Dyrekcja Instytutu Fizyki UŚ oraz Polskie Towarzystwo Fizyczne mają zaszczyt i przyjemność zaprosić na wyjątkowe sympozjum naukowe „Fizycy w Świecie SI: Przekraczanie Granic Dyscyplin”, która odbędzie się w dniu 12 grudnia 2024 roku w Kampusie Chorzowskim Uniwersytetu Śląskiego, ul. 75 Pułku Piechoty 1, 41-500 Chorzów w auli P/0/05.
Konferencja ta jest dedykowana najnowszym osiągnięciom z zakresu fizyki oraz sztucznej inteligencji, a jej celem jest zainspirowanie i zintegrowanie środowisk naukowych oraz technologicznych. Wydarzenie będzie doskonałą okazją do zaprezentowania, jak umiejętności i wiedza fizyków przyczyniają się do dynamicznego rozwoju sztucznej inteligencji w różnych dziedzinach, w tym w medycynie i optymalizacji procesów przemysłowych.
Szczególną atrakcją tegorocznego wydarzenia będzie segment poświęcony laureatom Nagrody Nobla w dziedzinie fizyki z 2024 roku, którzy zostali wyróżnieni za przełomowe badania w obszarze komputerów kwantowych oraz ich zastosowań w sztucznej inteligencji. Ich rewolucyjne prace otworzyły nowe horyzonty zarówno w fizyce teoretycznej, jak i stosowanej, stwarzając niespotykane dotąd możliwości rozwoju technologii SI. Ich osiągnięcia będą omawiane podczas konferencji, ukazując, jak te innowacje zmieniają różnorodne branże.
Prelegentami będą fizycy pracujący zarówno na znakomitych uniwersytetach polskich i zagranicznych, jak i w sektorze prywatnym. Przewidujemy wykłady w języku polskim i angielskim.
Aby uczestniczyć w wydarzeniu w formie stacjonarnej, należy się zarejestrować.
Wydarzenie będzie emitowane na kanale Youtube: kliknij tutaj
9:30 – 10:15 dr Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, wykład w języku angielskim
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that enable autonomous pattern recognition, influencing developments from image recognition to the transformative potential of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in materials science, particle research, and other fields. The speaker will also share insights from his own work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics efforts may shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 dr Anna Dawid, Wyjaśniać czy interpretować? Czyli jak uczyć się (fizyki kwantowej) od sieci neuronowych
Głębokie uczenie maszynowe rewolucjonizuje przemysł i obiecuje zmienić oblicze nauki. Ma jednak liczne ograniczenia, z których najistotniejszym z punktu widzenia rozwoju nauki jest to, że sieci neuronowe są „czarnymi skrzynkami”. Taka konstrukcja uniemożliwia nam zrozumienie jak dochodzą do podawanych wniosków czy odpowiedzi, co sprawia, że nie możemy się od nich uczyć. W tej prezentacji omówię różne ograniczenia sieci neuronowych, a także sposoby, by „otworzyć” te czarne skrzynki za pomocą metod wyjaśniania modeli uczenia maszynowego. Pokażę też na przykładzie prostego problemu z fizyki jak można budować sieci o interpretowalnej konstrukcji, przez zmuszanie ich do posługiwania się „językiem”, który rozumiemy.
11:05 – 11:35 dr Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, wykład w języku angielskim
In this presentation, I will highlight how the skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”.
Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed matter physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can we neglect? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step which allows to fully exploit the computational power that modern artificial intelligence frameworks offer us.
