<img height="1" width="1" style="display:none;" alt="" src="https://px.ads.linkedin.com/collect/?pid=2085737&amp;fmt=gif">
Tema: AI

#0995: Data fusion for smart environments

Gjest: Pierluigi Salvo Rossi

Professor of NTNU

Med Vert Silvija Seres

In this episode of #LØRN, Silvija talks to professor at NTNU Trondheim, Pierluigi Salvo Rossi. Pierluigi is originally from Italy, but has lived and worked in Norway for many years. In the conversation, Silvija and Pierluigi talk, among other things, about society's expectations of AI and ML, and what AI and ML actually can contribute in relation to expectations. Furthermore, Pierluigi explains that AI does not necessarily replace human knowledge and skills, but that technological development can help people to acquire new learning and more knowledge.

Full transcript

With Pierluigi Salvo and Silvija Seres

Velkommen til Lørn.Tech - en læringsdugnad om teknologi og samfunn. Med Silvija Seres og venner.

SS: Hello, and welcome to a Lørn chat. My name is Silvija Seres, and my guest today is Pierluigi Salvo Rossi who's a professor at the Norwegian Technical University in Trondheim. Welcome.

PR: Thank you. Hello to everyone, and thanks for inviting me. Very nice to meet you here.

SS: Very nice having you here, Pierluigi. And I see that we're going to talk both about your research, but also probably about your hobbies. And let's see where it takes us. I'm just going to say a few words about the series so people know what they're listening to. So basically, this chat belongs to a series we have defined as applied AI. And the idea is to give people examples of where machine learning, data, artificial intelligence, and things like IoT can be brought into their everyday life and everyday work. Many people still think that this is a little scary and a little distant, and we want to give them lots of examples with a tiny bit of theoretical experience around them. Sounds good?

PR: Yep. Sounds good.

SS: Okay, cool. So, then we'll start and my question is always basically. Who are you, and what has made you so?

PR: Yes, well, that's a strange question. But I consider myself like a person that is somehow involved in technology development, mostly doing research. I've been interested in research since when I was at university as a student. Before that, as a child, as a teenager, I was quite curious about many things, including music as he was mentioning, I was playing the guitar. But then since I started University and engineering, I became more focused on this domain that is called signal processing that somehow is related to AI. And, yeah, I've been focusing more and more on research, because I got excited with this kind of work. And then I've been working most of my life in academia, between Italy and Norway with some visits also in Sweden, and the US. And a couple of years of experience also in the industrial private sector. I've been a data scientist for a couple of years with the Kongsberg group. But still mostly doing research. Applied research work. Yeah. So maybe this is me from a slightly professional view.

SS: I have three questions. One is what do you do with the guitar? The other one is I see that you were interested in math, philosophy, and literature before you became a very pragmatic engineer. I'd like to hear you on that. And then I'd like to hear you about why Norway

PR: Yeah. Well, first, guitar. Nothing professional. It was just a hobby. At school, I was playing in a band with friends, as many people are doing. I really like hard rock and heavy metal music. So, this has been a steal. What I listen to now when I'm working as we listening to this, when I was studying is kind of my constant soundtrack. In my activities.

SS: Give me examples, what do you listen to now?

PR: Yeah, for instance, while stealing music from the 80s mostly like, it can be like Iron Maiden, very popular, or like more Los Angeles, kind of hard rock style. Then I was just playing with friends in some bands, this kind of garage bands, enjoying this kind of hobby. But I never became serious. I mean, of course, I didn't choose to go professional. Start music study and these things, as I was saying, I was also interested in literature and philosophy in high school. This was curious about, and also math. I've been always fascinated with math and in general, abstract thinking, this is something that typically I prefer, but for practical reasons, then I decided to go to engineering, just for assuming that it would give me better job opportunities. And then I discovered signal processing in this engineering. I enrolled in the telecommunication engineering program and then when I discovered signal processing, this basically applied math and applied Statistics. It's very abstract, but it's also it's a lot of implications in the practical world, then I fell in love with this topic. And then I've been mostly working with that all my time. And yeah, so this was what was the other question then? Why Norway?

