Demystifying Machine Learning: What is it and why is it important?

In today's data-driven world, machine learning (ML) has emerged as a powerful tool for businesses across industries, including finance. As a finance professional, understanding the basics of ML is essential to unlocking its potential for informed decision-making.

In this interview, Silvana Albert, PhD , Product Owner Big Data Technologies Analytics at swissQuant, explains in simple terms the fundamentals of ML, its different types, and why it's important - to help you navigate this transformative technology.

Machine Learning in a Nutshell

Silvana, can you briefly explain what machine learning is and introduce us to the basic types, including supervised and unsupervised learning?

Machine learning is a subfield of artificial intelligence (AI) that focuses on developing algorithms and models that allow computers to learn and make predictions or decisions based on data without explicit programming.

In essence, it's about training computers to get better at tasks by giving them more and more data.

There are three basic types of machine learning:

• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning

These three types of machine learning cover a wide range of applications in fields as diverse as healthcare, finance, robotics, and natural language processing. They have transformed the way we approach complex problems by allowing computers to learn and adapt from data, ultimately improving decision-making and automation across multiple domains.

Guided Learning for Predictive Insights

What is supervised learning and what problems can it best solve?

In supervised learning, the algorithm is trained on a labeled dataset. This means that each example in the training data is associated with a known target or output.

The primary goal is to learn a mapping from input data to the correct output so that the model can make accurate predictions or classifications on new, unseen data.

Supervised learning is suitable for problems with clear outcomes and goals, such as image classification and speech recognition. Imagine predicting whether an email is spam or not (classification - assigning categories), or estimating the price of a house based on features such as size and location (regression – predicting numerical values).

Examples of algorithms are: Linear Regression, Decision Trees, Support Vector Machines, Neural Networks.

Can you give a simple example of a real-world problem that supervised learning can address?

Supervised learning uses labeled data to train models that can make predictions, classifications, or decisions based on patterns and relationships learned from the historical data. Let me give you a real-world example:

• Handwriting recognition: Converts handwritten text or characters to digital text. It's used in applications such as optical character recognition (OCR) systems.

In handwriting recognition, supervised learning is the preferred approach because it learns from labeled data how to map handwritten words and their correct transcription to digital text.

This trained model can then convert handwritten notes from various sources into digital format, making them accessible, searchable, and editable for various applications, such as historical document preservation, text analysis, or automated data entry.

• Model selection: Choose an appropriate supervised learning algorithm for text recognition, such as Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs). CNNs are effective at capturing spatial patterns in images, while RNNs can handle sequences of characters or words.

Discovering Hidden Patterns

What is unsupervised learning? How is it fundamentally different from supervised learning?

In unsupervised learning, algorithms are trained on unlabeled input data without explicit targets or predefined outcomes. The primary goal is to find inherent patterns, structures, or relationships in the data.

Unsupervised learning can be used to explore data or detect anomalies. Common tasks include grouping similar data points (clustering) and reducing the number of features while preserving information (dimensionality reduction). Think grouping news articles into topics or reducing the complexity of data for visualization.

Essentially, unsupervised learning focuses on discovering hidden patterns or structures in data without predefined labels, while supervised learning aims to make predictions or classifications based on known target labels.

Can you give an example of a situation where unsupervised learning would be the preferred approach?

Here are two real-world examples for unsupervised learning:

1. Detecting Fraudulent Financial Transactions

Imagine you have a large dataset of customer transactions and your task is to detect unusual or fraudulent transactions. You don't have data explicitly labeled as "fraudulent" or "normal. Instead, you need to identify anomalies based on patterns in the data.

In this case, unsupervised learning techniques such as clustering or autoencoders can be used to identify anomalies. They learn what's typical and then flag anything that deviates from the norm.

Unsupervised learning is preferred in this scenario because it doesn't require predefined labels. You want the algorithm to learn the normal patterns and then identify deviations from those patterns as anomalies. This approach is valuable for detecting new, previously unseen types of fraud that may not follow predefined rules or labels.