11:45 – 12:15 dr Emil Kaptur, Użycie metod uczenia maszynowego w optymalizacji efektywności paliwowej linii lotniczych na podstawie danych z rejestratorów szybkiego dostępu (tzw. czarnych skrzynek)
StorkJet zajmuje się tworzeniem oprogramowania wspierającego linie lotnicze w zwiększeniu efektywności paliwowej oraz zmniejszeniu emisji CO2 oraz pozostałych czynników wpływających na klimat. Operacje lotnicze są skomplikowane. Na szczęście, można je rozważać jako sumę prostszych składowych, w których decyzje dyspozytora, pilota, autopilota, itd. wpływają w różny sposób na spalanie paliwa i osiągi samolotu. StorkJet stworzył modele uczenia maszynowego, które opisują prawie każdy aspekt operacji lotniczych. Użycie tych modeli w produktach StorkJet pozwala linii lotniczej na określenie optymalnej procedury wykonania operacji oraz oszczędności związanych ze zmianą procedur. Podczas prezentacji StorkJet przedstawi:
- Stworzenie prostego modelu uczenia maszynowego. Od odczytu „czarnych skrzynek”, do (niezbyt dokładnego) wyznaczenia trybu ciągu użytego podczas wznoszenia samolotu.
- Symulacja wznoszenia samolotu. Jak zmieni się zużycie paliwa jeśli pilot użyłby innej prędkości wznoszenia.
- Symulacja kołowania na jednym silniku używająca modeli uczenia maszynowego do przewidywania zachowania pilota.
12:25 – 12:55 dr hab. Marcin Kostur, prof. UŚ, Zobaczyć niewidzialne z pomocą głębokiego uczenia: o rekonstrukcji zastawki aortalnej w tomografii komputerowej bez kontrastu.
Dokładna segmentacja zastawki aortalnej (AV) w tomografii komputerowej (CT) bez kontrastu jest kluczowa dla oceny nasilenia chorób AV oraz identyfikacji pacjentów, którzy mogą skorzystać z interwencyjnych zabiegów chirurgicznych. Jednakże, niska widoczność AV w tego rodzaju obrazach medycznych stanowi poważne wyzwanie. W niniejszym wykładzie przedstawimy innowacyjną metodę semi-automatycznego generowania danych referencyjnych (Ground Truth) opartą na rejestracji obrazów, która umożliwia trenowanie modeli sieci neuronowych w procesie słabego nadzoru, zdolnych do precyzyjnej segmentacji AV jedynie na podstawie obrazów CT bez kontrastu. Ponadto, zaprezentujemy nowatorskie podejście do oceny dokładności segmentacji, polegające na rejestracji sztywnych masek segmentowanych w obrazach kontrastowych i bez kontrastu dla każdego pacjenta. Wyniki przeprowadzone na otwartym zbiorze danych wykazują, że nasz model potrafi zidentyfikować AV z średnim błędem mniejszym niż 1 mm, co wskazuje na znaczny potencjał zastosowań klinicznych.
dr Stuart Gibson został mianowany na stanowisko wykładowcy w Szkole Fizyki i Astronomii na Uniwersytecie w Kent w 2007 roku. Jest współtwórcą systemu kompozytowego twarzy EFIT-V, który jest obecnie używany przez większość brytyjskich jednostek policji i w wielu innych krajach. Głównym tematem badań dr Stuarta Gibsona są zastosowania przetwarzania obrazu cyfrowego i uczenia maszynowego w kryminalistyce.
dr Anna Dawid jest jest naukowczynią zajmującą się fizyką kwantową i uczeniem maszynowym w Leiden University, a także entuzjastką teatru i gier. Chętnie bawi się interpretowalnym uczeniem maszynowym w nauce, ultrazimnymi platformami do symulacji kwantowych i teorią uczenia maszynowego. Jest pasjonatką kształtowania zautomatyzowanych podejść w unikalną nową soczewkę naukową i patrzenia przez nią na ustalone trudne problemy kwantowe.
dr Ciro Taranto jest liderem ds. optymalizacji i uczenia maszynowego w HYDROGRID, znanym ze swojej pasji do liczb i tłumaczenia rzeczywistych problemów na modele matematyczne i kod. Dzięki doświadczeniu w uczeniu maszynowym i sztucznej inteligencji wnosi analityczną precyzję i pragmatyczne podejście do rozwiązań. Ciro jest fizykiem, oddanym programistą Pythona, komunikatorem naukowym i członkiem zespołu, którego napędza ciekawość i determinacja.