SS: Norway for the weather or the lovely Norwegian ladies?

PR: Well, I will say that I don't mind and I really like the Norwegian weather. Although this might be uncommon, Norway was as many things in other topics in my life. It's been a little bit, maybe by chance, because I remember the first time it was with my girlfriend, but now it's my wife. Many years ago, we were looking for summer holidays planning in a travel agency. We went there and we were planning. We wanted to be fit in Stockholm and Helsinki. There was a big sports event. I don't remember exactly. I think if everything was fully booked, and then looking at the brochures and this thing, we looked at a picture of Bergen and my wife, she said: “Oh, that's the kind of place I want to visit”. So, we arranged it moving from Stockholm and Helsinki to visit Oslo, Bergen, the fjords and the strip on the fjords, and we went also to Flåmsbanen, this old railway. So that was just our summers. And that kind of holiday had a huge impact on our life because I was almost finishing my Ph.D. studies at that time. And so, looking for a postdoc, I was considering moving abroad. And as a child, I've always had the dream to finish my studies and then move to the US. But then after that, my wife really enjoyed Norway and these sceneries and that and of course, I'm in Scandinavia is quite popular for its high quality of life. We thought, Okay, why don't we apply for a postdoc position in Norway, and then I got the position at NTNU as a postdoc in 2006. We came to Trondheim for a couple of years. After that, I went back to Italy, because I got the positions that are in academia in Italy, but we have been still keeping in contact with NTNU and when visiting for projects and collaboration. And after a while, we decided, well, we miss Norway. And we like the way we were living here in Trondheim, especially Trondheim. We decided to change our lives and move back to Trondheim. And why Norway? The main thing I will say is the very nice quality of life. I mean, this is not just things that you read in the ranking and listings. In my opinion, you feel it when you're living here. It's just that ordinary life is very joyful, I will say. So, and of course, now we have children, even more. Especially in Trondheim that we consider a very children-friendly city. It's something that we will never change, I guess.
SS: Yeah, no, I agree. I'm originally from Yugoslavia and have lived in England, the US, and France in different places. But at the moment where I had to decide on where my kids are going to grow up. I thought that there is no better system in the world than in Norway. And I don't think people necessarily appreciate it if they've lived all their life in Norway. how well the system actually works.

PR: Exactly. As you said, this is exactly the place where we want to grow our children. I will say this is the main motivation.

SS: Yeah, yeah. A little ad for talents coming. I think. I believe that there is space for more. But Pierluigi. Tell us a few words about what does signal processing means. What picture should people have in their heads when they listen to you talk about signal processing.

PR: Yeah, more specifically, I work mainly with a branch of signal processing that is related to data fusion and statistical signal processing. And this is basically combining information from different sources that can be sensors or it can be another kind of source of information, combined this information to make reliable decisions. So, this is basically if I should summarize in one sentence, what I'm doing, this will be the sentence. And to be more a little bit concrete in simple ways. I was mentioning to you before this silly example, that I typically use in the first lecture in my course, in which I consider this silly game. Let's assume that we had like two jars in a different room that are filled with all marbles that can be red or blue. And then we asked a child to pick one marble from one of these two jars that we can call the alpha, beta, or whatever. These two names. He picks one marble from this jar, comes to our room, and chose this marble. And then we have to guess which one was the jar that he selected for picking the marble, just by not seeing. This is a sort of inference problem in which we have an observation of something related to something else that we don't have direct access to. And then we have to make a guess, how, based on knowledge, how do we do a reliable guess. This is basically my job, and this has implications in AI, in IoT, and in all the kinds of smart devices that are transforming our society into a smart society, if you want.

SS: So, if I now go from jars and red and blue balls to an oil platform. An oil platform might have some wave sensors, it might have some location sensors to know whether it's horizontal or at an angle, it might have some wind sensors, you might have some temperature sensors, and it could read all of the sensors and from that, try to infer if it's much at an angle.

