Machine learning – what is it and why does it matter?

Why is machine learning important?

Machine learning enables computers to automatically learn and improve from experience as opposed to being specifically programmed. As a result, it’s possible to create highly accurate and efficient models that can solve complex problems and make predictions based on new and unseen data. 

Advances in computing power, data storage and algorithm development have made machine learning more accessible and scalable, making it an essential tool for a wide range of applications across fields such as finance, healthcare and transportation. 

Where machine learning is used today 

You are likely interacting with some form or another of machine learning tech in your daily life, perhaps without even realising it. Applications currently using the tech include:

• Image recognition in social media apps
• Personalised recommendations in e-commerce websites
• Fraud detection in financial transactions
• Speech recognition in virtual assistants, such as Siri and Alexa
• Medical diagnosis and treatment recommendation systems
• Self-driving cars and autonomous vehicles
• Predictive maintenance in industrial settings
• Natural language processing for translation and sentiment analysis
• Credit scoring and risk assessment in lending and insurance 
• Energy load forecasting for efficient energy management.

What are the main types of machine learning?

There are three main varieties of machine learning: supervised, unsupervised and reinforcement learning. Let’s take a look at each.

• Supervised learning

With this approach, the algorithm is trained on a labelled dataset, where we already know the correct output. The algorithm then learns to map input variables to the correct output and can then make predictions on new, unseen data based on this learning. Supervised learning examples include image classification, speech recognition and regression analysis. 

• Unsupervised learning

Unlabelled datasets are used in unsupervised learning, so the algorithm needs to work to find patterns or structure within the data on its own. Tasks such as anomaly detection, clustering and dimensionality reduction are examples of this method.

• Reinforcement learning

With reinforcement, an agent is used to interact with an environment to learn how to make decisions that maximise a reward. The agent then receives feedback in the form of a reward signal and then uses this feedback to adjust its behaviour over time. This kind of learning is often used in game playing, robotics and other dynamic decision making tasks. 

What’s required to create quality machine learning systems?

Several key elements are required to create good learning systems. We’ve covered some of the top contenders to get you started. 

High-quality data

It’s crucial that the data used to train a machine learning model is of decent quality. Data ideally should be clean, representative and relevant to the problem of focus.

Appropriate algorithms

Choosing the right learning algorithm for a given problem is important. Each algorithm has its own strengths and weaknesses and the accurate choice will depend on the nature of the data and the problem being solved. 

Feature engineering

The process of selecting, transforming and scaling the input variables to a machine learning model is feature engineering. Quality feature engineering can greatly improve the accuracy of a model. 

Robust evaluation

It’s important to evaluate a machine learning model on a separate, unbiased dataset to ensure that it is not overfitting to the training data.

Iterative development

The development of good machine learning systems very often involves an iterative process of experimentation and refinement, where different algorithms, parameters and features are tested and improved over time. 

Domain knowledge

A deep understanding of the problem domain and the data is often critical to developing effective machine learning systems. This can help guide feature engineering algorithm selection and evaluation.

Adequate computing resources

Machine learning often involves processing large amounts of data, and may require substantial computing resources such as powerful GPUs and cloud-based computing services.

What is bootstrapping machine learning?

A technique of resampling data to improve the stability and accuracy of a model is known as bootstrapping. It involves randomly sampling the data with replacements to create multiple data sets, training a model on each dataset, and aggregating the results to get a more robust estimate of model performance. 

What are some popular machine learning methods?

There are a wide range of learning methods to choose from, with the best method being dependent upon the specific problem being solved and the nature of the data. Here are some of the most popular:

• Linear Regression
• Logistic Regression
• Decision Trees
• Random Forests
• Naive Bayes
• Support Vector Machines (SVMs)
• K-Nearest Neighbours (KNN)
• Neural Networks (including deep learning)
• Gradient Boosting 
• Clustering (such as K-Means or Hierarchical clustering)

What is AlphaGo?

A computer program developed by a British artificial intelligence company, DeepMind, which was acquired by Google in 2015, AlphaGo is a deep learning neural network. It uses a combination of supervised learning and reinforcement learning to play the board game ‘Go’. In 2016, AlphaGo defeated Lee Sedol, one of the world’s top-ranked Go players, in a five-game match, making the first time an AI program had defeated a human professional Go player. This was considered a significant milestone in the field of AI and demonstrated the power of machine learning for complex decision making tasks.