That’s true that everyone is talking about AI. Most precisely about GenAI. GenAI is a type of intelligence that can produce synthetic data like texts, pictures, words, etc. In a simple term, Generative AI is a type of AI that can generate new content based on a learning method from existing content. I think it’s a good start to look at what Google is proposing in the course for Generative AI. One of the biggest differences to understand is the difference between AI and Machine learning. AI is a branch of computer science that is in Systems that can act, reason, and autonomously behave like humans. Machine learning as a subset of AI gives the computer the ability to learn without manual programming effort.
The most common Machine Learning models are Supervised and Unsupervised machine models. In the Supervised, there is a tag with the data whereas in the Unsupervised model, there are none. Supervised data is all about “getting somewhere” with meaningful data whereas Unsupervised data from an Unsupervised machine model is all about “Discovery and research”. Let’s look at the various learning models.
Supervised Learning
Quite easy to understand the basic concept. In this learning model, data are input into a model which the model outputs based upon the training. Then the same output data is compared with what the model was taught. In the event the output data is very different from what the model was taught, it’s an error in the system. However, the model can re-input the data to itself until the closest and most accurate data is reached when compared. Cool right? Data are also associated with a tag. case example: A dog training app wants to classify different dog behaviors (e.g., sitting, barking, lying down) based on sensor data collected from a collar worn by the dog. Accurate behavior classification can help owners understand their dogs’ activities and needs better.
Unsupervised Learning
Data are input to a model that generates examples and there are no tags associated with it. case example: A dog training app wants to understand different patterns in dog activities throughout the day based on sensor data from collars worn by dogs. The goal is to cluster similar activity patterns together to provide insights into typical daily routines and identify unusual behavior that may indicate health or behavioral issues.
Reinforcement Learning
In a Reinforcement learning model, data are input to achieve a goal. The agent takes actions, observes the results of those actions, and receives feedback in the form of rewards or penalties. case example: The goal is to train a robotic dog (or a simulated dog in a virtual environment) to fetch a ball. The dog needs to learn to identify the ball, approach it, pick it up, and return it to the trainer.
Deep Learning
This model uses an AI neural network. The idea of neural networks came from the field of Cognitive psychology. The basic units of a neural network, are also called nodes or units. Each neuron receives inputs, processes them, and produces an output. Deep learning models have many layers of neurons which allows them to learn more complex patterns and it can use both label and non-label data. So new examples are output with the unlabel data and label data provide accurate information for the inputs. Deep learning is thus a semi-supervised learning. case example: The goal is to develop a system that can automatically recognize dog commands given by a trainer and the corresponding actions performed by the dog using video data. This system can help in assessing the effectiveness of the training and provide real-time feedback to the trainer.
GenAI is a subset of Deep learning
- It uses an AI neuron network (deep learning )
- Both labeled and unlabeled data
- Supervised, Unsupervised data and semi-supervised method
LLM – Large Language Model is a subset of deep learning. Large language models are a specific application of deep learning within the domain of natural language processing.
Making it more clear to understand
A deep learning model refers to any neural network architecture that consists of multiple layers (hence “deep”), enabling the model to learn hierarchical representations of data.
Large language model specifically refers to a type of deep learning model that is tailored for natural language processing tasks.
Discriminative and Generative model
The deep learning model or Machine learning model can be divided into 2 model types: Generative and Discriminative model.
Discriminative models focus on classifying models for data points. They focus on drawing a boundary between different classes or categories in the data. Think of them as tools for deciding which category something belongs to based on its features. When a discriminative model is trained, it can be used to train and predict the label for the new data.
Generative models generate new data based on what they learned. They focus on understanding how the data is generated. They try to model the entire process that could produce the data. Think of them as tools for creating new data points that look like the ones you’ve seen before.
Introduction to Generative AI by Google is definitely interesting and I believe it can set anyone for success who do not understand anything about it.
