Story by Boat Charunthon Limseelo, Advised By Professor Aye Hninn Khine

Hi everyone, welcome back to MLOps The Series.

In last episode, I ensured that you guys might study about some simple concepts of Machine Learning Operations before coming to this episode. For this week, Aj. Santhitam is the lecturer for guiding us for learning about the algorithm within machine learning types, which consists of supervised, unsupervised, and semi-supervised training. But, we would like to know what would be their example for each type and how would it work. In this blog episode, we will discuss the example from three of them.

Diving into the World of Machine Learning

Machine learning (ML) is rapidly changing how we interact with technology, and it’s becoming increasingly important to understand its core principles. But what exactly is machine learning? Simply put, it’s a method of data analysis that automates the building of analytical models. Instead of explicit programming, ML uses algorithms that learn from data to uncover hidden insights. This iterative learning process is what allows computers to make predictions and decisions without being directly told how to do so.

Why Does Machine Learning Matter?

The impact of machine learning stems from its unique capabilities:

• Automation: Once trained, ML models can run automatically without further human intervention. This can save time and resources, making it a powerful tool in many applications.
• Speed: ML algorithms can analyze large datasets much faster than manual methods. This speed is particularly beneficial when dealing with big data.
• Accuracy: ML models can often make predictions with greater accuracy than traditional manual methods. This accuracy can be crucial in fields where precise insights are needed.
• Scalability: ML is desinged to handle large data, making it suitable for complex problems that require processing massive amounts of information.

The Machine Learning Workflow

Developing a machine learning model involves several key steps:

1. Data Acquisition: Gathering the necessary data is the first step.
2. Data Preparation: This is a crucial stage that involves cleansing, shaping, and enriching the data. Data annotation (establishing ground truth) is also performed at this stage.
3. Model Training: The prepared data is used to train the ML model. This process is interative and uses training, test, and validation sets.
4. Model Testing: Once trained, the model’s performance is evaluated and optimized.
5. Deployment: Models are then deployed to production and continuously delivered. Out-of-band (OOB) testing, also known as shadow deployment, might be used at this stage.

Types of Machine Learning

In normal, we do have 4 types of Machine Learning that we have known: Supervised, Semi-supervised, unsupervised, and reinforcement learning (RL). For this class, we are going in deep for Supervised and Unsupervised Learning.

Supervised Learning: This involves training models with labeled data, where the desired output is known. Examples include:

• Classification Models: Used for categorizing data into predefined classes.
• Regression Models: Used for predicting continuous values.

Unsupervised Learning: This involves training models with unlabeled data, where the goal is to discover hidden patterns or structures. Example include:

• Clustering: Grouping similar data points into clusters.
• Anomaly Detection: Identifying unusual or rare data points that deviate from the norm.

Challenges in Machine Learning

While machine learning offers many benefits, it’s important to be aware of some challenges that are often encountered:

• Overfitting: When a model learns the training data too well, resulting in poor performance on new, unseen data.
• Specialized vs. Generalized Learning: A model can become too focused on specific training data patterns and fail to generalize well.
• Unbalanced Data: When data is biased and one category is overrepresented.
• Noisy Data: Dealing with uncleaned or inaccurate data can affect model performance.
• Noisy Data: Dealing with uncleaned or inaccurate data can affect model performance.

The Foundaions of Machine Learning

ML drawns from a diverse set of fields:

• Mathematics & Statistics: Concepts such as probability, entropy, regression, Bayes theorem, distributions and normalization are essential.
• Computer Science: Algorithm design, data structures, optimization, planning, and programming are all essential.
• Artificial Intelligence: Reinforcement, computer vision, natural language processing, and neural models are used in different aspects of ML.

Practical Example

The sources offer some real-world examples of machine learning:

• Consumer ML includes applications like face detection, gender prediction, age estimation, face recognition, and expression recognition.
• Smart Home applications use ML for voice interfaces.
• Product Offering ML can be used to determine the next best product offering for custormers.
• Loan Customer Lifetime Value Prediction: ML can also be used to predict the lifetime value of loan customers.
• Predicting Contract Values: ML can also be used to predict the length of contracts, interest calculations, and overall value of a customer by looking at data like demographics, previous contracts and payments.

Key Takeaways

Machine learning is a rapidly evolving field, but understanding these basic concepts and processes can help you appreciate its power and potential.