Welcome to learning edition of the Data Pragmatist, your dose of all things data science and AI.

📖 Estimated Reading Time: 5 minutes. Missed our previous editions?

• K-Nearest Neighbor; Claude 3 better than GPT4?

• Sampling Techniques in Data Analysis, 5 Essential Lessons from a Junior Data Scientists at Spotify

Today we delve into Machine Learning methods. As part of our learning series, Top AI Tools for Productivity.

Do follow us on Linkedin and Twitter for more real-time updates.

— Arun Chinnachamy

Data Science/Analysis Job Roles to Pursue Right Now

• Data Mining Expert at Plant-A Insights Group

• Data Engineer & Web Crawler at hush.ai

• Quality Analyst (FinCrime) at Revolut

• Data Collector - WFH - No experience needed at TransPerfect

• Data Science - Technical Lead at One97 Communications Limited

🧠 Machine Learning methods

Machine Learning is a subset of AI that focuses on enabling machines to improve with experience using statistical methods. It involves designing algorithms that can learn and enhance over time by observing new data. The goal is to derive meaning from data, making it the key to unlocking Machine Learning. With more qualified data, Machine Learning algorithms become more accurate in making decisions and carrying out tasks autonomously.

Machine Learning methods

Even though there are a number of approaches are used in Machine Learning, the most popular ones are as follows:

1. Supervised Learning

   Supervised Learning uses labeled data to train machines, allowing them to learn from past experiences. With a larger dataset, machines can gain more insight into the subject. Once trained, machines can predict outcomes when given new, unseen data. This method is often applied in scenarios where historical data predicts future events, such as detecting fraudulent credit card transactions or assessing insurance claim likelihood.

   

   An example for Supervised Learning based on topics, picture by Google Developer Tutorial

2. Unsupervised Learning

   Unsupervised Learning involves training machines without labeled data. Similar to listening to a foreign language podcast without understanding, the algorithms learn from the data's inherent structure without explicit guidance. While one podcast may not provide much insight, exposure to numerous podcasts enables the brain to form a language model, recognize patterns, and anticipate certain sounds. Techniques like self-organizing maps, nearest-neighbor mapping, and k-means clustering are utilized to explore the data and identify underlying structures.

   

   An example of clustering in k-means algorithm, picture by Google Developer Tutorial

   K-means clustering is one of the most popular unsupervised machine learning algorithms. that aims to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean that is called cluster centers or cluster centroid.

3. Reinforcement Learning

   Reinforcement Learning, akin to unsupervised learning, lacks labeled data but receives graded outcomes based on actions taken. Through iterative processes, such as playing numerous games, the system learns to optimize rewards, eventually devising winning strategies. The objective is to learn the best policy for maximizing expected rewards over time. Reinforcement Learning finds applications in robotics, gaming, and navigation.

   

   An example of Reinforcement Learning in dog training

4. Semi-supervised Learning

   Semi-supervised Learning combines labeled and unlabeled data for training, suitable for applications like Supervised Learning but with reduced labeling costs. It utilizes a small amount of labeled data alongside a larger pool of unlabeled data, which is more cost-effective and easier to acquire. This approach can be applied to classification, regression, and prediction tasks. An example includes identifying a person's face on a webcam.