordingly.
4. Training the model:
Training is the most important step in Machine Learning. During training, you feed the prepared data to your machine learning to find patterns or characteristics and make predictions.
In this way, it learns from the data in order to accomplish the task it has been given. Over the time, with training, it gets better at making predictions.
5. Evaluation of the model:
After training your model, you need to verify its performance.
To do this, you test the model’s performance on unpublished data.
If the test is performed on the same data that you used for training, you will not get an accurate measurement. This is because your model is already used to the data and finds the same characteristics in it as before.
When used on test data, you get an accurate measurement of your model’s performance and speed.
6. Adjustment of the parameter:
Once you have created and evaluated your model, see if its accuracy can be improved in any way.
To do this, adjust the parameters in your model.
Parameters are the variables in the model defined by the programmer. Accuracy will be highest with a particular value of one of your parameters consists in finding these values.
7. Making predictions:
Finally, you can use your model on unobserved data to make accurate predictions.
A concrete example of these steps?
Consider a system chose input data contains pictures of different kinds of fruit. You want the system to group the data according to the different types of fruits.
First, the system will analyze the input data. Then, it tries to find patterns, such as shapes, size and color.
Based on these patterns, the system ill try to predict the different types of fruits and separate them.
Finally, it keep track of all the decisions it makes in the process to make sure it learns.
The next time you ask the same system to predict and separate the different types of fruit, it won’t have to go through the whole process again.
This is how Machine Learning works.
How many Machine Learning is there?
Good question !
Because there are several “families” of Machine Learning:
- With supervised learning
- With unsupervised learning
- With reinforcement learning
Supervised learning uses labaled data (labeled examples) to train machine learning models.
In labeled data, the outcome is already known. The model just needs to match the inputs to the respective results.
This method needs to supervision to train machine learning models. Hence the name supervised. These machines need to be guided with additional information to get the desired result.
If you need to train a model or a system with computer vision to recognize an apple.
First, you provide dataset containing images of a type of fruit for example apples.
Then, you provide another data set that lets the model know that they are images of apples.
Next, provide a new dataset containing only images of apples. At this point the system can recognize the fruit in question and will remember it.
This is how supervised learning works. You train the model to perform a specific operation by itself.
Where do you find machines with supervised learning?
They are typically used to solve classification and regression problems. Among their main applications, you will find:
- Automatic language processing,
- Weather forecasting
- Stock price analysis
- Speech recognition
- Computer vision
Is there such a thing as unsupervised Machine Learning?
Yes, and you will discover that it works quite well…
Unsupervised Learning :
Unsupervised learning uses unlabeled data (unlabeled examples) to train machines. The unlabeled data has no fixed output variable. The model learns from the data, discovers patterns and features in the data, and returns the result.
How does it work?
Unsupervised learning finds models and understands trends in data to discover the result. Thus, the model attempts to label the data based on the characteristics of the input data. The training process used in unsupervised learning techniques requires no supervision to build models. They learn on their own and predict the results.
Unsupervised learning allows systems to identify patterns in data sets that would otherwise not be labeled or classified using AI algorithms.
Let’s dig in…
Let’s illustrate with an unsupervised learning machine that uses images of vehicles to determine whether it is a bus or a truck.
The model learns by identifying part of a vehicle, such as the length and width of the vehicle, front and rear hoods, roof tops, types of wheels used, etc.
Based on these characteristics, the model determines whether the vehicle is a bus or a truck.
For which applications?
Unsupervised learning is used to solve grouping and association problems.
You’ll find in a wide range of industries, including finance, e-commerce, healthcare, engineering, industry 4.0, security and gaming.
One example: Customer segmentation.
Based on customer behavior, likes and interests, you can segment and group similar customers into a group.
Now let’s look at the third type of Machine Learning: reinforcement learning.
Reinforcement learning is a mode of statistical learning.
It trains a machine to take appropriate actions and maximize its rewards in a particular situation. (As if AI wanted to improve and optimize its grades).
It uses an agent and an environment to produce actions and rewards. This agent has an initial state and a final state. But they may be different paths to the final state, like in a maze.
In this learning technique, there is no learning technique, there is no predefined target variable.
How does this machine learn?
Reinforcement learning follows trial and error methods to achieve the desired outcome. After completing a task, the agent receives a reward.
We could compare this to training dog to catch a ball. If the dog learns to catch the ball, you give it a reward, such as a cookie.
You want to train a machine that can identify the shape of an object from a list of different objects.
You gave this machine a set of data and ask it to identify a particular type of object (in this case, a suitcase). The machine tells you it’s a box, but that’s the wrong answer.
As feedback, you tell the system that it’s wrong: it is not a box, it is a suitcase. The machine then learns from this feedback and keeps this in mind. The next time you ask the same question, the system will give you the right answer; it will be able to tell you that it is a suitcase. This is a reinforced answer.
This is how reinforcement learning works: the system learns from its mistakes and experiences.
This model is used in the world of games. With reinforcement learning, the level of difficulty increases as you get better.
This type of Machine Learning is used to program robots for example. You no longer need a long and tedious development work. The computer will learn to operate, to react to this or that even or request by itself.
Whether the robot is physical (for industry or an autonomous vehicle) or virtual (for finance or security management), the learning phase will be executed in the form of digital simulations.
Machine Learning: A unique technology?
Machine Learning has been around for decades now.
And its role in our daily life is only growing.
- Computer power is increasing.
- The volume of data increasing.
- Internet bandwidth is expanding.
- Data scientists are improving their expertise…
Computers can learn more, remember more, and generate more accurate results through machine learning.
These machines increase efficiency and safety at work and at home.
The enable business to make critical decisions:
– To analyze and streamline their actions.
– To optimize their current operations.
– And to find new ways to improve their performance.
Need to dig deeper into this topic?
Then deepen your knowledge with one of the following articles:
[article coming soon] Deep Learning: is it a world of autonomous AI?
Why is this technology becoming indispensable for your organization?
[Upcoming article] Computer Vision: How does AI see?