Machine Learning

Everyone is talking about machine learning, but the technology is actually older than Sputnik 1 and the Miracle of Bern. Why is that? The potential of machine learning for the development of breakthrough innovations—ranging from autonomous driving to early cancer detection and open banking—has only just started to be exploited in recent years, but its rise is as inexorable as the metaphorical ‘data flood’.

Machine Learning The engine for innovation

Machine learning (ML) is involved in some way whenever people communicate with Alexa, Cortana or Siri, as it is in autonomous driving, speech and facial recognition, machine-based early detection of illnesses, crime and share prices. ML is a sub-field of artificial intelligence (AI) which is predicted to facilitate phenomenal innovations, even disruptions, profits, progress and/or threats. What are the implications of ML, and what is the story behind the hype surrounding this technology, which is not even that new? And what companies and startups are using ML in particularly promising innovations?

Humans prefer to delegate any tasks that are taxing, boring, annoying, or simply impossible for biological reasons to machines, such as plowing, sorting screws, transportation, calculating, navigation, flying and deep-sea diving. Machines are faster, more precise and stronger than a human could ever be, but the hierarchy is preserved as long as homo sapiens has one crucial advantage: being intelligent and capable of learning. Regardless of whether this is actually true of all humans, the question of whether machines can also be intelligent and capable of learning arises. Keyword: Machine learning, along with other buzzwords, such as artificial intelligence, deep learning, artificial neural networks etc. Does human intelligence outside of homo sapiens exist? Can activities like thinking, learning and drawing conclusions be mechanized to such a degree that a machine can perform them as well? These questions are not new: The first evidence for contemplations of this kind in the modern era is thought to be L’Homme Machine by the French doctor and philosopher Julien Offray de La Mettrie published in 1748. The American computer scientist John McCarthy first coined the term artificial intelligence (AI) in 1955. Alan Turing is probably even more famous. The Turing test, named after him, determines whether a machine is intelligent: According to the test, this is the case if a human communicates with a computer via keyboard and screen for 30 minutes without noticing that the counterpart is an artificial being.

Data + Algorithms = Intelligence?

Until today, there has been no agreement whether a machine has ever passed the Turing test and can therefore be considered as intelligent. For decades, the topic of AI/ML lived its existence within the spheres of Science Fiction, outside of specialist circles, until around the beginning of the decade, when the entire public domain became abuzz with the word: Some hope for this (not-so-new) technology to provide plenty of (honey) rewards, while others fear the attack of the killer bees or machines. Why the new hype?

Machine learning is nothing new, and strictly speaking machines do not learn. Rather, having no data to work with initially, they are given instructions on how to proceed in the form of algorithms, which have been in use in some areas for decades. Machine learning is however a tool for innovation, because now algorithms have an unimaginable volume of data to draw on, enabling the development of highly attractive applications from a business perspective. For ML depends on data and the computing capacity to process that data. Each post, each YouTube video and Instagram photo increases the amount of nourishment available to ML, not to mention the data collected by companies in a targeted fashion. From this data, the learning machines generate their output: Pattern recognition. This is because ML is a sub-area of artificial intelligence, which primarily serves to detect patterns. IT systems (basically consisting of hardware and software) use data and algorithms to detect patterns and rules and develop solutions. This is impossible without humans, as the data and algorithms first have to be fed into the system. Only then can such machines find, extract and compile the data which is impossible to find for human brains among this flood of data. Based on this, machines can make predictions and calculate probabilities, and even correctly generalize new, unknown input data and optimize processes with example data after some training.

Theoretically, that is. Now, what do we do with that, and how do we use it to generate innovative, successful business models? (successful in the sense of producing major revenue.) For example, e-commerce: Thanks to learning search engines and filters, between 2007 and 2017 the number of online shoppers in China soared from 46 million to 533 million. In 2018 (as of April 2018) approximately 529 billion euros of revenue were generated in the e-commerce market in China (source: statista).

Startups developing ML applications in the area of …

Autonomous Driving

www.aurora.tech
On its website, the company Aurora declares “We do self-driving cars”. The company was founded in Palo Alto in 2017 by three leading figures in autonomous driving—Sterling Anderson, Drew Bagnell and Chris Urmson—with the aim of developing technologies for Level 4 autonomous cars, i.e. fully automated vehicles where no driver is required.

www.mobileye.com
Mobileye develops driver assistance systems using optical technologies effective in reducing the frequency of accidents.

www.understand.ai
understand.ai is a machine learning startup based in Karlsruhe which processes image and video data for autonomous driving using self-learning algorithms.

Fintech/Tech

www.candis.io
The software of this company can be used to partially automate accounting for small and medium-sized companies.

www.crealogix.com
Not a start-up but rather a company founded in Switzerland in 1996 which went public in 2000 that offers an array of digital banking products for the financial industry, including online and mobile banking systems, online banking security, finance portals and digital financial advisory tools.

www.deposit-solutions.com
This Hamburg-based startup operates the open-banking platforms Zinspilot and Savedo. Private savers and investors can use these platforms to invest with foreign banks and lesser-known German institutions offering somewhat higher interest rates than regular German banks.

www.smacc.io/de
The use of artificial intelligence enables the finance and accounting solution by Smacc to gain updated daily insight into the company finances, making these transparent in the process. The software helps small and medium-sized companies to handle financial tasks with more efficiency, thereby facilitating their daily business.

www.squirro.com
Founded in Switzerland in 2012, Squirro has developed the cloud platform Cognitive Insights, which uses data engines and AI technologies to compile and automatically analyze unstructured data from internal and external sources. Users include financial service providers, insurers and telecommunications providers, which employ the platform to increase customer loyalty and open up new sales opportunities.

www.quantilope.com
This company developed the agile insights software quantilope to simplify market research processes. The software combines quantitative market research methods with machine learning and artificial intelligence, giving companies real-time insights into consumer needs, purchasing motivations, trends and customer habits.

Digital Health/Medicine

www.boca-health.com
The Milan-based startup Boca Health has developed a non-invasive solution which can monitor and measure body water content in humans. The purpose is for users to be able to use it to prevent heart and kidney diseases.

www.fuse-ai.de
In association with the Wuppertal-based radiology center Radprax, the startup Fuse-AI from Hamburg has developed a prototype for the automatic detection of prostate cancer.

www.inveox.com
Inveox digitizes and automates pathology laboratories, thereby increasing the certainty and reliability of cancer prognoses while enhancing lab efficiency and profitability.