PR: Yeah, for instance, I mean, this kind of processing may be targeting like optimization in the extraction in the use of resources. For instance, one of the projects that we are currently working on is for improving and leakage detection. For instance, you have the acoustic sensors that are detecting if that could be leakage and then by processing all this information, of course, when you have sensors that are detecting some events, sensors are not perfect so their information might be wrong. Simply because there is something failing or the risk of false alarm is to be more technical. And of course, so the point is how you combine all this information to make safer decisions and this could be, for instance, designing safer equipment that will reduce the number of useless maintenance for instance, and then we'll detect exactly when the problems are happening.

SS: One of the interesting things here is if we go back to industrial problems, or let's go back to the platform so they have these cables at the bottom of the sea that transfer gas or oil and that very bad idea to have a leakage there. To try to again put the picture in our heads we could measure the pressure in the cables we could maybe even do you measure the quality of the water. Could you discover that there is some oil coming out? Is it chemical or is it-

PR: What we are doing is mostly a more theoretical thing. Like my group we are not designing the sensors so assuming that there could be chemical sensors or temperature sensors, acoustic sensor or other kinds of sensor that people that are mostly working on sensor development is more on the electronics. They are the experts in the design so given the sensors that will then send something and collect information. What I'm doing with my research group is how to process this information more effectively. So that actually when they are telling an alarm, can you trust or not? Is it an alarm or something else? Can I detect in advance that an alarm will come in because now it's something strange going on from this measurement so I can expect that there is an anomaly if this is an anomaly behavior that could lead to a problem? These are the kind of things that I'm working on. Processing data from a system that someone else has been designing

SS: So, it's almost like you are creating the formula in a way that given we have these kinds of sensors, I can then in theory conclude or in real-time understand what's happening with the location or with the quality of service. Given this reality of sensors becoming cheaper and very, very much more available than just three years ago. I mean, now we can buy off-the-shelf sensors from China if we wish that were very recently just for the defense sector. So now you can actually create systems that could potentially be a little lab on a chip that don't control things that have to do with, let's say my blood or my urine or some health material from my body or they could be used in industrial production or they could be used for security in a city.

PR: Yeah, I mean, this dimension is very relevant now because this is partially the explanation of why machine learning and AI In general, this AI is receiving all this emphasis now. Because let's say, from my perspective, I think I'm mostly used to combine information. If we want to say more technically, this is applying statistics to combine this information reliably. Statistical functions. The nice thing is this is happening really in everything, even in our discussion now, this real-time discussion is happening because the device is able to make inferences on what I'm saying. What I'm saying is translated into an electrical signal that is corporate and then you are receiving this signal and based on the signal, this is transferred back into a voice-based on some statistical guess of the device that is just applying statistics in making this. This works because we can discuss in real-time, even if you are far away, right?

SS: I have to stop you, sorry. And I'm going to just be silly for a moment. In another example, where you can see this while you and I talk for those who are actually looking at the video. So Pierluigi is Italian and he uses his hands a lot. And with his hands, he can create a light show behind him. I don't know if you can see it on your own camera, I see it's very sunny in your room. But as you move your hands, the system is trying to optimize the light balance. Sometimes when you have your hands in front of the camera, the picture will go very light or very dark.

PR: Okay, maybe I should-

SS: No, no, you are you're looking perfect. And I actually am enjoying the effect. But I just wanted people to realize that that's another example of what you just were saying.

PR: Okay. It's this availability of sensors. This is nice because now a lot of information that we are gathering is actually coming, as you were saying, from the availability of a lot of sensors that are becoming cheap, we have a lot of devices that are deployed in our environments, and we have a lot of devices that are connected because now we have everything that is connected, everything is becoming a sensor speaking as a smart device. We have a huge amount of data that are collected from the real world that can process. And actually, now the processing capabilities are also becoming much cheaper. Now we just have a few $100 we can afford, what 10 years ago was or 20 years ago was the largest supercomputer in the world. Now, this can be everyone can afford with a few $100 and buying similar equipment. Even the processing capabilities now have been increased a lot. So, this is why now there is a lot of emphasis. It's not surprising that these AI and data-driven methodologies now are really seeing a nice summer, if you will, and they are flourishing because now we have all this data available, we can store and then we can process all this data. This partially explains why AI now has become so popular and so relevant, because actually, we have the possibility to exploit what data has to offer.