dr Emil Kaptur jest szefem działu badań i rozwoju w StorkJet. Posiada tytuł doktora fizyki eksperymentalnej, a jego wiedza specjalistyczna w zakresie analizy danych i uczenia maszynowego pochodzi z jego poprzedniej pracy w Europejskiej Organizacji Badań Jądrowych (CERN). Obecnie kieruje zespołem badaczy uczenia maszynowego o zróżnicowanym doświadczeniu, rozszerzającym modele wydajności StorkJet, aby obejmowały nie tylko wydajność w locie, ale wszystkie obszary operacji lotniczych, w których można oszczędzać paliwo.
dr hab. Marcin Kostur, prof. UŚ to wykładowca akademicki i naukowiec, którego badania obejmują szeroki zakres tematów. Jego najnowsze prace naukowe skupiają się na zagadnieniach zastosowań fizyki w medycynie i analizie obrazów medycznych.W swoich ostatnich publikacjach, badał między innymi transport lipoprotein niskiej gęstości w ścianach tętnic oraz wykorzystanie głębokiego uczenia w interpretacji i modelowaniu procesów istotnych w diagnostyce chorób serca. Jest autorem rozwiązań technologicznych wykorzystujących uczenie maszynowe w analizie obrazów tomografii komputerowej.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.
9:30 – 10:15 Dr. Stuart Gibson, Physics-Driven AI and AI-Driven Physics, Exploring a Symbiotic Evolution, lecture in English
This lecture examines the intersection of physics and artificial intelligence (AI), inspired by the pioneering contributions of 2024 Nobel Laureates John Hopfield and Geoffrey Hinton. We begin with an accessible introduction to machine learning and AI fundamentals, setting the stage for understanding for the Nobel-winning advancements in neural networks and associative memory. By drawing on principles from physics, Hopfield and Hinton created adaptive models that autonomous enable pattern recognition, influencing developments from image recognition to the transformative of deep learning.
The presentation will showcase AI’s impact within physics, highlighting applications in science, research, and other fields. The speaker will also share insights from own his work in chemometrics and AI’s role in the natural sciences. Concluding with a forward-looking perspective, we will explore how collaborative AI-physics may feel shape the future of scientific discovery and lead to new breakthroughs.
10:25 – 10:55 Dr. Anna David, Explain or interpret? How to learn (quantum physics) from neural networks
Deep machine learning is revolutionizing industry and promises to change the face of science. However, it has numerous limitations, the most important of which from the point of view of the development of science is that neural networks are “black boxes”. Such a design makes it impossible for us to understand how they reach their conclusions or answers, which means that we can’t learn from them. In this presentation, I will discuss the different limitations of neural networks, as well as ways to “open” these black boxes using methods to explain machine learning models. I will also show you on the example of a simple problem in physics how you can build networks with an interpretive design, by forcing them to use the “language” that we understand.
11:05 – 11:35 Dr. Ciro Taranto, Visualize, Predict, Plan: Synergies between Physics and Artificial Intelligence applied to Hydropower, lecture in English
In this presentation, I will highlight how skillset built as a researcher in theoretical physics has helped me to shape my motto in artificial intelligence: “Visualize, Predict, Plan”. Visualizing very abstract quantities is fundamental to interpret the results of complex simulations in condensed physics, but is also a key part of exploratory data analysis in machine learning. Predicting, one of the “standard” tasks of machine learning, can benefit from the art of approximating mathematical models in theoretical physics: which relations can be assumed to be linear? Which quantities can be in the ignore? Finally, in order to plan actions in the real world one needs to be able to translate real-life problems into mathematical models: this is a fundamental step that fully exploit the computational power that modern intelligence frameworks offer us.
