Types of AI models
Last updated
SkillfulAI_Docs_v1.0
Last updated
There is more to AI than just GPT models!
Below is a concise overview of six key types of AI models, showing their core features and capabilities:
Machine learning are algorithms that enable computers to learn from data without being explicitly programmed. These models identify patterns and make predictions based on the input data.
For example, to train an AI model to recognize different types of vehicles, a dataset with labelled images of vehicles, like cars, trucks, and motorcycles, is collected. The model learns from these images by identifying patterns and features that make each vehicle type unique. Once trained, the model can look at a new image and decide whether it shows a car, a truck, or something else.
Examples:
Predictive Analytics: A retail company can use ML to predict future sales based on past trends.
Recommendation Systems: Netflix suggests movies based on user preferences.
LLMs, like GPT-4, are built upon advanced ML techniques. They learn patterns from massive datasets to generate their output. The training process involves optimizing the model to predict the next word in a sentence, a fundamental ML problem. But they do require large datasets and can sometimes be a 'black box'.
Deep learning is a subset of ML that uses neural networks with multiple layers (hence "deep"). It excels at handling “unstructured data” like images, audio, and text.
The input layer takes raw data, hidden layers analyse and process it, and the output layer generates the result. Unlike basic neural networks with a few layers, deep learning networks can have many layers, allowing them to handle complex tasks and identify patterns that simpler methods cannot.
Example:
Image Recognition: Deep learning powers facial recognition systems on smartphones.
Speech Recognition: Virtual assistants like Siri or Alexa convert speech to text using DL.
LLMs are primarily deep learning models. They rely on transformer architectures, a specific type of neural network, to process and generate text. The deep layers of these networks allow LLMs to capture complex linguistic patterns and understand context better than traditional ML models.
It is one of the most widely used methods in machine learning.
In this, the model learns from "labeled datasets". Each input is paired with the correct output, and the model learns to map inputs to outputs.
It is called "supervised" because the model learns from datasets that have been labeled by humans. These labels guide the model, showing it how to classify or process the data based on the examples provided.
For example, a supervised learning model could be trained to identify fruit types using labeled images of apples, bananas, and oranges. The model learns the patterns in the images to match them with the correct fruit labels. After training, the model can look at a new, unlabeled image and predict what fruit it shows, based on what it has learned from the labeled data.
Example:
Spam Detection: Emails labeled as "spam" or "not spam" train a model to filter emails.
Image Classification: A dataset of images labeled as "cat" or "dog" helps a model identify animals in new pictures.
LLMs incorporate supervised learning during their fine-tuning phase. After pretraining on a vast corpus, they are refined using labeled datasets for specific tasks, such as summarization or sentiment analysis.
Unsupervised learning works with "unlabeled" data. The model identifies patterns or clusters within the data without explicit guidance or needing instructions from a human. The model is built to process and learn from the data independently.
An unsupervised learning model can group the various types of fruits without labels based on shared characteristics, such as size, shape, or color. Over time, the model can refine these groupings, making them more detailed, like distinguishing between small red fruits (e.g., cherries) and large red fruits (e.g., apples).
Example:
Customer Segmentation: Businesses group customers based on purchasing behavior.
Anomaly Detection: Detecting unusual transactions in financial systems.
The pretraining phase of LLMs is largely unsupervised. They are exposed to enormous amounts of text data without labels and learn linguistic structures, relationships, and patterns independently. This phase helps them understand the nuances of language.
NLP focuses on the interaction between computers and human language, enabling machines to understand, interpret, and generate text.
Example:
Chatbots: Customer support bots that answer queries in natural language.
Translation Tools: Google Translate converting text between languages.
LLMs are a core part of modern NLP. They take NLP to new heights by generating human-like text, understanding context, and performing a wide range of language-related tasks such as translation, summarization, and question answering.
Reinforcement learning trains algorithms by rewarding good actions and penalizing bad ones. In this approach, an agent interacts with an environment to achieve a specific goal.
For example, reinforcement learning is often used to teach robots how to perform tasks like sorting objects or assembling parts. It is also applied in game development, where AI agents learn to play games by trial and error to achieve higher scores or complete levels more efficiently.
