Hello, and welcome, everybody. This is Joe from Stella Bright. They're all coming to you today again from a very, very hot Frankfurt area. We are expecting to crash the 30 degrees Celsius today. It's gonna be damn hot. Nonetheless, I do have Danco here with me. Hey. Welcome. Hello. Totally. My pleasure to have you here because you're a very interesting guy. You are a brain researcher turned entrepreneur and working on machine Menninger.
And we get through this interview and, at the end, have a few ideas how machine learning can profit from your research and what your very interestingly named Startupradio, robots Joe Manta, actually does. But first, I would like to, to thank our enabler. This recording is supported by HDI, Essentrade and Invest, and Enterprise Europe Network.
These organizations have made tremendous contributions to helping start up businesses succeed and thrive, providing a range of services from helping to find grants to ongoing partnerships. By taking advantage of these resources, startup companies can network and develop innovative strategies for success on the international stage. The dedicated support of both entities is paramount in providing start up businesses with the
tools for lasting success. Look for our dedicated sub podcast in partnership with them called Tech Startups Germany. Now that we do have this out of the way, as always, I've been looking through your CV, and I found it extremely interesting. You have been teaching at the Technical University Darmstadt statistics. That Yes. Nonetheless, it it it takes it it takes quite a special person to really love
statistics and even teach that on university level. And we have to say, Technical University is one of the top technical universities here in Germany, maybe even in Europe. So, they don't take anyone as statistics teacher. Then she became a group leader, but the interesting twist here is in brain research for the Max Planck Institute, can you take us along your journey how you came from statistics to brain research? Because then it kind of makes sense how you ended up in machine learning and
artificial intelligence. Yeah. Yeah. Very gladly. So thank you very much for having me here. It's it's a pleasure to be here. So the story is a little bit different. So I was teaching it at, at only as a side gig when I was already at the at the, Max Planck Institute. So it was happening in parallel. Right? But how I came to the brain research, is is a story Okay. That's hard Let me get this straight. You you've been, at the same time, a BRAID researcher and a
statistics teacher? Yeah. Yeah. Yeah. Yeah. Yeah. Let's do a more. One can do that. I think I think about brain research is always about biology. But, apparently, you can also apply pretty much statistics. Sorry to interrupt you, but it's just Oh, yeah. It's quite uncommon. There is a lot of statistics in brain research. Trust me. Like, when you get all this data, when you record all these brain signals, you have to know what to do with that. Right? You have, like, bunch of
numbers in your computer, and you have to analyze them. You have to understand them. There's lots of statistics. Believe me. I was I was totally fine with all the statistics I've been taught in Germany and the US. Menninger high five through professor Owen from Midwestern State, by the way. He he was the only one who made it really entertaining. Disclaimer, I've never attended your lessons. Don't worry. And, also, statistics is in in many ways, statistics is understood as
as discipline that preceded machine learning. Right? So machine learning somehow built on top of statistical research, statistical analysis, inventions in in in statistics as well. Right? But that wasn't really my story. Right? My story was that I was first interested in artificial intelligence, like, back back then. Realized that artificial intelligence is struggling with doesn't have a good solution. There is no good way of of emulating brain, human brain in a computer.
And that's what triggered me to to try to understand the brain. So I started with studying psychology, getting PhD in cognitive psychology. So kind of first understanding the brain at the level of behavior. And then go then say, okay. Now I understand that, kind of. Now let's go into the brain. Then I went to to to, brain research. That's where I needed a lot of statistics.
And then after that, as I don't know if you mentioned, but I then I went to industry after some time at at the Max Planck Institute. I I would be curious. How was the interview with HR there? Because, for everybody, we we do have a quite a vast audience in in Europe where Max Planck Institution should be pretty well known. But for for all the thousands of people listening from abroad, Max Planck Institutes are one of the premier basic research institutions in Germany.
So you simply can't can't tell. I go there. You have to do your your undergraduate studies, your graduate studies, your PhD, and then do some impressive work that they will consider you. I do believe the vast majority of the noble arts in Germany is associated with the Max Planck Institutes. Oh, no. And then you you you go there to to to study towards HR. And what did she ask you? There is no HR in Max Planck. Yes. It's you. Believe it or not. It's actually no way. When you went
into industry, I I was curious. How did the interviews go? In the, in the industry? Yep. Well, quite well. You know, it took having coming from Max Planck Institute opens a lot of doors. Right? And nobody asked any more question whether I'm qualified to do statistics or machine learning. This is kinda assumed that you know. The questions that HR would in the industry ask was, like, how much do you know about industry specific things? Like, do you have
experience with customers and and things like that, which I didn't. Right? So I had to learn these type of things. But the the the competence, they don't question. They just respect. So that that's how I would should also give we should also give a mental shout out to, the the the place you did your PhD because it's not too far away from where I studied. They have a very well maintained rivalry with the university. I attended its University of Oklahoma. Did you, by the way, eat a lot of Oklahoma
onion burgers? Sorry, guys. The heat is really getting to me here. No. Well, I did eat some burgers, but I'm not great fan of burgers, I have to admit. So, yeah, occasionally, but, nah. Not too many. Not too many. Great. So we've been at the place we shifted into industry roles. What did you do there? Because I do believe, they would be they would be very happy to have you in any team, working on machine learning.