11:45 – 12:15 Dr. Emil Kaptur, The use of machine learning methods in optimizing the fuel efficiency of airlines based on data from high-speed access loggers (so-called black boxes)
StorkJet develops software that supports airlines in increasing fuel efficiency and reducing CO2 emissions and other climate impacts. Air operations are complicated. Fortunately, they can be considered as the sum of simpler components in which the decisions of the dispatcher, the pilot, autopilot, etc., affect in different ways the fuel and the performance of the aircraft. StorkJet has created machine learning models that describe almost every aspect of aeronautical operations. The use of these models in StorkJet products allows the airline to determine the optimal procedure for the execution of operations and savings related to changing procedures. During the presentation, StorkJet will present:
- Creating a simple model of machine learning. From the reading of the “black boxes”, to (not very accurate) determination of the thrust mode used during the aerial erect.
- Simulation of the aerial erect. How fuel consumption will change if the pilot used a different ascension speed.
- Simulation of taxiing on a single engine using machine learning models to predict the behavior of the remote control.
12:25 – 12:55 dr hab. Marcin Kostur, prof. U.S. See invisible with deep learning: about aortic valve reconstruction in computed tomography without contrast.
The exact segmentation of the aortic valve (AV) in computed tomography (CT) without contrast is crucial for assessing the severity of AV diseases and identifying patients who can benefit from interventional surgical procedures. However, low AV visibility in this type of medical image is a major challenge. In this lecture, we will present an innovative method of semi-automatic reference data generation (Ground Truth) based on image recording, which allows you to train neural network models in a poor surveillance process capable of precise AV segmentation only on the basis of CT without contrast. In addition, we will present a novel approach to assessing the accuracy of segmentation, consisting in the recording of rigid masks segmented in contrast images and without contrast to each patient. The results in an open dataset show that our model can identify AV with an average error of less than 1 mm, indicating significant potential for clinical applications.
Dr Stuart Gibson was appointed to the position of Lecturer in the School of Physics and Astronomy at the University of Kent in 2007. He is the co-inventor of the EFIT-V facial composite system which is currently used by the majority of UK police constabularies and in numerous other countries. The main theme of Dr Stuart Gibson’s research is forensic applications of digital image processing and machine learning. Specific areas of expertise include artificial intelligence in the natural sciences, facial composites for use in criminal investigations, medical image analysis, computer vision with security applications.
Dr. Anna Dawid is a quantum physics and machine learning scientist at Leiden University, as well as an enthusiast of theatre and games. She is happily playing with interpretable machine learning for science, ultracold platforms for quantum simulations, and the theory of machine learning. She is passionate about molding automated approaches into a unique new scientific lens and looking through it at established difficult quantum problems.
Dr. Ciro Taranto is the Optimization & Machine Learning Lead at HYDROGRID, known for his passion for numbers and translating real-world problems into mathematical models and code. With expertise in machine learning and AI, he brings analytical precision and a pragmatic approach to solutions. Ciro is a dedicated Python developer, scientific communicator, and team player, driven by curiosity and determination.
Dr. Emil Kaptur is Head of Research and Development at StorkJet. He holds PhD degree in experimental physics and his expertise in data analysis and machine learning comes from his previous work at European Organization for Nuclear Research (CERN). Currently, he leads a team of machine learning researchers with diverse backgrounds extending StorkJet’s performance models to cover not only in-flight performance, but all areas of aircraft operations where fuel can be saved.
Dr. hab. Marcin Kostur, prof. UŚ is an academic lecturer and researcher whose research covers a wide range of topics. His latest scientific papers focus on the issues of applications of physics in medicine and analysis of medical images.In his recent publications, he studied, among others, the transport of low-density lipoproteins in the walls of the arteries and the use of deep learning in the interpretation and modelling of processes relevant to the diagnosis of heart diseases. He is the author of technological solutions using machine learning in the analysis of CT images.