Machine learning 101

What is machine learning? 

Machine learning algorithms make decisions or predictions based on data. Algorithms find patterns in data and can be trained to improve over time as more data is provided without needing to be explicitly programmed for every task. Better quality data leads to better machine learning algorithms. 

 Various metrics are important to measure machine learning success. Some ML systems focus on prediction, where success is measured by prediction accuracy, and some focus on decision-making, where success is measured by achieving optimal outcomes. One approach to optimal decision making is reinforcement learning, where an algorithm learns through trial and error to maximize rewards. For example, if a robot was in a maze looking for the way out, reinforcement learning or other algorithms will find the right path for the robot. 

Is machine learning the same as AI? 

Machine learning is a subset of artificial intelligence (AI). AI refers to the aim to make systems that mimic human intelligence, with the ability to carry out tasks like understanding natural language, recognizing objects, or playing games. Machine learning is a subset of AI focused on developing models that learn from data. 

 Machine learning is often contrasted with rule-based learning, with ‘if this occurs then [do] that’ instructions. In many scenarios, machine learning is a superior system, as rules can never be fully exhaustive. In a medical scan, for example, you cannot say if pixel one is this color and pixel three is this color, then it is a tumor. It is not how we work as humans, and it is not how effective AI agents work. Instead, in this case, machine learning agents would be trained with data, shown thousands or maybe millions of scans with corresponding labels, and would learn the right patterns to spot the tumor. 

Machine learning and rule-based systems often complement each other, with each being suited to different scenarios. But for now, rule-based systems are still commonly used in robotics, or areas where there isn’t sufficient good quality data available yet, although even in areas like self-driving cars, machine learning plays a central role in perception and decision-making. 

What are the differences between artificial intelligence and machine learning? 

AI: A broad field that includes any technique enabling computers to mimic human intelligence. AI encompasses a wide range of applications like robotics, natural language processing, and computer vision. 

Machine learning: A specific subset of AI that focuses on building systems that learn from data to improve their performance on a specific task. Machine learning is one of the ways AI systems can learn, but it’s not the only way—AI can include rule-based systems, expert systems and more. 

What are the different ways to master machine learning? 

 To master machine learning, you could follow these steps: 

1. Learn the fundamentals: Start with understanding linear algebra, calculus, statistics, and probability, which are the backbone of many machine learning algorithms. Take beginner classes and subscribe to relevant content on YouTube and Coursera.
   
2. Understand the core concepts: Grasp essential machine learning concepts such as supervised learning, unsupervised learning, reinforcement learning and model evaluation.
   
3. Learn how to code: Python is the most common language for machine learning, with libraries like PyTorch, Numpy, Scikit-learn.
   
4. Practice with real-world projects: You can learn the basics by listening to lectures. But if you don’t try to program your own machine learning algorithm then learning will be slower. Apply your knowledge to real-world datasets, e.g., from Kaggle.
   
5. Stay updated: Machine learning is a fast-evolving field, so it’s important to follow the latest research and developments from conferences like NeurIPS or ICML. What are the current trends? What are the open questions? Where do current machine learning systems fail, and can we tackle that? A great way to stay at the cutting-edge by studying machine learning at MBZUAI. 

What are the differences between machine learning and deep learning? 

Deep learning is a subset of machine learning. Machine learning covers a broad range of algorithms, whereas deep learning specifically uses multi-layered neural networks to learn patterns, often from large datasets. Deep learning takes inspiration from how human brains process information via interconnected neurons.  

Before deep learning, machine learning encountered a problem in how to create the right features from unstructured data, like images, text and sound. Initially, it took cumbersome manual work to list features. For example, taking an image and saying that “there is an edge in this part, there is a circle in this area…”. The goal of deep learning was to bypass the need for this feature engineering, to take learnings from raw data input much like the human brain does. 

How do I become a machine learning researcher? 

Formal education is the place to start– like an M.Sc or Ph.D. at MBZUAI. Engage in research by reading papers (follow top conferences like NeurIPS, ICML, and ICLR). Contact researchers to try and get involved in research projects. Experiment with different algorithms and contribute to open-source machine learning frameworks. Join a research group, network with other researchers, publish papers and collaborate on projects to broaden your experience and visibility. 

What are some real-world machine learning applications? 

• Healthcare: Machine learning models can help doctors make better decisions, assist in diagnosing diseases, discover new drugs, predicting patient outcomes, personalize treatments, develop robotics surgeons to ease shortages in some areas of the world.
• E-commerce: Machine learning can power recommendation engines that suggest products based on customer behavior.
• Autonomous vehicles: Self-driving cars rely on machine learning to interpret sensory data and make driving decisions.
• Natural language processing: Applications like chatbots, virtual assistants, and machine translation use machine learning models to process and generate human language.
• Image and video recognition: Facial recognition, object detection, and video analysis for security and entertainment purposes.
• Energy: Predictive maintenance in power plants, optimizing energy consumption, and managing renewable energy sources rely on machine learning algorithms.