SS: Lots of data because of all of these sensors. And actually, I think that we have only seen the first five minutes of this summer in a way because I don't think we realize that we are going to be wearing many, many more sensors shortly. And that our houses are. We've been talking about smart houses, but it's only recently that we've started putting these sensors inside them. Lots of communicational power. Lots of networks, right?

PR: Yeah. There will still be a lot of things to come, although I guess there could be maybe a sort of misalignment between what is really coming and what is maybe the expectation of general people and common people. I typically use an example that was made. If you go back to some decades ago when this there was a lot of expectation about this if you want to call it robotic revolutions and a lot of like in the fiction and the general people were expecting these humanoid robots, what came out from them was mostly like washing machine, and all these kinds of equipment that are now popular in our home that is kind of absolutely useful and wonderful, but it's totally different from these humanoid robots that you could see in the fictions and these kinds of things. There is a kind of misalignment of what is really happening. And what are the science fiction and expectation around this is something similar that might happen with this AI in which maybe the expectation or more close to this, I don't know, general AI that will control maybe our life and something that will become extremely useful things that are much more practical and restricted to a specific application? At least I would expect this in the night in the recent in the next years in the short term.

SS: 100% agree with you. So basically, I'm a bit sick of the general AI discussion and the ethics of it, and so on. Because what I see is, especially in China, perhaps is this race to apply AI and applied it to health, to climate, to production, to transportation to many of the practical areas. And that said, I think that we humans, you know, we are a really bizarre creature, we can create technology that creates opportunities that are far beyond our current understanding. And I think there is always this lag between the potential and the real effects of technology. And I think that whoever manages to exploit that gap will also have enormously much to gain, which is why your research is super, super important. Where is this relevant? And, again, that said, I don't think we should underestimate the effect that, let's say, washing machines and fridges have had on our lives, or, or even on the deliberation of women. They end up doing something for us that we didn't anticipate, and it's in the management of these anticipations, that I think that the magic lies.

PR: Yes, exactly. In my opinion, is exciting, even if it's something completely different from maybe what was the expectation. It's nice that you said, I mean, you're a little bit sick of this kind of skepticism or criticism on the general AI. And another thing that I agree with you and another strange thing that you might also be a little bit sick of is this constant criticism like AI is a black box and something we cannot trust; we cannot apply it. This is also another strange thing to me. Maybe I shouldn't say sick when I'm listening to this, but I think it's not a fair use of fair criticism of AI when they think because if you want to take the fluttery black box, we move this to some other moving from machines and AI to humans. When we are inpatient as a user, in a hospital, or an airplane, we trust the pilot, we trust the doctor, although, for us, that's a total black box, we have no idea what he's doing, why is doing, but we trust, but what we trust is the training system that has led that person to be there. Because he has been undergoing some training school and other university and other things, we trust the training. If that person is there, we trust the training, and not even if I have no idea what it's doing and why it's doing, I trust the old system. And this is the same thing. The point the problem is not really that the machine is a black box or not. It's just a question, those machines should be able to work and explain what they're doing. And also, maybe so this is what is called explainable AI is a relevant topic. And finally, a fun topic that when I first listened to that, I was thinking it doesn't sound very interesting. But, after a while, I thought, well, this is something necessary, because of course, to have this trust in the system, we need to understand not only how this we need to trust the training process of that machine, we need to understand why this machine is taking decision. For instance, if we have two doctors that are not agreeing on something, they will be able to discuss, I'm failing this, because of that, this other guy is telling this because of something else, and then explaining their motivation they possibly able to converge. We have to provide the machine not only to make a decision but also to explain why they have taken this decision so that they could be a discussion, maybe with a human in case of conflict. And also, another relevant thing is assessing their level of confidence. I mean, taking a decision and then assessing I'm telling this is maybe the best way and I'm totally confident or this is the best way to do it, but this is an area that I'm not totally confident. So maybe some other discussion or some other information if needed. So explainable AI and assessment of confidence of uncertainty, I guess are two crucial research topics that hopefully will remove this black box criticism that sometimes in my opinion is used in a too shallow way.