1. Tumor subtyping in cancer research

In cancer research, scientists strive to understand the molecular and genetic heterogeneity of tumors. They aim to classify tumors into subtypes based on their genetic and molecular characteristics, as these subtypes may have different treatment responses and clinical outcomes.

Here, unsupervised learning is preferred because researchers lack predefined labels for tumor subtypes. Instead, they use unsupervised techniques to find natural groupings or subtypes within a dataset of tumor samples.

Key Differences & When to Use What

What are the key differences between supervised and unsupervised learning in terms of data requirements and model evaluation?

The distinction between supervised and unsupervised learning is important. This overview shows a comparison of the differences in data requirements and model evaluation for supervised vs. unsupervised learning:

Summary:

• In supervised learning, the model learns to make predictions or classifications based on known outcomes, making it suitable for tasks where you have labeled data and specific goals.
• In unsupervised learning, the model discovers inherent patterns or groupings in the data, which is valuable for tasks involving exploration, data understanding, and uncovering hidden structures, but it doesn't have predefined target labels for validation.

When facing a new problem, how do you decide whether to use a supervised or unsupervised learning approach?

The choice between supervised and unsupervised learning depends on the presence of labeled data and the specific goals of the problem or task.

• If there is labeled data with clear outcomes or targets, and the goal is prediction, classification, or regression, then supervised learning is typically the preferred approach.
• Conversely, if the data lacks labeled categories and the goal is to explore patterns, relationships, or clusters within the data, unsupervised learning is the more suitable choice.

It's essential to carefully assess the nature of the problem, the available data, and the desired outcomes in order to make an informed decision. In some cases, a hybrid or semi-supervised approach may offer the most effective solution.

Learning from Trial and Error

You mentioned three types of machine learning. What can you tell us about reinforcement learning?

Yes, apart from supervised and unsupervised learning, there's reinforcement learning.

Reinforcement learning is a type of machine learning where an agent interacts with an environment and learns how to make sequences of decisions that maximize a cumulative reward signal.

It's like teaching a computer program to learn through trial and error. The agent explores its environment, takes actions, receives feedback in the form of rewards or punishments, learns to make better decisions, and adjusts its behavior over time to achieve a specific goal or task and maximize long-term rewards.

Reinforcement learning is commonly used in applications such as gaming, robotics, and autonomous systems like self-driving cars, where the agent learns to make decisions to achieve goals by exploring different strategies and optimizing its actions based on past experience.

The Future of Machine Learning

The future of machine learning promises innovation and widespread integration into multiple aspects of our lives. Key trends include advancements in deep learning, reinforcement learning, and natural language processing, enabling more intelligent and context-aware AI systems.

Ethical considerations and responsible AI practices will be paramount as machine learning continues to transform industries such as healthcare, finance, autonomous systems, and personalized experiences. As it evolves, machine learning holds the potential to address complex global challenges and transform industries, ushering in a new era of innovation, automation, and data-driven decision-making.

The goal would be to transition to context-aware AI systems, and much depends on how we enforce ethical considerations and responsible AI practices.

We have collected so much data, and now it is time to make informed decisions from it.

Personally, I'm most excited about how it will further transform healthcare and personalized medicine because we still have so many challenges there, especially in the omics space.

Unveiling Tomorrow's Possibilities

As a finance professional, embracing machine learning opens doors to data-driven insights and informed decisions. Whether it's predicting market trends, optimizing investment portfolios, or detecting financial fraud, the applications of machine learning in finance are limitless. So dive into the world of machine learning and let it be your ally in navigating the ever-evolving financial landscape.

Sources:

Murphy, K. (2012). Machine Learning: A Probabilistic Perspective. The MIT Press Cambridge, Massachusetts, London, England.

Bishop, C. (2006). Pattern Recognition and Machine Learning. Springer New York.