What is deep learning?

Deep learning is a subset of machine learning, where computers process large datasets using neural networks modelled after the human brain.

In deep learning, the focus is primarily on the autonom­ous learning process of these neural networks. They consist of an input layer, one or more hidden layers, and an output layer. In­form­a­tion enters the input layer as a vector, is weighted through ar­ti­fi­cial neurons in the hidden layers, and finally produces a specific pattern in the output layer. The more layers a neural network has, the more complex the tasks it can handle, enabling ar­ti­fi­cial in­tel­li­gence to tackle intricate problems.

How does deep learning work?

Sorting images according to whether dogs, cats or people can be seen in them is a chal­len­ging task for a computer. Something that is im­me­di­ately clear at a glance to humans requires a computer to analyse in­di­vidu­al image char­ac­ter­ist­ics.

With deep learning the raw data input, in this case the image, is analysed layer by layer. In the first layer of an ar­ti­fi­cial neural network, for example, the system examines the colours in the in­di­vidu­al image pixels. Each image pixel is processed with its own neuron. In the following layer, edges and shapes are iden­ti­fied and, in the layer after that, more complex char­ac­ter­ist­ics are examined.

The in­form­a­tion collected is displayed in a flexible algorithm. The results from one layer are carried onward into the following layer and change the algorithm. In this way, the computer is able to use a variety of op­er­a­tions to come to the con­clu­sion of whether an image can be cat­egor­ised as a dog, cat or human.

At the start is the training period, errors in cat­egor­isa­tion are corrected by humans, allowing the algorithm to adapt. After a short time, it can improve its image re­cog­ni­tion in­de­pend­ently. As the in­ter­link­ing between the neurons in the network changes and the weighting of variables within the algorithm is adapted, certain input patterns (different kinds of cat pictures) lead more and more ac­cur­ately to the same output patterns (the cat being re­cog­nised). The more image material is available to the system for it to learn from, the better.

With deep learning, it isn’t always possible for humans to un­der­stand which patterns the computer re­cog­nised in order to reach its con­clu­sions, par­tic­u­larly since the system con­tinu­ously optimises its own decision-making rules.

History of deep learning

Deep learning is actually quite a recent term – it was first used in 2000 – yet the method of using ar­ti­fi­cial neural networks to enable computers to make in­tel­li­gent decisions is several decades old.

Basic research in the field goes all the way back to the 1940s. Ar­ti­fi­cial neural networks were first developed in the 1980s. Back then, though, the quality of the decisions was dis­ap­point­ing, because machines’ in­de­pend­ent learning – deep learning – requires large quant­it­ies of data, which at the time just weren’t available digitally. Only around the turn of the mil­len­ni­um did the age of big data begin, making deep learning in­ter­est­ing again for science and business.

Strengths and weak­nesses

Compared with earlier AI tech­no­lo­gies, deep learning is sig­ni­fic­antly more effective. Before the tech­no­logy can reach its full potential, though, some weak­nesses still have to be overcome.

Strengths of deep learning

One of the most important arguments is the quality of its results. In image re­cog­ni­tion and speech pro­cessing, in par­tic­u­lar, the tech­no­logy is clearly superior to all others. Provided with high-quality training data, deep learning can carry out routine work much more ef­fi­ciently and much faster than any human – without any signs of fatigue either, and with no change in quality.

With other forms of machine learning, de­velopers analyse the raw data and peri­od­ic­ally define ad­di­tion­al features that the algorithm is to take into account while learning in order to improve the AI’s fore­cast­ing power. With deep learning, the system itself re­cog­nises useful variables and in­cor­por­ates these into its learning process. After the initial training period it can learn without any human guidance, saving both time and money since skilled employees aren’t necessary for future de­vel­op­ment.

Up to now, large quant­it­ies of data had to be labelled manually in order to make machine learning possible. In image re­cog­ni­tion, for example, employees were required that would assign the label dog or cat to the images. With deep learning, the manual training period is sig­ni­fic­antly shorter. Above all this is relevant because, while general corporate practice certainly does involve col­lect­ing large quant­it­ies of data, only in rare cases does it exist in the form of struc­tured data (telephone numbers, address, credit cards, etc.). In most cases it is stored as un­struc­tured data (images, documents, emails, etc.). Unlike al­tern­at­ive methods of machine learning, deep learning can evaluate different sources of un­struc­tured data while con­sid­er­ing the task at hand.

The argument that the tech­no­logy is too costly in practice for it to be ap­plic­able on a large scale is losing traction. Services like Google Vision or IBM Watson are in­creas­ingly emerging, allowing companies to build on existing neural networks instead of having to develop them from scratch. With this, in the future deep learning will be more and more capable of playing on its strengths in corporate practice.

