Increasing amounts of data
Our world is becoming more and more digital. In almost all areas of life - private and professional alike - data is collected: we accept cookies as we surf the internet, we make purchase using credit cards, we use navigation apps and spend time on social media. Machines send data about use, operating temperatures and output. Weather stations collect information on precipitation and wind speed.
Machine learning: systems that learn from data
Companies have been using data for a few years now, by using business intelligence software for data analysis and reporting. This data could be used even more effectively by using machine learning. Machine learning falls under artificial intelligence, and is an umbrella term for the artificial generation of knowledge from experience.
An artificial system - software - uses examples to learn rules that it can generalise after this learning phase is over. A machine learning model is able to compare apples to pears, for example: the system recognises patterns and regularities - in this case, in photographs of apples and pears. This learning data is then used by the machine learning model or algorithm to recognise whether a brand new image contains an apple or a pear. It then checks whether this recognition was correct or not. The quality of a machine learning model is measured by how high the error rate is. The more relevant data a machine learning model has available, so the more ‘practice’ it has, the better it learns from this data and the more precise the machine learning algorithm will become.
The new era governed by data
Artificial intelligence (AI) allows computers and machines to “think and act like humans”. AI-based software has therefore already introduced enormous changes to the business world. Even if the general future is unclear, AI-based BI tools are certainly indispensable for a competitive presence in the business of tomorrow.
Supervised and Unsupervised Learning
In this example, machine learning takes place using verifiable data as supervised learning: the test data is given a correct label (apple or pear) by humans in advance. The algorithm can then compare its prognoses with the actual results and improve itself step by step.
This is different from machine learning using the unsupervised learning principle: no learning target is provided. The machine learning model is fed with data without any prior definition of what this data means. The machine learning algorithm recognises patterns and divides data up into clusters or categories without knowing what categories these are, i.e. under which labels they fall (such as ‘apple’ and ‘pear’).
The majority of machine learning processes used in practice entail supervised learning.
Deep Learning - what is that?
The so-called Deep Learning is a part of Machine Learning. This special method of information processing is based on neural networks. Thereby artificial intelligence is created by training methods. These methods are very similar to human learning. With the available information, the system can always link the learned information with new content and thus learn again. The machine is able to make decisions and forecasts and to question them again and again. This process then leads to a confirmation or a new start. As a rule, humans do not intervene in this learning process. Deep learning therefore teaches how to teach machines in practice and is particularly suitable for applications with large data sets from which patterns and models can be derived. The basis for this is formed by artificial neural networks, which are constantly being reconnected during the learning process.
Automated Machine Learning with DataRobot
Predictions made easy
DataRobot enables users to create and bring into production highly accurate predictive models in a fraction of the time of traditional modeling methods and tools. DataRobot identifies the most effective forecast model for the current case from a broad range of predefined models to best represent future forecasts. Furthermore DataRobot provides a continuous monitoring of the productive models in order to keep them up to date or to adapt them proactively.
Machine learning in practice
A whole range of technology that we use daily in our private lives is already based on machine learning algorithms. Examples are facial recognition in photos and videos, speech recognition on mobile phones, translation programmes online and computer spam filters.
Machine Learning for Manufacturing
In industry, ML is used to collect empirical values from machine data, such as after how many hours of operation a part must be exchanged. This predictive maintenance allows maintenance schedules to be planned in advance. Expensive downtime caused by part failure is therefore avoided. Machine learning methods also allow the prediction of price developments regarding end products in batch production based on the availability of raw ingredients and their costs.
Machine Learning in the energy sector
Energy suppliers use machine learning algorithms to coordinate supply and demand as well as to recognise anomalies in the power supply. .
Machine Learning for Financial Services
Banks can use machine learning algorithms to make credit decisions for retail and business clients, such as in relation to credit checks, establishing credit conditions and recognising credit fraud.
Machine Learning for Healthcare
In healthcare, machine learning-based similarity analyses of patient files, patterns and connections can be used to optimise the treatment of new patients.
Machine Learning for Retail
Retail can use machine learning for more efficient market research or personalised customer communication to reflect customers’ purchase preferences.
Machine Learning for Telecommunications
Telecommunications companies can search network traffic for anomalies indicating attacks.