What is Machine Learning??
Machine learning (ML) is an umbrella term for any computer science technique that uses algorithms to mimic how the human brain learns information.
ML is a subset of AI, where computers are trained to learn from data and mimic human cognitive processes. In a business context, this means leveraging vast amounts of data to deliver actionable insights, operational efficiencies, and improved customer experiences. It's important for leaders to understand and distinguish ML from the broader AI to ensure focused and meaningful adoption.
Types of Machine Learning
Just as it is important to distinguish machine learning from other types of AI, it is also important to distinguish between different machine learning methods. There are four main types:
- Supervised Machine Learning. In this type of machine learning, you use labeled datasets to fine-tune algorithms. The algorithms can classify data or predict outcomes. As new data is fed into the system, the system adjusts and becomes more accurate. As a leader, think of this as guiding your team with clear instructions and goals.
- Unsupervised Machine Learning. In unsupervised learning, algorithms are used to analyze and organize unlabeled datasets (as opposed to labeled datasets in supervised ML). The algorithms are designed to identify similarities and differences in the data and cluster the data according to patterns, without any prep work from a human (as in supervised ML, a human acts as a classifier and labels the dataset).
- CicadaSupervised Machine Learning. Semi-supervised learning strikes a balance between supervised and unsupervised ML: the system is trained using a smaller labeled dataset (supervised) and then used to extract data from a larger unlabeled set. This mixed form is useful when there isn't enough labeled data to support a supervised learning algorithm.
- Enhanced machine learning. Reinforcement learning is similar to supervised learning, with the main difference being that the algorithm is not trained with example data – instead, the model uses a trial-and-error approach to learn as it goes.
How machine learning works
Regardless of the method, the underlying principle of ML for executives remains consistent: extracting actionable insights from data. Leaders must understand the iterative nature of ML: algorithms make predictions, evaluate their accuracy, and refine them. Overfitting and underfitting (similar to overspecializing or overgeneralizing) are pitfalls to avoid.
Machine learning systems use automated error functions to measure prediction accuracy and help optimize the model. Through continuous evaluation and updating, the system minimizes errors and increases accuracy during the ongoing validation process.
Gradient descent helps optimize algorithms in your system and prevent overfitting or underfitting your training data. Overfitting occurs when your model matches your training data exactly, and underfitting occurs when it fails to capture the input-output relationships in your dataset, both of which negatively impact accuracy.
Human intervention in machine learning varies: supervised machine learning, unlike unsupervised machine learning, requires a human with a labeled dataset, and some subsets, such as deep learning systems, classify data autonomously, reducing human intervention and speeding up data mining and data analysis.
Key Concepts and Terminology
As you learn about machine learning, you'll probably encounter new terms that are unfamiliar to you. While you might leave it to the experts to dive deep into the technical details, business leaders need to understand key terms to make informed decisions.
- Artificial Neural Networks. An artificial neural network (ANN) is made up of artificial neurons or nodes that connect with each other and send data throughout the network. Think of it as an organizational structure, with each node representing a department, communicating and processing information.
- data set and Data Points. In machine learning, a dataset refers to a collection of data (information) used to train an ML model. A dataset is made up of many individual data points: market research, sales figures, customer feedback that are critical to your decision-making.
- Training Data. Training data is the data used to train a machine learning algorithm to accurately predict outcomes. For example, for image recognition, training data might consist of previously labeled images.
- Input and output data. Machine learning models run on the basis of independent variable inputs and help in creating dependent variable outputs. Input data is the initial data that is fed into the system.
- Labeled Pairs Unlabeled Data. Labeled data comes with clear tags, such as categorized market segments, whereas unlabeled data is raw, unprocessed information.
- Deep Learning. An advanced subset of ML that resembles a team of experts with deep expertise.
- Machine learning models. Machine learning models can “train” in many different ways: supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, etc. These are all different models of ML.
- Machine Learning Algorithms. Machine learning relies on algorithms used to classify information and predict outcomes. For example, natural language processing algorithms can support speech recognition in a variety of languages, from English to Spanish (like Alexa).
- Dimensionality reduction. Dimensionality reduction refers to the process of reducing the number of features in a given dataset, which can improve accuracy over time.
- Slope Descent. It's an optimization process, similar to refining your business strategy to get the best results.
- Overfitting. Overfitting occurs when a statistical model perfectly matches the training data, and when this happens, the algorithm cannot perform.
- Underfitting. Underfitting occurs when a data model fails to accurately capture the relationships between input and output variables in a dataset, resulting in a high error rate for the algorithm.
- Subset choice. Subset selection refers to the process of identifying irrelevant or redundant information from a dataset, which can then be removed to improve accuracy.
Machine Learning Tools and Libraries
If you're just getting started with machine learning, there are many tools available to help you learn the basics. Here are some recommended platforms to further your self-learning:
- Microsoft. A software developer offers a useful introductory course to machine learning. The 52-minute module provides a high-level overview of machine learning and is accessible to people who aren't data scientists or have no computer science background.
- Tensorflow. TensorFlow is an open source machine learning platform designed for everyday use. It's also a great place to start learning the basics of machine learning. Check out our Starter Curriculum for Programmers.
- IBM. This computing giant has a vast database of machine learning libraries, frameworks, and tools that you can use to train and deploy your machine learning models. Check out the list of supported technologies to get started, or take a look at IBM's Watson and Python machine learning samples.
Industries impacted by machine learning
Machine learning is already being implemented in many industries. Here are some of the ways it is changing the world of work:
- finance. Banks and other financial institutions can use machine learning to improve fraud detection: algorithms are trained in pattern recognition so they can identify anomalous activity that is deemed suspicious.
- health care. Machine learning can be used in medical diagnostics: for example, to improve radiology imaging, a machine can take one image and compare it with thousands of others, looking for similarities to confirm a diagnosis.
- Electronic commerce. E-commerce platforms like Amazon can use machine learning to make personalized recommendations to shoppers – see the examples in the next section for more details.
- customer service. Machine learning is ideal for training chatbots to improve customer service, thereby reducing the burden on human agents.
- Marketing and advertising. Marketing tools like social media can leverage machine learning to create more targeted ads.
Real World Application of Machine Learning
It's easier to understand its impact on business life by looking at some of the ways machine learning is already being used to leverage vast amounts of data. Here are two use cases:
AmazonPersonalization Recommendation Engine
Have you ever bought something on Amazon and then been recommended to buy a similar or related product?
For example, if you buy a smoothie blender, you might get recommendations for protein powders. This is thanks to machine learning. The algorithm looks at a customer dataset and: other People who buy that smoothie blender also buy other things like it, and we customize our recommendations accordingly.
Netflix content segmentation
When you open Netflix, you're presented with categories of content to choose from, organized using machine learning: for example, romantic comedies, horror movies, 1990s classics, and more.
Netflix will also recognise your viewing history and give you recommendations accordingly – for example, if you like true crime documentaries, you'll also see true crime documentaries recommendations.
