Introduction to Artificial Intelligence, Machine learning, and Deep Learning

Machine Learning is a process by which we train a device to learn some knowledge and use the awareness of that acquired information to make decisions. For instance, let us consider an application of machine learning in sales. It is pretty common to see a spike in sales during the festive seasons. Therefore, during festive occasions, various sellers give different offers to attract more customers and generate more sales.

All these adjustments to the usual sales process enhance the sales, leading to higher profits. Customers are inclined to buy more products under attractive offers like, big sales, discounts, free or 1+1 deals, refer and get bonuses. For humans, these adjustments in pricing are easily perceivable as these are a part of our real-life experiences. However, that’s not the case with machines as they should be trained to pick up these changes. 

 

Machine Learning

Machine learning (ML), is a form of artificial intelligence (AI), that prepares software programs and applications to predict outcomes more accurately without having to be explicitly instructed to do so. In order to forecast new output values, machine learning algorithms use historical data as input.

In the former example, machines can be taught to detect patterns or features in data that could potentially boost product sales. These patterns are extremely subtle and are not perceivable to human senses. This is where machines come into play. 

Can intelligent machines deduct patterns? 

Can we use it for predictions in advance, so that we have time for modifying our current process? 

These are some of the questions that need to be answered before we teach a machine. Thanks to technology, computation is faster nowadays. This means we can make predictions in real-time or in advance or at a later stage. 

But what goes into training a machine?

This is where machine learning comes in. Before getting into what machine learning is, let us learn a little about Artificial intelligence.

 

Artificial Intelligence

Artificial intelligence is a branch of studies that tries to make a machine think and act like a human. In short, any computer program that is intelligent is Artificial Intelligence (AI). Humans can respond to various stimuli. In Artificial Intelligence, machines are modified to respond to stimuli like humans. Stimuli present to us in various forms. It can be optical, auditory, or via touch. It can also be a combination of multiple sensory inputs. 

For instance, when you are reading this statement, your eyes see the optical output from the screen. As you read, you understand what is given in this article. This is possible because you are processing the language. Once you have processed the language, you pick up the important information from this statement.

 For instance, let me ask you "How are you?"

Your eyes send information and you identify that the first letter is H, the second is O third is W. So the word is how. When you are reading "How are you?" you process it linguistically and realize that somebody is asking about you and then you choose to answer it with yes I am fine or No I am not fine.

Instead of asking "How are you?" I asked "Como Estas?"

Your eyes will send the information and you will identify that the first letter is C, the second is O, etc. But then you will not be able to process it linguistically if you are not aware of Spanish.

 

There are three main ways our brains have worked, 

• One- processing the optical data 
• Two- processing the meaning of the statement 
• Three- making a decision and responding accordingly

Can we have a machine that can do the same? 

If we can accomplish that, we are much closer to artificial intelligence. The main task is to translate the human perception to the machine so that it can learn and try to make an appropriate decision.

And this is where we are now.

 

 

 Tackling the Challenge of Transmitting Human Perception through Machine Learning

 

   

 

 

Machine learning is a subset of Artificial Intelligence where traditional algorithms and statistical prediction models are used to make decisions. Some of the machine learning algorithms we use day in and day out are spam filters and recommendation systems.  

Broadly, machine learning algorithms are classified as supervised, unsupervised, and reinforcement learning.

In the supervised algorithm, we teach the model what is what. For instance, to classify between dog and cat pictures, let us assume we create a model. Now to this model we feed the data that a certain picture is a cat and another picture is a dog. Likewise, we do it for a large collection of pictures of dogs and cats. Each picture that we present to the model is labeled either a cat or dog. This mapping of the input to the output is the main characteristic of the supervised machine learning algorithm.

In an unsupervised machine learning algorithm, we do not label the input. We feed the pictures of the cat and dog to the model and ask the model to learn the difference between these two.

In reinforcement learning, the model is given a punishment or reward based on its prediction. If the machine predicts a cat as a cat, it is rewarded, while if the machine predicts a cat as a dog, then the model/algorithm/machine is punished.

 

 

Now, we require the model to distinguish between a cat and a dog. 

But how is the model going to identify the different animals? How are we going to teach a model what a cat is or what a dog is?

That’s where features become integral. In machine learning algorithms, we feed the distinct features to the model. Based on these features, the machine learning algorithms identify the cat or dog. For machine learning models, identifying the features is the critical element to arrive at accurate conclusions.

For example, an interesting feature to differentiate between a cat and a dog could be based on the ear position. For a cat, the ears are facing up, while for a dog the ears are hanging down. This can be entered into the model.

In case, we are unable to feed information on features due to lack of availability or awareness of details, deep learning becomes the answer.

In deep learning, we just give the input pictures to the model and let the model identify the features on its own.

The trade-off for not giving features needs to be compensated by large data. So yes, to set up a deep learning algorithm you need to have a huge dataset. Let us conclude with the following markers that we learned today.

                    

	Labeled	Not Labeled
Features Present	Supervised Machine Learning	Unsupervised Machine Learning
Features Not Present	Supervised Deep Learning	Unsupervised Deep Learning