SS: I completely agree. I think it's so interesting, you know when they made the self-driving car and you know the few accidents that they actually had, and so many people use them as you know the proof that this technology is not ready, and we are not comparing the accident numbers to you know, to what humans do. And humans are the ultimate black box I just finished the book on consciousness. It's super interesting for a guy like you. I'm going to send you the reference, but a very interesting model. That could be mechanized, actually. I'm thinking when it comes to this argument that we don't know, we can't explain exactly why the AI decided this or that. So we can't be unless we can have 110% explainable, we can't trust it, I think what we need to understand is what are the principles that are baked into the learning algorithm, as you say, as for the learning of the people, and then I think we need to stop using alternative excuses, like privacy, or like explainability or our lack of understanding of the ethics of this thing as innovation, breaks. Because I think that many of these things, we learn by applying and thinking very fast. Learning very fast.

PR: Yeah, I also have a similar attitude, I'm typically the person that likes to jump into something, and then maybe underestimating the risk and then discover, oh, but now we have to go for that.

SS: And then you learn very fast, and you have to fix very fast. But I think at the moment, we don't have enough perspective to anticipate all the future problems. But the way that we are working at the moment is, at least legally and politically, let's wait until we figured out all the risks and problems. And I think we're losing

PR: If this gap between technology, and then the legislation is increasing, this will eventually create even more problems, most likely, because I mean, I see the necessity that all the legislation and the rest should be trying to be as fast as possible as together with technology, just stopping or creating barriers is not working, will not work. And then as long as this gap will be increasing, I guess I just see this as a potential for new problems, not a solution.

SS: Yeah, my position on that is that creating barriers and waiting is, is only helping the others who are not. And unless you're in the game now, 10 years down the line, you will have even less influence over some sort of an end game. And the same thing with bias in data. You know, you said your sensors are not perfect. But well, you know, if we humans as creators of data, you know, are the main source of data in you know, previous employment history or previous legal case summaries, etc., we are not perfect either. Be careful because the data is biased towards men rather than women or something. Well, I guess our job is to recalibrate it now for the future, rather than say it's not useable data.

PR: Yeah, I totally agree.

SS: So, my question to you is, how do you apply for your work in practice? Do people come to you and say, I need the system? And help me think about how to solve it? Or do you say that, well, now this is possible? Does anybody want to try this? How can people get involved with your kind of work?

PR: Yes, well, I can say that maybe both things are possible. But some years ago, I will say that I was mostly working on theoretical aspects of this algorithm design. And then I was looking for collaboration for possible applications to apply these things that I was working on. I came back to NTNU now in 2019. For the third time, I've been moving in and out, you know, many times. And then after this experience, as a data scientist with Kongbsberg-group, and Kongsberg-group I've been mostly working on projects that are related to the energy and maritime domain and went back to NTNU, I've been trying to gain from that experience and working mostly in this domain. So now I've been maybe just for like, lucky coincidence or something. But it is more the other way that I have a lot of colleagues or people from other departments that are just calling me and then trying to well, we had these problems, can you help us in designing new algorithms that are solving problems and in this domain. That's why I have now a lot of collaboration in this energy domain in the maritime domain, and also moving a little bit towards also the civil engineering and construction building. Recently, I've been establishing some calibration in which again, you can imagine how the sensor data processing can be very useful in also construction building being for real-time monitoring. Every time there is something that makes sense to have real-time monitoring for an anomaly, so preventing problems, typically my expertise could be useful.

SS: So, if we just take a couple of concrete examples in maritime, it could be something that has to do with autonomous ships.

