Entry 4: AI, ML & DL

This week we explored Artificial Intelligence, Deep Learning, and Machine Learning. We also learned about State Space Search as well as ML algorithms.

Activity 1: Applications of ML and Deep Learning

As can be inferred from the above line, for this activity we had to search for examples where we can apply machine learning as well as deep learning.

Machine Learning involves using algorithms to create self-learning models trained on various data. It falls under Artificial Intelligence. Its applications are as follows:

• Recommendation Systems: This is one of the most popular applications of machine learning. Using different models, a website keeps track of your data and is used to predict the user's likes and dislikes.

• Image Recognition: Image recognition is another popular application of machine learning where a device is programmed to recognize various images based on the intensity of the pixels.

• Virtual Personal Assistants: This application involves both machine learning as well as speech recognition. One such example is the iPhone's Siri. It records your data and checks how accurate the data you get in return is.

Deep Learning is a subset of machine learning where the models learn by example (which is what we humans do) are trained by various sets of data. Its applications are as follows:

• Aerospace and Defense: It is used to identify objects such as satellites and to identify appropriate (safe) areas for troops.

• Automated Driving: Automotive researchers use deep learning to detect pedestrians which decreases accident rates and also helps to identify certain objects such as stop signs and traffic signals.

• Medical Research: It has helped immensely in the medical field. Cancer researchers use deep learning to identify cancer cells using high-dimensional data sets.

Activity 2: Depth-first Search

Depth-first search is an algorithm for searching tree or graph data structure. It involves exploring a node as much as possible before going back to the previous nodes.

Advantage: It requires less time and memory space to perform depth-first search as compared to other search algorithms.

Disadvantage: It does not guarantee a final solution. Due to the possibility of an infinite loop, it does not provide the shortest solution.

Activity 3: Teachable Machine

Activity 4: Classification Algorithm

This activity involved calculating the distance between two points to figure out which group the new point was pointing to.

Activity 5: ChatGPT

For this activity, we had to ask ChatGPT how it works and then ask it to explain the same as it would to a 5-year-old. Then, we had to decide which explanation made more sense.

How does ChatGPT work?

ChatGPT is a language model developed by OpenAI, and it's based on the GPT (Generative Pre-trained Transformer) architecture, specifically GPT-3.5. It works by using a combination of deep learning techniques, natural language processing, and large-scale training data to generate human-like text responses. Here's an overview of how it works:

• Pre-training: ChatGPT is pre-trained on a massive corpus of text data from the internet. During pre-training, the model learns to predict the next word in a sentence based on the context of the words that came before it. This process helps the model learn grammar, vocabulary, facts, and some level of reasoning abilities.

• Transformer Architecture: The core of ChatGPT is the transformer architecture, which is a type of neural network model. The transformer model is known for its ability to handle long-range dependencies in sequences, making it well-suited for natural language understanding and generation tasks.

• Fine-tuning: After pre-training, ChatGPT is fine-tuned on specific tasks or datasets to make it more useful and controlled. This fine-tuning process allows OpenAI to customize the model for tasks like text completion, question-answering, language translation, and more. It also involves adjusting the model's behavior to adhere to guidelines and policies set by OpenAI.

How does ChatGPT work? (if explained to a 5-year-old)

• Sure! ChatGPT is like a smart computer that can talk like people. It learned to talk by reading lots of books and stuff on the internet. It's like when you learn to talk by listening to grown-ups.

• You can ask ChatGPT questions, and it will try to give you answers that make sense. Just type in what you want to ask, and it will write back to you. It's like having a chat with a friendly robot.

• But the robot has rules to make sure it says nice things and doesn't say mean or silly stuff. It's like your mom and dad telling you to be polite and kind when you talk to others.

• So, it's a computer friend that talks like a person and follows some important rules to be safe and helpful.

I think both explanations make sense. But the first answer is more technical and requires you to have knowledge about certain terms. The second answer is less descriptive and is simple enough for people without such knowledge to easily understand how ChatGPT works.

References

• Coursera (2023) What Is Machine Learning? Definition, Types, and Examples Available at: What Is Machine Learning? Definition, Types, and Examples | Coursera (Accessed: 8 November 2023)

• Coursera (2023) 9 Real-Life Machine Learning Examples Available at: 9 Real-Life Machine Learning Examples | Coursera (Accessed: 8 November 2023)

• MathWorks (no date) What is Deep Learning? 3 things you need to know Available at: What Is Deep Learning? | How It Works, Techniques & Applications - MATLAB & Simulink (mathworks.com) (Accessed: 8 November 2023)

• GeeksforGeeks (2023) Depth First Search or DFS for a Graph Available at: Depth First Search or DFS for a Graph - GeeksforGeeks (Accessed: 8 November 2023)