Yeah. Exactly. As a machine learning is one big topic, and the other big topic is statistics in general, data analysis. This is what also what industry needs, what companies need. And, this is what I could help with, quite a bit. Right? So, you know, any anything that has to do with lots of data, trying to understand the data, systematize, get insight into data. This is where I could help quite a bit. But also then, of course, machine learning coming up with with, ways
of machine using the data, basically. That's what machine learning is. Taking lots of data, compressing the knowledge that this data contain into a machine so that the machine could make some predictions or decisions and things like that. Right? So these two things. Right? I Startupradio, basically, as a data scientist. That's what the job is, described today. I see. And then you got into
several positions. I see research fellow, head of AI, senior AI consultant, a a lot of different positions in start ups, in tech companies, also in research institutes, for example, Frankfurt Institute for Advanced Studies, all gearing toward, different areas, but always, kind of put together by either either psychology, statistics, or both of them. And then you started robots Joe Menninger back in 2016 here in lovely Frankfurt. But first, how did you get the name?
I was looking for for a cool name. And I there were a couple of companies that that I I ran into, and I thought, they have a cool name. I would like to have a company with such a cool man. Right? One of them was what was it called? Incredible Machines. I was like, ah, that's a nice thing. And and, couple of others, I can't remember now. So I was, like, thinking. I was I switched on my creative mind for a while, and it took me a month or so to come up with something
interesting. But, also, you know, the domain had to be available. That's also another challenge that you have. Right? And then when I came up with it, I was very happy. Interestingly, the first person I I told this idea to didn't get it. It's like, what? I don't get it. So not everyone finds the name interesting. You do. Many people do. Maybe, I don't know, 50%, 60%. But other people just some people just don't get it. So the the the idea. Joe, you know, it's a, it's a 5050
job. And what do you guys are actually doing there? So, we have developed a new machine learning algorithm, which was inspired by my understanding of how the brain works. So if you compare the neural networks in computers and and neural networks in in in the brains, human or animal, there's one big difference. Right? The brains in, in computer, once they they are in production, once you do once you train your neural network and you put
it in production, you don't train it anymore. It's not allowed. And it's a really bad idea. Some people tried, It always ended up with a disaster. And, a famous example is when Google put one language model out and allowed it to learn while people interact with it. The model very quickly became racist and had all sorts of bad behavior because people provoked it. People knew it's gonna learn, so they actually intentionally tried to gave gave it bad bad, things because that's what human
do. And and then you Because you operate. That's what's being human. You provoke everything. You try to break things. And, and the model was helpless against mean humans. And today, you don't do it. You don't do it anymore. Right? There's even legally legally forbidden for autonomous driving machines to have training while the car is driving for similar reasons. Now mathematically, there's I can explain to you statistically, mathematically Right? Because human brain is not like that. A
human brain can learn. We learn as we go, and it's not a total disaster. We're not perfect, but it's not such a disaster. May I just say the bottom line is you develop a new machine learning algorithm that is training itself why it's already in production without going mental, being a total dick hat, or, putting out all crate crazy kind of stuff and, fantasizing, I do believe, is a call in some models. Right? Yeah. Yeah. Yeah. Yeah. You you said it right. You
said it right. So, Okay. Can you tell us a little bit about this? Because I do believe a lot of AI and machine learning people, c level executives in different companies, the their ears are now really perking up here. Okay. Very good. Very good. They should. So, basically, the reason why human brain is not, Joe susceptible to manipulation and to random inputs is that we have internal control of what is relevant, what is not relevant. We can direct our attention towards learning.
We can make decisions. What do we pay attention? We we know what is important, what is not important. We know where we wanna go. Right? The classical AI algorithms are not do not know anything. They are just completely open. Just teach me anything. I'll take anything. Just give me data. That's practical. For some purposes, it's a disaster if you have very small number of data points and if these data points are somehow not what you wanna learn. Right? So we have developed a way to guide deep
learning, to guide neural networks. Right? So there is another layer of parameters, another layer of knowledge, which which knows what is allowed, what is not allowed, what could be done, what what what shouldn't be done. Right? Basically, each parameter in a neural network has its own parameter telling telling it whether it should be how much it should be allowed to learn. Right? And the made most part of the network is not allowed to learn, only some parts.
