What Even is Machine Learning?
I wanted to start exploring Chapter 3 of Learning AI with ChatGPT and Google Colab: The Easiest, Quickest Way to Start Learning AI by Dr. Kelsey with a clear understanding of the differences between Artificial Intelligence (AI) and Machine Learning (ML). Naturally, I went to ChatGPT for help.
What is Artificial Intelligence?
According to ChatGPT, "AI is a broader concept that refers to machines or systems capable of performing tasks that typically require human intelligence." Some well-known examples of AI are virtual personal assistants like Siri, Alexa, and Google Assistant. Other examples are the algorithms in your streaming platforms that recommend you what to watch next based on your previous viewing history. Making waves now are self-driving cars and drones that use AI to navigate the open roads or skies. If you've ever tried to get customer service through the Amazon app, you most likely talked to chatbot first. That's another example of AI. The list goes on and on. As you can clearly see, AI has become a natural part of our every day lives.
What is Machine Learning?
How about ML? Well, ML is a subset of AI. As ChatGPT explains it, ML "is the method through which artificial intelligence is achieved. ML focuses on the development of algorithms that can learn from and make predictions or decisions based on data. Rather than being explicitly programmed for a specific task, ML systems learn from data and improve their performance over time." ChatGPT and DALL-E are perfect examples of ML at its finest, going through the testing data again and again to refine results as each simulation is performed.
It seems as if ML depends on AI, and not the other way around. So I again asked ChatGPT about this relationship between AI and ML. It turns out that not all AI required ML. According to ChatGPT, "Early AI systems were based on hardcoded rules and logic (like expert systems) without the capacity for learning. These systems are still considered AI but do not involve ML." Some examples are basic home automation systems (like programmable thermostats or sprinkler systems) and industrial automation systems.
Lucky for me, I just finished a course on ML here at Benedictine, so I have a good base understanding of how these algorithms work, albeit, we use R in this program as opposed to Python.
There are four types of learning that can be achieved with ML: supervised learning, unsupervised learning, semi-supervised learning, and reinforcement learning.
Supervised Learning
Supervised Learning happens when all data points are already associated with a label or outcome. Take for example a credit scoring algorithm. All of the individual data points (history, revolving utilization, available credit, missed payments) individually determine whether a person has a good or bad credit score. In other words, supervised learning is ideal for creating a map from inputs to outputs from the training data, and create a prediction for new outcomes. Thinking back on our credit score example, inputting credit history, revolving utilization, available credit, missed payments for a new client, the ML can accurately predict the new client's credit score.
There are two types of problems solved with supervised learning: classification and regression. Classification, I think, is the simplest. It refers to assigning a value a label. For example, if an email contains the word "viagra," then it is likely to be spam. On the other hand, regression problems require the ML algorithm to work with continuous values, such as our credit score example.
The algorithms associated with supervised learning are decision trees, linear regressions, logistic regressions, support vector machines, and neural networks.
Unsupervised Learning
Unlike supervised learning, in unsupervised learning algorithms there are no clear-cut answers. Unsupervised learning seeks to explore the structure of data to find patterns groupings or correlations. Some examples of unsupervised learning are clustering, association, and dimensionality reduction. Unsupervised learning is actually very important to marketers, as this is how we can come up with customer segmentation and market basket analysis.
The types of algorithms you can create using unsupervised learning are k-means clustering, hierarchical clustering, priori algorithm, and principal component analysis.
Semi-Supervised Learning
Reinforcement Learning
Reinforcement learning might be the most exciting type of algorithm that we have heard news from in a while. The news article about the algorithm that beat the world champion of the game "Go" comes to mind. Reinforcement learning is where the algorithm learns to make decisions by performing actions within a controlled environment to received rewards or penalties. The ultimate goal for this model is to learn strategy or policy, and maximize results over time.
Other applications of reinforcement learning are in robotics (such a robotic arm learning to pick up objects) and self-driving cars.
Conclusion
I again asked ChatGPT what type of ML algorithm is the most interesting to learn. In true ChatGPT nature, it gave me a very political answer for each. So if you're interested in learning more about ML, this might help you choose where to start:
- Supervised Learning: If you're interested in applications where you have labeled data and your goal is to predict outcomes (like image classification, spam detection, or predicting market trends), supervised learning is a great choice. It's the most common form of machine learning and is widely applicable in many industries.
- Unsupervised Learning: Choose this if you're fascinated by discovering hidden patterns and structures in data, especially when you don't have labeled data. It's ideal for segmentation, association, and dimensionality reduction tasks, like market basket analysis, customer segmentation, or feature extraction.
- Semi-Supervised Learning: This is a middle ground between supervised and unsupervised learning. If you have limited labeled data and a large amount of unlabeled data, and you're interested in problems like natural language processing or image recognition where labeling data is costly, this could be a good choice.
- Reinforcement Learning: If you're intrigued by the idea of agents learning to make decisions in an environment (like robotics, game playing, or autonomous vehicles), and enjoy problem-solving in a dynamic setting, reinforcement learning is an exciting field to explore.
Comments
Post a Comment