Camo newsroom

Industrial analytics thought leadership-series.

How will industrial analytics, Industry 4.0, IIoT and digital transformation change businesses? And ultimately the world? We ask industry leaders to share their view on these technologies, how we use them, and where they will take us.

Harald Martens:

We need more love for real life

It is still like walking a slack line to connect mathematics and statistics to all other sciences. The concept of filling this gap with black box AI is an illusion, that will create dangerous alienation. To professor Harald Martens multivariate analysis promises the needed two-way bridge to make a successful crossover. To understand his view-point we have to go back to Uganda in 1972.

“My first job as bio-chemist was in Kampala in Uganda in 1972, with my first wife, Magni Martens. It was a very political period in the years after 1968, and it was extremely incorrect to go for a career: instead, we wanted to improve the world! In Kampala we analyzed 50 local grain types for content of amino acid nutrients. When we came forward with the result, we were told it was “too much data”. I was furious, and left the meeting in protest. But it started a thought process, and the analytical approaches I have been working with ever since,” says Harald Martens.

From “precisely wrong” to “approximately right” modelling

By trial and error, he taught himself enough linear algebra to be able to analyze simple mixtures of known chemical constituents, by mathematically “unmixing” them using a mathematical model based on their known individual multichannel properties. This worked, but only under ideal conditions. Data from the food samples, that he analyzed, first at University of Kyoto and later working for the Norwegian Food Research Institute (now Nofima), often seemed to reflect also unknown chemical constituents, and physical effects as well. How to model the unknown, mathematically?

Gradually he found that by extending the linear algebra of theory-driven “unmixing” into a more pragmatic “unscrambling”, it became much easier to model and interpret multi-channel data from real-world samples. This was done using data-driven modelling techniques from psychometrics. The term “unscrambler” is associated to the notion of scrambled messages during WW II, that you would want to decrypt and understand. Ten years after the Uganda experience he presented a working version of “Unscrambler” for Camo Analytics, and the company took in the product and concept, and now this analytical software has a worldwide user community of domain experts working with scientific data.

“It’s like nature or reality is having a message. It may drown in detail, but you want to find the dimensions and the angle in which the message stands out. Finding the pattern in data – the harmonies and rhythms – is what unscrambling is about,” says Harald Martens.

The taste of a pattern

As a child Harald and his brother were quizzed by their mother about ingredients in the food. She may have prepared a meal, and the boys were frequently and seriously asked to look, smell, taste and chew and guess the ingredients using their senses. They became good at it over time, and they learned to see the patterns in the complexity of ingredients, taste nodes and processes in cooking.

“I left Uganda with the distinct feeling that it would be possible to work with data in another way to capture the patterns and meaning in all the data. I came to think of my mother and the food quiz in my childhood. What we did was a qualified guess on ingredients based on data from sight, smell and taste. We became pretty good at finding the pattern, and I got that same feeling about the data about grains and amino acids. There is a pattern, we just haven’t found a name for it,” he says.

The music in data

Another clue came from music. Harald was an extrovert child being raised by an introvert father. The need for recognition and mirroring was frustrated except for piano playing.

“We played in duet with four hands on the piano, – relatively simple stuff, but this was a true connection. We improvised and had an emotional contact, while working with a limited set of musical patterns. The way we did that and the harmonies created was again patterns growing out of something else. Obviously important to me growing up, but again the ingredients of tones forming patterns and making new meaning.”

Axis rotation

The quest for a way to understand complicated datasets like the dataset from Uganda led Harald to Japan, where he worked with programming in Fortran. Back in Norway he confidently applied his software to other types of data, and problems popped up: totally different results coming out of the same data, depending on how he initialized his home-made software. In desperation, he also tried to run it with IBM’s quadruple precision, with no improvement. The algebraic idea of axis rotation was then presented to him from psychologists doing psychometrics: You can read a map upside-down or sideways or any way you want; the landscape it depicts is still the same.