Right? So what we have in addition to that, we have an algorithm that teaches this guiding algorithm. Right? So there are 2 levels of learning, basically. 1st, you have a phase where you learn how to learn, how to how to direct your learning. And after that, you can go in production and learn safely and know where you are going and being being directed and not being open to any kind of random input that you that you receive. Right?
So If if if I put it in my very, very layman's term, you put a parent in there. Because I do have a toddler here at home, and if he's not supervised, he'll do all crazy kind of stuff. Not to mention, he has, like, a little shopping cart, in which he sat backwards brushing his teeth or stuff like this, and this is just the tip-off the iceberg. So, basically, you develop kind of a a parental algorithms for toddler AI. Is it something like that? It's it's
yeah. It's whatever is left in in us after we have been parented. Right? So, like, you know, you gotta get parented, but then after that, you internalize these rules. And you're like, you don't do this. No. No. That's a no no. Right? So these internalized things we give we give to the network. Right? Yes. That's that's correct. Right? I Another another way before we proceed, would it be possible that you because a lot of people this is, this is not necessarily
like a coding deep tech podcast per se. We try to keep it on a very Bonnie level. Could you, before we proceed, kind of split a little bit for us what you mean by talk about AI, machine learning, and deep learning because they're not necessarily all the same even though people use them interchangeably. Yeah. Okay. So machine learning is a discipline on how to make machines learn
something. Right? It's a very general discipline. There's lots of different algorithms, lots of different ways to make machine learning. Deep learning is one subset. It's one set of algorithms. There are many more that are not deep learning, but deep learning is the most popular one. It everyone talks about deep learning. And the reason is is that deep learning is exactly open to learn anything.
And you can expand the number of parameters. You can expand the amount of knowledge practically infinitely, and it will still keep learning. It can't just take anything in in itself. The price is that it requires lots computation, lots of data, but it can learn anything. That's exactly the problem we needed to solve. Right? Now AI is an term where you that you refer to in a popular way in the resulting machine that you have built. Joe you you put
some deep learning, some other machine learning algorithms. You put some other algorithms that are not machine learning and yet are AI algorithms. You put everything together. You pack it. You make interface. Now you have an AI. So that's how I I, see it, how I understand it, how I use it. Okay. Can you give us some hints? Because you've you've been talking about, you've been talking about let let me take one
step back. Let us because a lot of people who are out there own companies, work for companies, that would profit from a new AI teaching methodology algorithm, like you said, when when they can still learn, when they're in production. The thing is, let us get a little bit through that by first telling what you are allowed to learn, what not, and maybe we drew your knowledge from to do that, and then we could expand into where you think,
theoretically, this could be applied. Because my understanding is you develop the algorithm, but it's not, broadly applied yet. Yeah. That's correct. Actually, you know, the challenge we have right now is to find the right use cases. So it's it's not the same you know, one step is to create great technology, and the other step is to find a great use case cases
for the great technology. And you can know really be sure that you have a great technology, but still it takes time, and it's still an effort to find the right use cases, to find the right market fit. And that's our challenge right now. And and, I think we have, like, lots of examples in other technologies as well. Like, the first computers people did not know how they will use it. They knew they have some amazing machine that can do so many different things. But how
exactly it's gonna be used, they didn't know back then. Right? They didn't know that that we are gonna use it for to play to to for word processing or Excel sheets or or it took time to invest the invent those. I know computer games. I don't know. But similar happens here with large language models. We are still searching the right use cases. Although we we have some we know we have some amazingly powerful machine there. So the same challenge
robots Go Menninger is facing right now. Find the right use case. We have identified several potential ones that we are working on. Right? For example, a car that is adjusting itself, its visual system constantly while it's driving. Like, for example, when a when a when you are driving on a sunny day and suddenly you enter a tunnel. The the the laws of visual, shades, the lightning, and so on, it completely changes suddenly. Right?
And what our human brain does, it adjusts. It takes a few seconds until you can see clearly when you enter it to know. Right? The the AI solutions that we have today, they don't adjust. They have no capability of adjusting. Right? With our solution, it should be possible. Right? Similar thing with when you when you have, say, conversation. When machine has a conversation with humans and has to listen to to to, people's speech, different people have
different accents. Right? And we adjust to accents. Right? So when I start talking to you in the morning, like, the first few sentences, you may not be really it would be hard for you maybe to understand because I have a unique accent, which comes from my Croatian language. But after a few minutes, you get adjusted. Oh, doctor, don't worry. Be before 10 AM, I barely under understand anybody. I'm not a boring person. Okay. So, you know, that's another use
case. Right? So we are we are we are looking for those, and and the challenge is to find the right market fit. Mhmm. Can you talk about industries? We just talked about self driving cars. I do know there was a lot of AI involved, for example, during the development of the COVID vaccines to end the pandemic. Talk about health care, biotech. Would you think your algorithm could also be applied there, financial services and so on and so forth? Yeah. I I I'm sure it can be.
