Machine Learning describes the ability of certain algorithms to autonomously learn from data using pattern recognition
The amount of data is constantly increasing. Authorities, corporations and even the sports club next door collect, record and process data today. But what can be done with these enormous amounts of data? And how can companies use the data commercially? The answer to these questions is: with machine learning.
Machine Learning is a sub-area of Artificial Intelligence (AI). Machine learning-based systems learn autonomously without having been specifically programmed for their later task.
Machine Learning has experienced an upswing in recent years.
On the one hand, this was due to the fact that computer capacities have increased enormously: Computers process large amounts of data ever faster. On the other hand, the knowledge of interrelationships has increased. Scientific progress has been immense.
The technology is now being used in many areas of everyday life, such as search engines and speech recognition. And even manufacturing companies are already using the intelligent machines. Machine learning helps in robotics or image recognition.
We are currently on the verge of the technology's introduction even in areas far removed from production. This applies, for example, to the area of finance and accounting. The first products - such as the plug-and-play assistants for AI in SAP from Nooxit - are ready for the market and are now being used.
An expert in machine learning is Ethem Alpaydín, professor at the Faculty of Computer Engineering at Özyegin University in Istanbul. He defines machine learning as follows: "Machine learning means programming computers in such a way that a performance criterion is optimized on the basis of an example or previous experience".
Machine learning thus describes the ability of machine systems to learn from examples and data. The AI technology can then draw conclusions and find solutions autonomously. However, specialists have not programmed these systems specifically for this purpose, as is usually the case in software development. Only a special algorithm is programmed.
A computer is trained with the help of data. It recognises patterns, dependencies and connections in large amounts of data. The system generates its own knowledge based on experience and links data "intelligently". It therefore learns autonomously and thus comes up with its own results.
These recognised patterns and regularities then serve the system - on the basis of complex mathematical calculations - to predict a certain behaviour or to solve a certain problem.
An overview of how machine learning works:
AI is a branch of computer science. Artificial intelligence is fundamentally concerned with the question of how intelligent human behavior can be imitated and automated using computers. AI uses sophisticated methods to simulate human behavior.
Machine Learning is a sub-area of AI. The focus is on pattern recognition and learning from large amounts of data.
The term Deep Learning is a special case, an advanced area of machine learning. Here, experts use so-called "deep" neural networks. These networks can classify information autonomously and can process enormous amounts of data. DeepLearning is suitable for particularly complex tasks and leads to significantly better results than pure machine learning.
A computer can learn to recognize certain patterns and assign objects or even people to certain categories. To achieve this, programmers must first develop a suitable algorithm. Then the computer must be trained and "fed" with data and examples.
Machine learning algorithms use mathematical and statistical models and methods. The difference to conventional programs is that the self-learning algorithm can find new solutions. The system generalizes what has been learned and draws its own conclusions.
The algorithm generates new knowledge from experience and can thus also correctly solve new queries with a high hit rate - for example, assigning an image of a previously unknown person to a certain category.
Basically, symbolic and sub-symbolic approaches can be distinguished in machine learning: The symbolic approaches provide information about the learned knowledge and rules. The sub-symbolic approaches (based on neural networks), on the other hand, do not allow any insight into the solution paths.
In the learning process itself, a distinction is made between two types of learning. There is the process of supervised and unsupervised learning. Supervised learning means that artificial intelligence reproduces rules on the basis of given values. The system is trained with known input and output data. In unsupervised learning, on the other hand, the system independently forms corresponding categories. The output data is not classified in advance.
According to a study by Lufthansa Industry Solutions, machine learning technologies are currently being used in many companies in the fields of image recognition, language and text analysis and text translation.
In larger companies, topics such as the automatic capturing of large amounts of documents or the upgrading of planning systems are in the foreground. For medium-sized companies, on the other hand, machine learning is above all a tool for improving customer orientation and customer service.
Machine learning (often in the form of deep learning) is already being used in many areas of everyday life.
Companies such as Google, Apple, Facebook or Amazon use the technology to recognize faces or make buying recommendations. Intelligent speakers like Alexa are also based on machine learning.
Machine learning can help to better understand the needs of customers - for example, based on their shopping behavior. In the financial sector, the technology also helps analyze stock markets, in the automotive sector in the development of self-driving cars, and in medical technology in the detection of cancer cells.
With the help of machine learning, many problems can be solved for which no human experience is available or for which a suitable computer program cannot be written immediately due to the complexity. These are tasks for which conventional software is not sufficient.
Machine Learning offers a whole range of advantages for companies. The technology enables a better use of existing data ("BigData") or to make use of it at all.
The data alone is basically worthless - only its evaluation with the help of suitable software makes it commercially viable.
Algorithms trained by machine learning reduce the number of errors, relieve people of monotonous tasks and perform highly complex tasks that humans cannot perform in the same quality - for example, in detecting possible damage in production (predictive maintenance) or in medical diagnostics.
Overall, the advantages of machine learning lie primarily in the improvement of processes and in a general increase in competitiveness.
With the help of AI, new business models can be developed. Due to a better data basis, companies can make better decisions. In addition, companies can improve customer relations, reduce costs and increase efficiency.
Especially where people make mistakes (such as when entering data manually), there is enormous potential. Intelligent assistants monitor the entries, reduce the number of errors and thus improve the quality of work. This saves time and money.
With SAP Leonardo, SAP has created a platform on which all activities for digital innovation are bundled - from machine learning to block chain and analytics. SAP speaks of a "Digital Innovation System".
SAP already offers its customers AI and machine learning technologies in many areas.
These include, for example, the use of AI in robot-controlled process automation. With the help of automation and machine learning, companies can reduce the proportion of manual activities.
Other areas of application of AI in SAP are voice-controlled chat bots. Artificial intelligence is also used in the SAP Cash Application (automation of the reconciliation of receivables).
The plug-and-play AI assistants also use machine learning technologies.
The reversal assistant, for example, is fed with millions of historical financial and material postings from SAP source systems, which it analyzes and processes.
In this way, the AI for SAP recognizes patterns and regularities. The more data is available, the better and the more is learned about the probability of incorrect postings.
The reversal assistant can then apply the acquired knowledge to new data - and recognizes a likely incorrect posting as soon as a it is entered.
Our other SAP assistants are also based on machine learning technology.