An overview of the strengths

• Better results than with other methods of machine learning
• No feature de­vel­op­ment and no data labelling necessary
• Efficient execution of routine tasks without affecting quality
• Problem-free handling of un­struc­tured data
• More and more services to make it easier to use ar­ti­fi­cial neural networks

Weak­nesses of deep learning

Deep learning requires an enormous amount of pro­cessing power. This largely depends on the com­plex­ity and dif­fi­culty of the task to be ac­com­plished and the size of the data set used. Up to now, that made the tech­no­logy expensive and only prac­tic­able for research and a handful of mega-cor­por­a­tions.

There has indeed been ob­serv­able progress in this respect. What won’t change in the fore­see­able future, though, is the fact that decisions made by deep learning are no longer trans­par­ent to humans. The neural network is (so far) a black box. For some ap­plic­a­tions where trans­par­ency is decisive, this makes the tech­no­logy ir­rel­ev­ant.

For deep learning to work at all, large sets of training data are required. If these quant­it­ies of data aren’t available, computers aren’t yet able to deliver reliable results with the help of deep learning. The first libraries of neural networks are indeed being published, making the ap­plic­a­tion of deep learning easier for the general public. However, the services are not suitable for every ap­plic­a­tion, meaning that the de­vel­op­ment of learning al­gorithms for deep learning still demands a lot of time in­vest­ment, and po­ten­tially takes more time than using al­tern­at­ive methods.

An overview of the weak­nesses

• Requires high pro­cessing power
• De­vel­op­ing learning al­gorithms is re­l­at­ively time-consuming
• A large data pool is necessary
• More training data needed than with other methods of machine learning
• Decisions difficult or im­possible to un­der­stand (black box)

Ap­plic­a­tion areas for deep learning

Deep learning is already being im­ple­men­ted in various sectors, and in the future we will come across it in many more areas of our day-to-day lives.

• User ex­per­i­ence: Some chatbots are already optimised using deep learning and leverage Natural Language Pro­cessing to respond better to customer inquiries, easing the workload on human customer support teams.
• Voice as­sist­ants: As mentioned, deep learning is used in various voice as­sist­ants like Alexa, Google Assistant, or Siri through speech synthesis. These systems autonom­ously expand their vocab­u­lary and improve their language com­pre­hen­sion.
• Trans­la­tions: Deep-learning-powered trans­lat­ors, such as DeepL, produce high-quality trans­la­tions. Thanks to this tech­no­logy, dialects and text from images can be auto­mat­ic­ally trans­lated into other languages.
• Content creation: LLMs like ChatGPT use deep learning to generate text that is not only gram­mat­ic­ally correct but can also mimic an author’s style—provided they have suf­fi­cient training material. Early ex­per­i­ments have seen AI systems create Wikipedia articles and re­mark­ably authentic Shakespearean texts using deep learning.
• Cy­ber­se­cur­ity: Deep learning-powered AI systems are par­tic­u­larly suited for detecting ir­reg­u­lar­it­ies in system activity, helping to identify potential hacker attacks.
• Finance: The ability to detect anomalies is es­pe­cially useful in financial trans­ac­tions. Properly trained al­gorithms can help prevent attacks on banking networks and credit card fraud more ef­fect­ively than tra­di­tion­al methods.
• Marketing and sales: AI systems can use deep learning to perform sentiment analysis and autonom­ously implement defined actions to restore customer sat­is­fac­tion.
• Autonom­ous driving: While fully autonom­ous vehicles remain a vision for the future, the tech­no­logy already exists. It combines various deep learning al­gorithms: one to recognise traffic signs, another to detect ped­es­tri­ans, and so on.
• In­dus­tri­al robots: Robots equipped with deep learning AI could be deployed across numerous in­dus­tri­al sectors. By simply observing a human operator, these systems could learn how to operate machines and optimise their own per­form­ance.
• Main­ten­ance: Deep Learning offers sig­ni­fic­ant potential in in­dus­tri­al main­ten­ance, where complex systems require con­tinu­ous mon­it­or­ing of numerous para­met­ers. Ad­di­tion­ally, it can predict which com­pon­ents of a system are likely to require servicing soon (Pre­dict­ive Main­ten­ance).
• Medicine: Deep learning AI systems can scan images for anomalies far more ac­cur­ately than even a trained human eye. As a result, diseases can be detected earlier than ever on CT or X-ray images using these in­tel­li­gent systems.

Deep learning has great potential but isn’t a universal solution

In public discourse to some extent there is the im­pres­sion that deep learning is the only tech­no­logy of the future for AI. It’s true that, in many ap­plic­a­tion areas, deep learning makes much better results possible than previous pro­ced­ures did.

However, deep learning is not the best tech­no­lo­gic­al solution for every problem. There are other strategies to make computers ‘in­tel­li­gent’ – solutions that can also work with small datasets and where the decision-making is trans­par­ent for humans.

Some AI re­search­ers view deep learning as a trans­ition­al phe­nomen­on and believe that better ap­proaches, not based on the human brain, will emerge. Google’s company strategy proves that these critical voices are not to be ignored: There, deep learning is just one part of the AI strategy. Alongside it are also further methods of machine learning, like the de­vel­op­ment of quantum computers.