But how exactly? I don't know yet. Right? But what what we we have a very nice piece of evidence that it the algorithm can reply there because we have, performed one project together with NASA, exactly in this field. And, what we did, we trained the neural network to learn, how to deal with RNA sequences. Joe you basically one one general problem that is relevant for for biotech industry is to process RNA information. You have a sequence of of nucleotides, and you need to work with
that as an AI solution. Now like every other AI, you need a lot of data to train anything. But NASA had 6 mice only in space, and they couldn't increase the data amount of data because it's just prohibitively expensive. And they they just can't pay, I don't know, 100,000,000 to send another 600 mice in space. So they had to find a solution how to train a model with RNA coming from 6 month month mice in space, and they had it at the same time as a control 6 mice on the ground. And they
had to train the model with it. So they discovered us. We worked together, and and it worked. It worked. So we could use the Earth, a lot of massive Earth data to train the model, to learn how to learn. And then after that, it could deal effectively with just, 6 data points from space. And and, we wrote this paper recently Joe one can find it. Of of course. We link it down here in Joe show notes on our blog post. You'll find it at stethubreight.oforward/block. And and then I
I was curious. So your model was basically your algorithm was first applied to teach the other algorithm how to learn? Well, there's a yeah. There are 2 algorithms. One algorithm that, knows how to learn, but this one has to learn first. So there's the first an algorithm that teaches the second algorithm how to learn. Right? So there there is a hierarchy of algorithms. Like you said, the the parents and the there's a parent and the The parent algorithm algorithm.
I see. I see. I see. Very, very interesting. Joe, everybody who'd like to learn more about your algorithm, we'll link down here in the show notes. First, your company's website. Secondly, your LinkedIn profile. As always, we will be ending with a few questions. First of them, are you looking for talented people, HR, new hires? I'm I'm looking for talented people all the time. Right? There is, it's great pleasure to find talented machine learning people. They are not easy to find, I have to say.
And every time I, discover someone, it's a great pleasure to work with them. Right? And if there are, any, say, PhD students interested in a project, people who have finished PhD interested in a great project, please contact me. We have we have really interesting problems to solve. Right? That that are academically interesting, commercially interesting. I would be interested. You're likely not looking for the normal coders here.
Are you looking more for people with a statistics, computer science background, or do you look more for people with a psychology background? More statistics slash machine learning, which includes, of course, a lot of computer science understanding. Right? That's the is the the rarest breed. These are the hardest to find. That is already pretty good. Would you guys be open to talk to new investors?
Yes. Definitely. Definitely. Yes. Yep. Yes. And the last question, this is, supported by Hessen Trade and Eves and the European Enterprise Network Hessen. You would now have the chance to address a question, a concern, an idea for improvement for the state's Startupradio scene. If you could address those decision makers, what would it be? Well, one experience I had by, you know, starting a company was that in general in Germany and in Hessen,
it's very hard to start a deep tech company like we are. And deep tech basically means that you have to develop a technology before you have a use case, before you have a market fit, before you have a customer. There's a lots of work behind that. And my feeling is that this is this is really hard to get in Germany. The and that's probably the reason why we see most of the companies that come out with such successful concepts are not coming from Germany, but from United States.
From UK is also much more, effective in that way. And we somehow, it would be really great if we found a way that works in Germany for deep deep tech. Right? It's not just there is a lots of talk about deep tech, but there's not lots of it's not coming out really in a great way. We don't have a list of great German or Hessian companies that have actually brought new AI
technologies to the world. Right? Everybody's trying to just pick up whatever he's using and just go and make a make a business model, which is great, which is, of course, useful, but it's not how you become a leader. Right? We need to become a leader. And to become a leader, you cannot do it with with the system the way how it works right now. I don't know why what is broken. It's not only the government. The governments can help a lot. But, also, I think it's generally mindset. I don't know.
Also, investors are generally not don't have the stomach for such risks. They would like to go quickly, you know, have your customer first, and then we'll talk. And, fine. Reducing risk is fine. To become a leader, the world leader, that's not possible that Right? That's my that's my, message. Find out a way to become a leader. World leader help us become a world leader. We wanna be world leader in the technology, in developing fundamental technology for AI and help us.
And we don't we did get a small grant from Hassan. Thank you very much. But we would need also some some bigger support. If if not, we'll fight. We'll keep fighting. Awesome. Danco, it was a pleasure talking to you. Thank you very much. I really enjoyed the interview. Yeah. Me too. Thank you. Thank you. Have a good day. Bye bye. Bye bye.