Machine learning: the importance of decision automation ~ TecnoBlog

What is machine learning and how do I know when to use it?

When we are at work, in our company or elsewhere, we face different problems. Some of these issues can be: how to improve the product, how to gain more customers, or how to make better use of resources. In addition, you are probably the person responsible for solving them. Could you solve it with Artificial Intelligence? And if so, how could you do it?

If you want to fix a problem and are wondering how to do it with artificial intelligence, read on. The goal is that you can:

• Understand the basic concepts to propose a solution with Machine Learning.
• Decide when to use and when not to use machine learning methods.

What is machine learning?

Let’s start by defining Machine Learning. Without being so rigorous, we can say that a machine has artificial intelligence if it can to interpret Data, to learn of them and use it awareness adapt and achieve specific goals.

Pay attention to the highlighted words in the previous paragraph: Interpret, Learn and Know. These words are the key to understanding what Machine learning and the value it can generate. All of this, of course, orbiting the data … the wonderful data. Let’s analyze this with an example.

• Data as the primary source of learning depends on the problem and can be structured as arrays of values, or unstructured as images, text, voice, among others.
• Which can interpret and learn data is varied, such as the recognition of faces in an image or the speed at which an autonomous car must go.
• The knowledge use it focuses on solving the problem. A machine learning system can learn to identify faces and, in turn, this knowledge can be used, for example, to grant access to a property or authorize a purchase. The advantage of machine learning is that it gives computers the ability to learn how to solve a problem without having to be explicitly programmed to do so.. How can we use machine learning to solve a problem and generate value?

How to Generate Value with Machine Learning

All around us we find different uses of machine learning. For example Netflix, Amazon or Spotify, use recommendation systems to suggest products or content. We can also find artificial intelligence in applications as different as autonomous cars developed by Google or in medical diagnostic systems.

Think about your job or your daily activities, what application do you think machine learning could have? Are you sure the app can be solved using machine learning methods? Let’s see how you can answer these questions. Let’s go back to our examples, think about what they have in common?

• Recommendation systems should decide what product to offer after reviewing your search history information.
• An autonomous car must decide, among other things, what speed is safe after knowing information about the characteristics of the terrain.
• A medical diagnostic system must decide if a patient has skin cancer after learning about, for example, the characteristics of a mark or mole on the skin.

Well. The reason machine learning is useful and generates value is that it works for automate decision making using the data related to the problem as input.

Our previous examples could be solved by humans, but if the information or tasks acquire a very large volume, these actions become impossible. Just as some manual tasks can be automated with robotic arms, with machine learning we are replacing mental data analysis tasks with machine learning systems. Now, from the example you thought of, what decision can be automated? What data can be used?

So far, we have talked about two elements necessary for a machine learning system: What does my system need to learn to decide? and What data do I need for this?
A third element is a data analysis model or machine learning method. This method is the one that takes the input data and processes it to provide the automatic decision as output. What steps should you take to resolve a machine learning issue?

Machine Learning VS “Classic” Programming (Explicit)

You can find this type of example on my blog Blog – Machine Learning Blog

As we have already commented, artificial intelligence generates value by automating decisions.

Just as a robotic arm automates manual tasks, machine learning (a subdomain of artificial intelligence) automates “mental” tasks. A classic example is spam or junk mail filtering. A person could identify which emails are spam and which are not, but due to the huge amount of emails received, this task becomes impossible to perform. This is where artificial intelligence comes in, learning, from examples, to automatically identify spam.

Consider the following problem: convert from Celsius to Fahrenheit, where the approximate formula is:

  f = celsiustimes 1.8+32

Now let’s see the solution:

• a) Using structured (normal) programming
• b) Use machine learning

a) In “normal” programming, you must give the instructions explicitly. The solution in Python would be

def celsius2fahrenheit(celsius:float) -> float:
	fahrenheit = celsius*1.8 + 32
	return fahrenheit
f = celsius2fahrenheit(23)
print(f)

The exit here is 73.4

b) In Machine Learning, it is not necessary to explicitly indicate the algorithm. You only skip examples and the machine learning technique only finds “the algorithm” for the solution.

Let’s start with the examples:

import numpy as np
celsius_q = np.array([[-40], [-10], [0], [8], [15], [22], [38]], dtype=float)
fahrenheit_a = np.array([-40, 14, 32, 46, 59, 72, 100], dtype=float)
X = celsius_q
y = fahrenheit_a

Then

from sklearn.svm import SVR
#TÉCNICA DE MACHINE LEARNING
svr_lin = SVR(kernel="linear", C=100, gamma="auto")
#ESTE MÉTODO "ENTRENA" AL MODELO PARA QUE APRENDA
svr_lin.fit(X, y) 
# PREDECIMOS YA QUE "APRENDIÓ"
svr_lin.predict([[23]])

The exit here is 73 185

As you can see, we got closer. It’s a good solution, but we can improve it.

In the previous example, we saw how machine learning learns by examples without needing to explicitly state the steps to take to solve a problem. So, after reading this article, how would you answer the opening question… what do you think is the value of artificial intelligence systems?