PR: Yes. And for instance, autonomous navigation, if you can imagine how it's relevant to understand this, there are objects and other things and then to help to have, let's say, what I'm doing is supporting what is called situational awareness so that the ship's as to have an understanding of what is the running abandonments to take a proper decision were to navigate waves to stop and this kind of thing so that we see the definition of objects and other dangers.

SS: And it has to read camera data, but it can also read the sea data

PR: Yeah, it could be that. It could be a camera leader or radar data. This is typically what we are using now in this I'm involved in one of the centers of excellence at NTNU on terminal shipping, which I think is very relevant. There are a lot of activities happening in Trondheim, relative autonomous ships. And I think this is a very exciting topic that can be something in which Norway could easily be worldwide leading. This is one of the topics that I consider very strategic for Norway.

SS: And in buildings, it could be, for example, sensors that read the temperature and find out what sudden temperature rises or falls or basically rooms that never get properly used because the temperature never rises.

PR: Another project that we want to get developing is not started yet. But we are proposing this also operations using satellite data for monitoring ground motion. And this could help in preventing landslides and these kinds of events.

SS: What kind of answers is that? You will take pictures from satellites, and you will be able to see if there is a movement of things that shouldn't be.

PR: Exactly. Yeah, this kind of thing. I'm not involved in the sensor designing. But once we have the understanding of what is the sensor measuring, then based on the measurements, we develop algorithms that are combining decisive information, these measurements to make decisions or make predictions.

SS: A very expensive part of these projects for the customers, until recently was putting the sensors in the system. Is there some sort of standardization going on there? And can you just say, well, if you have a sensor that can give me this kind of data, I really don't mind what sensor it is. And it's easier for them to get going. Now-

PR: Well, this is a critical part, I would say maybe the most critical so it's very nice that you pointed this question this point in the discussion because maybe this is the most critical part in a project. But this is also a part of the project in which the expertise from the people that are domain experts is really necessary because and this I also would like to emphasize this part. Because sometimes the research feeling in which like data science could be sort of threatening expertise of people in their specific domain like that could be in a medical domain or in the geoscience or something else. But this is absolutely not true. I mean, listening to people that say, okay, you just provide me with the data, and that will give you the answers, this is 99% of the time just will result in nonsense results or in something that is already known from ages in that domain. This is not the way to proceed. Having interaction with domain expertise and data scientists that are working together iteratively is necessary to have a successful project. And the first part is just this data collection. You need realistic data, that are reliable meaning that are not biased and that they are meaningful in a way that they represent possibly without bias and without other strange artifacts, the scenario that you are going to have in operation and there is no way out of involving expertise from that domain in building a proper data set that is meaningful. Understanding what does the relationship and what are the artifacts and what are the things that you can exploit. If I know that temperature and pressure in these things are particularly linked, or in another scenario, they are not and they can be neglected. This is a piece of information that will save me a lot of computation in my approach, and this is something that is typically coming from domain expertise. So, we typically as data scientists will rely a lot on the expertise of the people working in the domain. And finally, sometimes it's just not possible to have enough data you can think of, for instance, in the other water domain is not easy to place sensor underwater and collect a lot of data in different situations. So, you have to rely on simulations. And typically, simulations and these projects for simulation are developed by people, again, that are experts in their domain or you know, they can have expertise in thermodynamics if this is a process engineering or they can have expertise in maritime engineering or something else. You really need this interaction with people from that domain. So, data science is not threatening the expertise in any domain is just enlarging the possibilities. The main effect is just providing this group instead of having much more resources and 10 times more experts working full time on the topic, so enabling their power. This is basically what is not assigned, but it's not replacing anyone.

SS: It is, in a way, giving them extra horsepower for the manual processes that they have to do before. And I'm thinking of an example, before AI, we had this, you know, we had the mobile and the internet revolution. But even before that, we had the spreadsheet revolution. And I remember when Excel came up, and then people were saying now there's no more work for accountants because you know, the system will do everything for you. And yet they became more useful and more necessary in larger numbers than ever before, because their subject evolved, it became much more complex. And so, you needed people who know the, you know, the legalities of accounting, and then you had these advanced tools that would help them develop and deliver. And I think something similar is happening in health, in engineering in many, many industries. But these are necessary tools for tomorrow, you can't say I don't want to use them.