“I got three pieces of stencil paper and read about matrix algebra and axis rotation. One afternoon of reading, and the problem was solved. This led to different prototypes and the early stages of Unscrambler,” explains Harald Martens. As a bio-chemist, he just loves the power and simplicity of matrix algebra.

The observation of a mix

“Whether it is flavor of spices, harmonies in music or chemical components in grain or on Mars, we are observing different mixed signals. We may observe a theme, rhythm and trend, but it is to some degree unknown to us, and we will have to unmix or unscramble the data to identify the message. We have to be open for systematic patterns in the unknown, as nature and the world is sending us these mixed signals.”

Harald Martens got great help from patient mathematicians and statisticians. But their ways of thinking sometimes felt alien. And he is still angry about how uninspiring his own math studies were as a bio-chemist. He feels that mathematics and statistics is still sometimes in a high tower with too little cross-over to other sciences. From talking to him, you sense that the initial crusade of Unscrambler was to give the power of mathematics also to scientists in other disciplines.

Blind machine learning?

The soft-modelling chemometrics culture successfully survived some hard storms from the traditional, overly hard modelling cultures in science. But Martens and his friends then met strong winds from the opposite direction – machine learning.

“It is still the idea, that we should discard all prior knowledge and experience and let all meaning derive from data. I think it is a mistake, and the talk right now about explainable AI is related to this criticism. One of the recent issues of Wired Magazine is battling AI head on, as it needs to be explainable to be socially acceptable. I really enjoyed reading it,” he says.

“To me it is about building on the knowledge and experience already here and then get data and build new knowledge from there. To me Unscrambler may be seen as explainable AI.”

Do it yourself

“Data is not just data. They contain the effects of several causalities, only some of which we know how to model explicitly. But domain experts have a lot of implicit, tacit knowledge. We should not exclude that knowledge when we want to interpret real-world data,” says Harald Martens.

“The effects of big data can lead to important breakthroughs. But to describe complex systems you should also apply knowledge to start off intelligently and use both logic and intuition in the analysis.”

He believes in the combination of theory-driven and data-driven modelling, with proper statistical design to ensure informative data, and statistical validation and lots of graphics to avoid over-optimism.

“To the domain expert my message is “do it yourself”. In most cases, if you don’t do it, nobody will analyze your data. Unscrambler is about that, and to me this concept of multivariate analysis to be used by non-mathematicians is still the biggest “non-celebrity” in Norwegian technology.”

Harald Martens has a hope for the future. We should have more love for real life, have more self-respect, have more respect for the expertise of others and more self-irony, and we should be a little less fearful of trying something new.

Harald Martens:

Professor and entrepreneur

When the analytical software tool Unscrambler from Camo Analytics saw the first daylight, Harald Martens was in the delivery room. His ideas of bridging the gap between scientific domain knowledge and hardcore statistics is still at the center of Camo Analytics, and Harald describes the Unscrambler as his love child.

Today Harald Martens is retired from the Copenhagen University, Norwegian University of Life Sciences, and the Nofima food research institute. He presently works as research leader in Idletech AS and as professor emeritus at the Department of Engineering Cybernetics at the Norwegian University of Science and Technology in Trondheim, Norway. He has written several hundred papers and several books, which have been cited more than 20 000 times.

Harald’s background is in biochemical engineering and bio-chemometrics, and his research has primarily focused on new methods to convert multi-dimensional measurements into understandable displays and quantitative predictions. He is now engaged in combining chemometrics, cybernetics and artificial intelligence in what may be called Big Data Cybernetics (BDC), and in new ways to teach math and statistics. Harald is a member of the Norwegian Academy of Technical Sciences and has received several international prizes.

“To the domain expert my message is “do it yourself”. In most cases, if you don’t do it, nobody will analyze your data. Unscrambler is about that, and to me this concept of multivariate analysis to be used by non-mathematicians is still the biggest “non-celebrity” in Norwegian technology.”

Find the right solution for your analytics needs.

Get in touch if you have questions about our products, platform, how to get started or how best to address your analytics needs.

Contact form