Table of contentsKey takeawaysCollaboration platforms are essential to the new way of workingEmployees prefer engati over emailEmployees play a growing part in software purchasing decisionsThe future of work is collaborativeMethodology
Dr. Andreas Maier joins us on Engati CX to talk about Machine Learning and its interdisciplinary nature.
Dr. Andreas Maier studied Computer Science, graduated in 2005, and received his Ph.D. in 2009. He developed the first online speech intelligibility assessment tool, PEAKS. – that analyzed over 4.000 patients and controlled subjects so far.
He was a postdoctoral student at the Radiological Sciences at Stanford University. Where he started working on a flat-panel C-arm CT as Laboratory in the Department of Radiology. He then returned to the University of Erlangen-Nuremberg. As head of the Medical Reconstruction Group at the Pattern Recognition Lab. And finally became a professor and head of the Pattern Recognition Lab.
In 2018, he was awarded an ERC Synergy Grant for the “4D nanoscope.” His current research focuses on medical imaging, image, and audio processing. As well as, digital humanities, interpretable machine learning, and the use of known operators.
Interview with Andreas Maier
This section will contain a summary of our interview with Dr. Maier. But, if you’d rather hear the full interview, we have the Spotify podcast embedded below.
The current state of Machine Learning is fascinating to Dr. Andreas Maier. He worked in pattern recognition and machine learning for quite some time, and the field is lifting off.
There’s a lot of interest. And a lot of investment in machine learning by industries currently. What stands out, in particular, are the ideas everyone is coming up with these days!
The state of Machine Learning has changed. A couple of years ago, for example, we were limited to classification problems of 5-10 classes. So applications were much smaller.
But today, computer programs are capable of much more! So, it’s really great for Dr. Maier to see how we have been advancing in advancing in the last few years.
It’s a great time to be a researcher.
AI Chatbots are an interesting technology. With a background in speech recognition and dialogue systems, Dr. Maier thinks we can work on integrating these technologies to automate call centres. We have to find the sweet spot where the application makes sense.
Finding the sweet spot: an example
There was a spin-off company that wanted to put a telephone system in their larger companies. One of the features of their telephone dialogue system was that if one calls and asks for a specific person, they’ll immediately get connected. Almost like a directory service.
Suddenly, the number of inquiries for this system had tripled. According to Dr. Maier, it seems that all of the employees started calling this system to be connected to other members. So the automation of these particular tasks is really helpful in the long run.
Dr. Maier thinks it’s very good technology. It’s an interesting technology to find the information we need when we need it.
The capabilities are endless. When chatbots get to a certain level of sophistication. i.e, when they become smarter and can detect emotions, it’s going to be a game-changer.
Why are we living in the spectacular now? Because right now, we’ve realized how much we can accomplish in a digital way.
For example, we don’t have to travel to conduct interviews. Many things are done remotely which saves time and money. “It’s going to be interesting,” says Dr. Maier.
We’re also realizing that we can automate so many tasks. Which again saves time that can be invested for other things.
Dr. Maier finds that it’s hard to predict. No one could have foreseen this virus. And it’s currently forcing everyone to embrace digitization. Dr. Maier advises everyone to keep what people really want in mind.
Now we see that digital methods are accelerated dramatically. My world is exploding, it feels like a surge.
The Big Data of the Past for the Future of Europe is a European time machine project. Its aim is to help research in history. Essentially, his team is planning to build a repository of about 2000 years of historical data. It’s currently in its infancy.
The Big Data of the Past also hopes to make data accessible to tourists for virtual time travel. For example, one can virtually visit an area and experience what it looked like years ago. Or to not travel, but to stay at home and experience a completely virtual vacation.
It’s a huge effort. And Dr. Maier finds that people don’t realize that in a centralized system. So now the team is planning to adopt techniques like standardization, and request for comments. This allows everyone to explore their own history and make it available to others.
Of course, the team has to be mindful of licensing concepts. Yet, institutes like European are already publishing archives and cultural heritage. so that it can be made available for the general public. It’s a large scale project, so we’re hoping to learn more about it soon.
“Artificial Intelligence for Reinventing European Healthcare” is actually a program for summer schools. Its purpose is to assemble people across Europe to teach developments in deep learning and machine learning in medical data.
However, one of the drawbacks of the program was not having access to data. So one of the things Dr. Maier and his team wanted to pursue data donation. To build a system to donate data and share it for scientific purposes.
Why do we need data?
Well, to be successful in the healthcare market, one must develop a product. But if it’s not medically certified, it will never reach a patient’s hands. To receive certification for these products, we need data.
So after three years of trying to get approval from the GDPR, a system was born. We can now donate image data and reuse it for crowd sourcing.
What’s brilliant about this system is that donors can withdraw consent at any point. Dr. Maier believes that this is a crucial building block as it builds trust in how data is collected and used.
Dr. Maier often hears critiques that GDPR is limiting Machine Learning in Europe. One doesn’t have access to large data sets like you would in the US.
If you look closely at these large data sets, you’ll notice that label quality is a huge concern. You can’t solve tasks with deep learning because of this concern. In healthcare, you have to ensure your data has the right labels so that you make the right predictions. So now Dr. Maier raises a new question-
If you have access to all this data, will it really bring this advantage? Or are we putting patients at risk without guaranteeing that we can get this advantage?
This is a contract that must be discussed. This is also why Dr. Maier is so passionate about data donations. You can build good systems once you have approvals and the right labels. Even if it’s for small data sets.
Currently, deep learning and machine learning need large data sets. But can we reach a point where we can develop systems that achieve similar performance?
Can we build better models with more generalized points?
In conclusion, Dr. Maier can’t guarantee that demolishing GDPR will bring a large surge to medical research.
It’s not that research is impossible. But we have to pay attention to regulations. We already showed that we could also gather data at large scales, even if GDPR applies.
Essentially, the idea is to mix deep learning methods with traditional processing methods. This could reduce error and enable us to build systems, trained on smaller amounts of data.
Current algorithms' paths are analyzed through classical error analysis and measurement techniques. Techniques that we know from classical sigma processes. Of course, classical methods can't explain everything. So there are a few bits of unexplained data.
We can exchange the unexplained data with deep learning methods to get results.
This is why Dr. Maier suggests integrating Deep Learning as well. You have the advantage of being explainable. Especially when you don't have access to large amounts of data.
This technique could benefit healthcare, an industry where you don't have access to large amounts of data. You have the advantage of being explainable.
This allows you to explain what your system is doing, and why it's safe to use in a clinical practice.