PR: Exactly what and this tool is, you say it's important, I mean, this will not replace the need of people will just make the need, the people will need to provide something different, maybe at a higher level or in more complex, but you will still need people like in the with legal expertise, or with medical expertise, or with the thermodynamic expertise or whatever. So there, there will be no replacement of this kind of expertise, it's just the death task will change. You need people that are used to be able to change what is their task. And this is why I stress that what is important, the most important thing to learn is not a specific topic or something. But the best thing to invest in is the capability to learn new things, fast and deep. This is what will always be needed. Because we know it's, even more, we cannot be expected to be doing the same thing for all our working life, similar tasks, but we need to adapt to a new way of working and new things and a new way to applying our expertise. So, the ability to learn, new things will be crucial, is even more than before.

SS: This is music in my ears, of course, given what the mission of Lørn is, and it's actually to make people want to learn new stuff, and also kind of dare to say that, you know, what you're doing Pierluigi, is not just for students at NTNU. I think as you said, there are many, many people both in Kongsberg, but also in many other engineerings, building, shipping, energy companies that will need to start thinking in these terms. And then my question is, how do they get started? I guess it's through projects that we have to learn. you recommend the book called statistical signal processing. But the little I remember from that kind of literature and you have the estimation theory and detection theory. It's not exactly kind of summer. poolside reading, right.

PR: Yeah. Yeah, that's right. But I mean, I guess now really, I mean, if someone is curious about things, you have a lot, you're overwhelmed with a lot of different opportunities at different levels. You can focus on books, old classical style, you can look at lectures on the web, you have seminars, and then you have even classes online from MIT that you can find on YouTube that are so there are really a lot of I mean, I guess each one can find the proper way that feels more suitable for his level of entrance and find the proper way, but I guess having this curiosity to learn more and to be exposed to this new domain in which there will be this interaction between, let's say classical topics and collaborating with data science is really something worth investing in my opinion because I cannot see that this will. I mean, I cannot imagine any sector that will not be affected by this kind of interaction of data science together with that specific topic.

SS: Very interesting. Yeah, I asked, I asked you if you have a quote that you could leave with our listeners as a little parting gift and

PR: I was selecting one by Woody Allen which is "Confidence is what you have before we understand the problem". I like this because it's just somehow, I'm like, crazy in this perspective. I mean, I think people should jump with enthusiasm into new things. Even if they might seem scary at the beginning once you've jumped but once he jumps, you need to swim but then you will enjoy swimming and because any kind of new challenge will bring you some kind of adrenaline shot maybe but you will find a lot of things to enjoy and at least for me when I'm learning new things I'm always enjoying. I go back to your first question about how you will define yourself maybe I like to feel like a student All my life I mean I always seem looking for new things to learn. Having this feeling of being a student that I'm learning something is maybe something that I'm really looking in every aspect and day of my life, so the feeling of being a student That is something that in my opinion is very nice because this is just to be willing to learn something new.

SS: I have this vision both for myself and for Lørn, to be honest, mix match and never stop when it comes to new knowledge. And I think you also mentioned at some point that you really like reading classical books as well. I turned 50 about a year ago and in the last two three years, I've started rereading and reading for the first time some of the classics and I think that being able to do that. At the same time do something that's more mathematical or you know the business. It's such a good therapy actually, for your brain that I think we all should do more of that

PR: In general diversities usually always bringing some positive aspects to everything. I really believe this also in a small sense, even when I'm focusing on my legal research area and world. I typically prefer to have at least two different projects that are different at the same time. If you just focus on one thing in your life, I mean this sometimes you just get bored or sometimes you just get stuck with something and having the opportunity to just have a break and move to something else. And then going back then you will look at that with new eyes and things typically will proceed faster and even better. And this is in general. This is just now in my personal attitude to research but I guess you can have a lot of diverse examples in every aspect of life where personnel in research groups in other activities and I cannot see any real bedside apart from you need to be able to manage a little bit more complex when you are including diversity, but I guess the benefit will are so much more than this little overhead that is always good to increase the level of diversity in any kind of dimension in your aspect of life.

