What are some deep learning use cases

Deep learning in practice: 5 use cases for deep learning algorithms

Deep learning is a data science method that is particularly important in connection with artificial intelligence. More and more everyday applications such as facial recognition software in smartphone cameras are using this method. In this blog article, we explain the high relevance of deep learning using five practical use cases.

What is deep learning?

Deep learning is onespecial class of optimization methods of artificial neural networks. That is why they are sometimes also referred to as "deep neural networks". The main difference lies in the complexity of the intermediate layers, the so-called "hidden layers".

Deep learning has become one of the key development drivers in the field of artificial intelligence in recent years for two reasons: First, because it achieves particularly good results when large amounts of data (big data) are available with which a network can be trained. Second, because the deep learning algorithms made it possible to represent intellectual and mental processes that were long assumed to be reserved for humans.

With a deep learning algorithm or a deep neural network, there are numerous intermediate layers ("hidden layers") between the input and output layers.

Two of the most prominent examples are Speech and face recognition. Siri, Cortana & Co., chatbots or the new Google image search are application examples that would not exist without DL. The algorithms of chatbots, for example, learn with every question they are asked and thus improve themselves. It is precisely this learning ability of deep learning algorithms that sets them apart from “normal” artificial neural networks.

Example to explain

In the movie"Her“Spike Jonze confronts his viewers with a form of artificial intelligence with which people can not only speak in a completely natural way, but with which one can even fall in love. Deep learning is a central key with which we can actually succeed in the future interact with digital personalities can.

Because anyone who tries to have a conversation with digital assistants like Siri or Alexa today will notice how quickly the limits of what is possible are reached here. Understanding and imitating human language is still one of the greatest challenges for computers. At the same time, the advances that are being made in this area are enormous.

How deep learning algorithms can understand language and images

0 and 1, yes or no - that's it basic binary operationon which all IT is based. In order to understand language or images, an extremely large number of gray levels, ambivalences and complex processes of understanding are necessary. Therefore, DL is a very promising path that could make it possible in the future that that is exactly what can be achieved.

One of the great strengths of these algorithms has been shown in recent years, for example, in the area of ​​image recognition and video analysis. The learning ability of this class of algorithms enabled them to continuously learn to understand the content of images. The following illustration shows in a simplified form how a deep learning algorithm “sees” something.

Example 1: Analysis of image data in the case of disease diagnoses

One of the most prominent areas of application of deep learning algorithms is in the field of image recognition. Impressive progress has been made in recent years, particularly in the field of medicine. These algorithms can be trained with different types of image data - this opens up very different areas of application and algorithms can be trained on them, Examine x-rays or CT scans for abnormalities.

So they can help doctors Support diagnosis of diseases. Because even if specialists have gained many years of experience, they can never view the same amount of image data that is used during training. Data sets used to train deep learning algorithms can contain many millions of images. So it's no wonder that intelligent analysis programs based on deep learning see better than humans.

Example 4: Sales and aftersales

In the area of ​​sales and aftersales, deep learning is used through language and sentiment analysis Improvement of the customer experience. But not only images in static or moving form can be used as a basis for training deep learning algorithms. Intelligent algorithms can also understand language in the form of texts or natural, spoken language better and better thanks to deep learning methods. For example, it is extremely difficult to understand irony - not just for machines or programs.

At first glance, ironic sentences do not differ from serious sentences. Sentences like “You did that great!” Or “Today you made yourself particularly stylish.” Can be meant both seriously and ironically, without the sentence itself changing. The context often provides the information that is crucial for the interpretation.

For service, for example, it is important to know whether an email from a customer is a normal request or whether the customer is clearly annoyed. Here it can happen that hundreds of e-mails from customers arrive in a day. A filter that does the Requests pre-sorted according to priority, helps to improve customer service tremendously.

With suitable training data, algorithms can be trained so that they can be used as intelligent filter system can be used. For example, they can identify those of angry customers from a mass of thousands of messages. The sentiment analysis, in which the feelings of the customers are measured, is used for customer care and can minimize the risk of customers dropping out.

Example 3: Improving security architectures

The more networked the world becomes, the more important the topic of cyber security and data security becomes. Deep learning can help Close security gaps in systems. Due to the ability to learn, the method is particularly suitable for distinguishing normal activities from attacks or other irregularities. This ability makes deep learning interesting, for example, when securing sensitive locations such as airports.

Real-time monitoring is all about monitoring the live video footage and identify suspicious eventsn. The longer a deep learning algorithm watches normal airport activity, the better it learns to distinguish which behaviors are unusual or noticeable.

Example 4: Minimizing risk in financial transactions

Deep learning sees things that humans cannot. But deep learning algorithms also have advantages in other sensitive areas that can be targeted by attacks. For example when monitoring bank transactions and securities trading. With anonymized training data, algorithms can be trained in such a way that they specifically recognize unusual activities that occur within a banking network. In addition to credit card fraud, attacks by malware and other malicious software can also be averted.

This makes one of the greatest challenges in the field of cyber security manageable: that Detection of first-time attacks by previously unknown malware or individual attackers. Even spam filters can be trained in this way to identify emails with malicious attachments. Because of the amount of data that needs to be checked, people cannot keep up with systems based on deep learning in this task.

Example 5: Industry 4.0: Tool for mastering big data

Deep learning is also used when it comes to analyzing big data and the issues are very complex at the same time. Due to its strengths, the data science method is an important tool for mastering big data. For example, when evaluating sensor data, such as is available in the case of maintenance data that accumulates in a wind farm. Here, measurements are sometimes carried out at various points every second, so that the amount of data quickly ends up in the petabyte range.

In such complex industrial ecosystems, the algorithm facilitates that Unit forecastthat need to be serviced. You can find out more about this in our free white paper on: Predictive Maintenance. Deep learning also makes it possible to establish complex relationships between industrial processes, data about interaction with customers and sales data.

The full effectiveness comes with time

Generally speaking, deep learning is a data science method that is used in particular when there are large amounts of unstructured data in which certain patternsbe recognized should. Above all, large amounts of audio data, video data or image data have been the focus of interest in recent years.

The intelligent solutions in this area open up areas that were reserved for humans for a long time. What sets deep learning apart from most other methods, and what makes them so intelligent, is the learning aspect.

Because this is not the case "Out-of-the-box solutions"that are programmed once and then ready to use. Rather, the algorithms inevitably need a certain training phase, during which they learn to perform their respective tasks. To do this, the algorithms sometimes make assumptions and check these assumptions in comparison with the test data. In this way, they learn from their mistakes or successes and get better and better over time.

For this reason, chatbots or digital assistants based on deep learning, for example, get better and better the more often they have received feedback from users as to whether their answer was helpful. The capabilities of Siri, Cortana & Co are still relatively limited today.

But if you just look at the successes that have been achieved through deep learning in recent years, it is only one question of timeuntil they'll talk to us naturally and maybe even have a sense of humor.

Michaela Tiedemann

Michaela Tiedemann has been part of the team at Alexander Thamm GmbH since the start-up days. She has actively shaped the development from a fast-paced, spontaneous startup to a successful company. With the establishment of her own family, a whole new chapter began for Michaela Tiedemann. Quitting the job was out of the question for the new mother. Instead, she devised a strategy for balancing her role as chief marketing officer with her role as a mother.