What is 'Deep Learning' that helped scientists win the VinFuture Prize 2024?

Five scientists have received the $3 million VinFuture 2024 award for their research on "Deep Learning". So what is "Deep Learning" that helped scientists receive this prestigious award?

At the VinFuture 2024 award ceremony held on the evening of December 6 in Hanoi, the VinFuture Grand Prize worth 3 million USD was awarded to 5 scientists , including Professor Yoshua Bengio and Professor Geoffrey E. Hinton, from Canada; Mr. Jen-Hsun Huang, Professor Yann LeCun and Professor Fei-Fei Li, all from the US.

These scientists won the prestigious award for their groundbreaking contributions to advancing the advancement of Deep Learning .

Scientists receive the VinFuture Grand Prize 2024 for groundbreaking contributions advancing the advancement of Deep Learning (Photo: Manh Quan).

So what is "Deep Learning" that helped scientists receive the VinFuture Grand Prize 2024?

In fact, "Deep Learning" is not a new concept and has been mentioned a lot recently, especially when the race to develop artificial intelligence (AI) is becoming more exciting than ever.

However, not everyone clearly understands the concept and practical applications of this technology.

What is "deep learning"?

Deep Learning is a branch of Machine Learning and Artificial Intelligence (AI). Deep Learning focuses on teaching computers to learn and improve their ability to perform tasks through an artificial neural network that simulates the way the human brain works.

What sets deep learning apart from traditional machine learning methods is its ability to automatically extract information from data without manual programming intervention.

Deep learning enables computer systems to make decisions based on learned data (Illustration: Pinterest).

In a simple way, you can think of deep learning as teaching a baby to recognize the world around them. For example, you teach a child to recognize a cat.

Initially, when a child sees a picture of a cat, the child's brain will gradually learn the characteristics of a cat such as pointed ears, whiskers, long tail, four legs. Every time a child sees a new cat, the child will automatically recognize "This is a cat" based on the learned characteristics.

Deep learning works in a similar way. It's a method of teaching computers to 'learn' from lots of examples, just like the human brain. The computer automatically figures out important features from the data (like a cat's ears, whiskers, or tail) through multiple layers of processing (hence the name 'deep learning'), and uses those features to recognize new things.

An example of a deep learning system that records the features of a cat so that it can recognize a cat if it encounters one (Illustration: AI).

For example, if you give a deep learning model 1,000 pictures of cats, it will learn about the characteristics of these animals. Then, if you give it a new picture that the deep learning system has never seen before, it can recognize "This is a cat!" based on what it has learned, just like a child would.

The main difference between deep learning and previous methods is that instead of humans having to specify each specific feature (such as entering the command "find pointed ears", "find whiskers", etc.) , the computer will freely explore and filter out which features are important. This helps it handle complex tasks that are difficult for humans to describe with specific rules.

'Deep learning' is like the learning process of a student, the more illustrations the student sees, the better he or she will learn, figuring out how to solve exercises without the teacher having to show him or her step by step. That is why AI systems often require a lot of data to be able to learn by themselves.

History of the Development of "Deep Learning"

The origins of deep learning date back to the 1940s, when two American scientists, Warren McCulloch and Walter Pitts, built the first mathematical model of neurons.

Professor Geoffrey Everest Hinton is considered one of the "fathers of AI" (Illustration: LinkedIn).

However, it was not until the 1980s, with the advent of the "backpropagation" algorithm invented by Canadian computer scientist Geoffrey Everest Hinton, that multi-layered mathematical neural networks truly became feasible and effective. Geoffrey Everest Hinton is also one of the five scientists who have just received the VinFuture 2024 special award.

The real explosion of deep learning began in the early 2010s, driven by three main factors: the dramatic increase in computing power of computer hardware, the massive amount of data available, and major improvements in neural network architecture.

Practical applications of "Deep Learning"

Today, deep learning has been practically applied in many areas of life.

In natural language processing , AI-integrated chatbots such as ChatGPT, Gemini, Claude AI… have made great progress in translation, text summarization, question answering, and even art creation. These AI chatbots are becoming smarter by understanding context and interacting naturally with humans.

In the field of computer vision , deep learning has revolutionized applications such as facial recognition, object detection, and more. Intelligent surveillance systems can detect unusual behavior, while photo editing applications can automatically beautify or create entirely new images.

Deep learning helps doctors diagnose diseases based on clinical signs more accurately (Illustration: Getty).

In medicine, deep learning is helping doctors diagnose diseases earlier and more accurately through the analysis of X-ray, CT, MRI images… Deep learning models also aid in the development of new drugs and prediction of protein structures.

In industrial manufacturing, deep learning is used in automated quality control and production process optimization. AI-powered robots can perform complex tasks that require high flexibility and adaptability thanks to deep learning.

Another prominent application of deep learning is in the field of speech recognition. This technology has transformed the way people interact with machines, from voice-based virtual assistants to tools that help people with disabilities.

Additionally, deep learning is also used to analyze big data in areas like finance, fraud detection, and e-commerce, where it recommends products based on user behavior.

However, the rapid development of deep learning also poses many challenges. Among them are ethical and privacy issues, when personal data is used to train models. In addition, deep learning is energy-consuming and uses a lot of computational resources, raising concerns about sustainability.

More importantly, understanding and controlling complex deep learning models remains a major challenge, especially when they can make decisions on their own that are difficult for humans to explain.

In the future, deep learning promises to continue to develop, opening up many new opportunities in fields such as medicine, energy, education. However, to maximize the potential of this technology, there needs to be close cooperation between researchers, governments and businesses to build standards of ethics as well as social responsibility.

Deep learning is not just a technological tool, but also a driving force for human progress if properly directed and managed.