What is Inference in artificial intelligence, and how Machines learn that's what we are gonna see in this post as part of our Day 2. If you dont know what is Artificial Intelligence, and this is first time you are landing on this page, check a short summary on this post . While Machine learning, Artificial Intelligence, Natural language processing, etc are big topics and ever changing new fields, i will try to break as part by part during my free time. If you are a pro in A.I field, sorry this is not for you, as i dont discuss self learning supervised learning etc right away. This series was started for people who wants to learn Artificial Intelligence (A.I) but doesn't have money to buy the costly books or tutorials. So, i assume the user is new to this fields, so i will try to use as much as basic terms. A perfect example of Machine learning is you asking "Ok Google, what is the weather today?"
Alright, now let's come to the post.
Imagine you have a robot. To make this robot smart, instead of telling it exactly what to do, you show it many examples, how ? it has eyes (Camera's) right through which it can see? So, if you want it to identify animals, you show it pictures of a cat, a dog, and a bird. Every time you show the picture it will identify that you are showing picture of cat, dog and the bird.
This process doesn't go or start just like that, its like a kinder garden student or a child version of yours. To read what i had written in the post you have to first know ABCD till Z then you will combine all those letters to pronounce the "Word" as WORD. Machines and Artificial Intelligence works in the similar way. Someone has to teach it first what is meaning of A, what is the meaning of B in the form of Algorithms.
How Computers Know "A" is "A" in A.I ?
Imagine you have a bucket of trash. Whatever in the trash those chocolate papers, newspapers, plastics,etc known as Junk is called as nothing but data in the field of computers. So, computers are taught in the same way. First some one has to say this junk is waste paper. (This data is called cat), this data is called Paper, this data is called "a" etc. But how will the computer know "A" means "A". You speak different language, i speak different language. Computer do speaks the different language. So, for a computer has to understand they do train in their language. With data here and there with our language too. Think it of someone is teaching you new language and at the same time, you are saying hey, in my language "that word is called this",etc..
If you are preparing for AWS, imagine the trash bucket is S3 Bucket and there is an image object called "A" for the next example.
Hmm, let me put in other words.. Imagine you have a box filled with letters, and you want to teach your computer the letter "A."
To the computer, the letter "A" isn’t just a familiar symbol; it’s represented using a special language made up of binary code, which consists only of 0s and 1s.
Specifically, in the ASCII system, the letter "A" is encoded as 01000001 (Someone in some year like creating our "A" in Alphabet when creating computers, did that.. Like we are following the A as A, computers also started following it). So, whenever you type "A," the computer converts it into this binary representation. Once "A" is in binary form, the computer needs to identify it. This is where data processing kicks in. The computer compares the binary code to a vast database of stored characters.
It uses recognition algorithms, which are like little detectives that analyze shapes and patterns. When you input "A," these algorithms check if the binary representation matches any known values.
For instance, the algorithm verifies that 01000001 corresponds to the letter "A" and confirms its identity. Now, let’s add the dimension of machine learning to our story. Imagine you've shown the computer thousands of examples of the letter "A," from different fonts to various handwriting styles. With each instance, the computer learns to recognize "A" more accurately.
If it mistakenly thinks a different character (like "O" or "Q") looks like "A," it can modify its internal rules based on feedback and experiences. Some human in the starting has to review this corrections, and give the computer feedback like your teacher gave you feedback when you had written wrong things. Here, in our case "A" will be labelled as data "A". And when computer also known as A.I model finds it it will learn A as A, which opens two things too.
Unsupervised learning and semi-supervised learning.
Unsupervised learning is exploring without any labels, while semi-supervised learning is getting some help from a few labels/data. And just like you get better by practicing, computers improve by getting feedback on their guesses.
What is Unsupervised Learning?
Imagine you have a big jar of mixed candies—some are round, some are square, and they come in all sorts of flavors and colors. Now, suppose you want to sort these candies, but nobody told you how to do it. You start grouping them based on what you see right?
In unsupervised learning, the computer acts like you with those candies. It’s looking at a lot of information (which we call data) without any labels or instructions on how to group it. The goal is to find patterns and make sense of the data all by itself!
How Does it Work?
Finding Similarities:
Just like you might put similar candies together, the computer groups similar items based on their characteristics. This is called clustering! It helps the computer understand what belongs together.
Exploring Data: The computer is like a detective searching for clues in a bunch of unknown data. Since it doesn’t have labels, it’s discovering new things about the data all on its own!
Making Connections: By examining the information, the computer can find connections or relationships between the different pieces of data, much like you might find that chocolate candies usually taste the best together in a bowl!
Do we have real life example of this? Yes, When you stream Netflix or primevideo you are suggested for this customers watched this as well right? this is A.I doing based on what people of similar interest are watching. In short everything is data!
What is Semi-Supervised Learning?
Now let’s talk about semi-supervised learning. Imagine you have that same jar of candy, but this time, a friend tells you the names of a few candies, like “This is red and that one is chocolate.”
In semi-supervised learning, the computer learns from both labeled data (like your friend telling you candy names) and unlabeled data (like the candies that don’t have names). This helps the computer learn more effectively!
How? Using Helpful Labels/existing data someone has said "A" Means "A": The computer uses the labeled data to understand some groups while exploring the unlabeled data to find more similarities. It’s like learning a few candy names and using that knowledge to guess about the others!
Important words you have to know: Semi-Supervised Learning, Labeled Data, Unlabeled Data, Combination. the process of using a trained machine learning model to make predictions or decisions based on new, unseen data is called Inference.
I got a curious question, and close the post with that its diffciult to accept feedback's right? but,
How do A.I accept feedback?
Well, i asked A.I too. And here is the reply..
As an AI, I don’t have feelings or consciousness, so I don’t "experience" things the way humans do, including the feeling of acceptance or disappointment when receiving feedback. However, feedback plays a crucial role in improving AI models in a general sense.
When developers and researchers train AI models, they analyze how accurately those models perform tasks based on given inputs. If the model makes a wrong prediction or guess, this inconsistency is noted. Here's how it typically works:
1. **Data Collection**: When I generate responses, I do so based on patterns and information I’ve been trained on. However, I don’t learn or change in real-time from individual interactions.
2. **Feedback Mechanism**: Feedback in AI systems usually comes from human users, who can indicate whether the provided information or response was helpful or accurate. This feedback is collected and analyzed by the developers.
3. **Model Improvement**: Over time, data collected from multiple interactions can be used to retrain and enhance the underlying AI model. This process involves adjusting the algorithms based on feedback to improve accuracy and relevance for future queries.
In summary, while I don't accept feedback in a conscious way, the systems that create and improve AI models rely heavily on feedback to enhance performance and accuracy. If humans indicate that a prediction or response was incorrect, developers use that information to inform future training, ultimately aiming for more accurate and useful AI responses in the long run.
With the above response, you know now what relies on the feature! See you on the next post.