SS: And that's a good statistician speaking, I guess. Pierluigi, thank you so much for joining us here in this Lørn chat for teaching us the statistical signal processing new school

PR: Thanks a lot for inviting me. Hope it was useful to the listener.

SS: I guarantee it was.Du har nå lyttet til en podcast fra Lørn.Tech - en læringdugnad om teknologi og samfunn. Nå kan du også få et læringssertifikat for å ha lytte til denne podcasten på vårt online universitet på Lørn.University.

Read full transcript

Who are you and how did you become interested in AI and the technology around it? 
I would say I am a person with a good level of curiosity and a desire to learn. At school, I was mostly interested in math, philosophy, and literature. Then I decided to enroll in engineering, assuming it would have given me better job opportunities. In my third year, I discovered the world of “Signal Processing” and I loved it … it is basically applied math/statistics, and it is behind most of the technology we use every day. 

What is the most important thing you do at work? 
Research: includes writing project proposals for research funds (also in collaboration with research institutions and industries), developing and validating new methodologies, coordinating the work of other researchers (including supervision of doctoral students), disseminating the results through scientific publications. 

Teaching: I am currently responsible for introducing the basic concepts of statistical signal processing (estimation, detection, and classification) to the students in the study program Electronic Systems Design and Innovation at NTNU. 

 What do you focus on in technology/innovation? 
I have long experience with data fusion in sensor networks, mostly algorithms design, and performance evaluation. Recently, I am focusing more on sensor-data processing for monitoring applications mostly in the energy and maritime domain using both model-based and data-driven approaches. 

Why is it exciting? 
It has a good level of abstraction (I like math ) and also a direct impact on practical applications. Basically, it offers a large variety of opportunities from theoretical modeling to experimental research. 

What do you think are the most interesting controversies? 
If I should select one, maybe the skepticism towards AI/ML (Artificial Intelligence/Machine Learning) approaches, often quickly and unfairly labeled “black-box” with negative emphasis. XAI (Explainable AI) and uncertainty quantification will be crucial for the deployment of AI/ML-based solutions in safety-critical applications. 

What do you think is relevant knowledge for the future? 
As (in my opinion) it has always been: the ability to learn (possibly fast and deep). Assuming that literacy nowadays includes basic knowledge of digital tools, then the ability to acquire and master new knowledge remains the most important issue. Such ability can be developed by means of any topic, e.g. studying literature, math, music, philosophy, etc. 

What do we do uniquely well in Norway within AI? 
There are many talents in various AI-related fields, so naming one field would be unfair to many others. However, I would like to stress again the strategic relevance of Norwegian research in the maritime domain. To make an explicit example, even though it is relevant to contribute to the research enabling autonomous cars, I assume that no one expects that Norway will be leading this sector. Differently, Norway has good chances to be the world leader in autonomous ships. 


Samle deg med en venn eller en kollega for å se om du klarer å svare på spørsmålet nedenfor.


Is the saying “AI is a black box” a fair criticism or not? Explain!  


Want to show off this case to your friends and coworkers?

Download summary (Available soon)

This is what you will learn:

Machine Learning

Confidence is what you have before you understand the problem

- Pierluigi Salvo Rossi

Recommended literature:

Fundamentals of Statistical Signal Processing:  
Volume I (Estimation Theory) 
Volume II (Detection Theory) by S.M. Kay (1993) 
Data Fusion in Wireless Sensor Networks:  
A Statistical Signal Processing Perspective by D. Ciuonzo and P. Salvo Rossi (2019)

This is NTNU

NTNU is an internationally oriented university with headquarters in Trondheim and campuses in Gjøvik and Ålesund.