DS
D.D.D.T.D. Sitaldin
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
2 records found
1
Relevance Detection of Unknown Classes through Cluster Distances
Based on Statistical Distance Measures in Feature Space
In the open world, machine learning (ML) models can encounter a multitude of unknown or novel classes. In a surveillance, safety, or security use case, unknown samples can pose potential threats that are hard to detect since those samples have never been trained on. At the same time, most of the unknowns that will be encountered by a surveillance ML model will be harmless. This results in too many unwanted alerts and manual analyses, of harmless unknowns that have been flagged.
Through this thesis, for the first time (to the best of our knowledge), a method is developed that can automatically assess the relevance of unknown classes, by modelling their image features as clusters (or distributions) and comparing them using statistical distance measures. Our use case lies in computer vision for military applications, where based on the user input, relevance is defined. We define road vehicles as relevant classes and use those for our training set. Our aim is to build a model that can successfully classify new unseen road vehicles as ‘relevant unknowns’, while also successfully classifying harmless unknown birds that are not part of the training set, as ‘irrelevant unknowns’. On the DomainNet data-set, we demonstrate that our novel method can very accurately determine the relevance of unknown classes at test time for both low and high-dimensional data, with AUC scores ranging from 0.99 to a perfect 1.00. ...
Through this thesis, for the first time (to the best of our knowledge), a method is developed that can automatically assess the relevance of unknown classes, by modelling their image features as clusters (or distributions) and comparing them using statistical distance measures. Our use case lies in computer vision for military applications, where based on the user input, relevance is defined. We define road vehicles as relevant classes and use those for our training set. Our aim is to build a model that can successfully classify new unseen road vehicles as ‘relevant unknowns’, while also successfully classifying harmless unknown birds that are not part of the training set, as ‘irrelevant unknowns’. On the DomainNet data-set, we demonstrate that our novel method can very accurately determine the relevance of unknown classes at test time for both low and high-dimensional data, with AUC scores ranging from 0.99 to a perfect 1.00. ...
In the open world, machine learning (ML) models can encounter a multitude of unknown or novel classes. In a surveillance, safety, or security use case, unknown samples can pose potential threats that are hard to detect since those samples have never been trained on. At the same time, most of the unknowns that will be encountered by a surveillance ML model will be harmless. This results in too many unwanted alerts and manual analyses, of harmless unknowns that have been flagged.
Through this thesis, for the first time (to the best of our knowledge), a method is developed that can automatically assess the relevance of unknown classes, by modelling their image features as clusters (or distributions) and comparing them using statistical distance measures. Our use case lies in computer vision for military applications, where based on the user input, relevance is defined. We define road vehicles as relevant classes and use those for our training set. Our aim is to build a model that can successfully classify new unseen road vehicles as ‘relevant unknowns’, while also successfully classifying harmless unknown birds that are not part of the training set, as ‘irrelevant unknowns’. On the DomainNet data-set, we demonstrate that our novel method can very accurately determine the relevance of unknown classes at test time for both low and high-dimensional data, with AUC scores ranging from 0.99 to a perfect 1.00.
Through this thesis, for the first time (to the best of our knowledge), a method is developed that can automatically assess the relevance of unknown classes, by modelling their image features as clusters (or distributions) and comparing them using statistical distance measures. Our use case lies in computer vision for military applications, where based on the user input, relevance is defined. We define road vehicles as relevant classes and use those for our training set. Our aim is to build a model that can successfully classify new unseen road vehicles as ‘relevant unknowns’, while also successfully classifying harmless unknown birds that are not part of the training set, as ‘irrelevant unknowns’. On the DomainNet data-set, we demonstrate that our novel method can very accurately determine the relevance of unknown classes at test time for both low and high-dimensional data, with AUC scores ranging from 0.99 to a perfect 1.00.
Bachelor thesis
(2019)
-
Dewwret Sitaldin, David de Best, Salar Mollaie Daryani, Van Cong Nguyen, Joey Gelderblom, Joris van Roessel, Peter van Duijsen
The main goal of this Electrical Engineering Bachelor project is to build a solar-power system for a quad-copter that will extend its battery life or rather its flight time. The complete system is comprised of a PV system (PV), a micro-controller (MC) and a DC/DC converter (DC) which was mounted onto the drone. On each subsystem, a separate thesis was written and this paper serves as a general yet complete overview of the design process, simulations and test results of a fully functioning solar drone with the theses attached as appendices for reference. The original (optimistic) aim of an extension of at least 25% of the battery lifetime was set by our supervisors. For the PV part SunPower C60 IBC cells were used (no specific selection was done) together with a (borrowed) custom-built drone (not built by this team, it was borrowed from another research group) as a starting point. After analysing the limitations of the drone and the cells, multiple configurations were designed and a mathematical model that determines power usage, energy costs per solar cell and the optimum amount of cells was developed. %The optimum amount of cells for this specific drone was found to be a total of 28 cells. A SEPIC converter will extract solar energy from a PV-module in order to charge the battery of the drone. The converter will be controlled by the micro-controller subgroup using MPPT (Maximum Power Point Tracker) algorithm and this will be done by supplying a PWM signal to the converter. Since the drone was not specifically designed for the project (thus not optimised when it comes to lift capacity and room for cell placement), the efficiency of the solar cells was not sufficient to extend the fight time by 25% (15.1% in summer, 5.6 in winter). Since these bottlenecks can easily be eliminated by replacing the drone and the cells, these results serve as a proof of concept and are an excellent starting point for future research.
...
The main goal of this Electrical Engineering Bachelor project is to build a solar-power system for a quad-copter that will extend its battery life or rather its flight time. The complete system is comprised of a PV system (PV), a micro-controller (MC) and a DC/DC converter (DC) which was mounted onto the drone. On each subsystem, a separate thesis was written and this paper serves as a general yet complete overview of the design process, simulations and test results of a fully functioning solar drone with the theses attached as appendices for reference. The original (optimistic) aim of an extension of at least 25% of the battery lifetime was set by our supervisors. For the PV part SunPower C60 IBC cells were used (no specific selection was done) together with a (borrowed) custom-built drone (not built by this team, it was borrowed from another research group) as a starting point. After analysing the limitations of the drone and the cells, multiple configurations were designed and a mathematical model that determines power usage, energy costs per solar cell and the optimum amount of cells was developed. %The optimum amount of cells for this specific drone was found to be a total of 28 cells. A SEPIC converter will extract solar energy from a PV-module in order to charge the battery of the drone. The converter will be controlled by the micro-controller subgroup using MPPT (Maximum Power Point Tracker) algorithm and this will be done by supplying a PWM signal to the converter. Since the drone was not specifically designed for the project (thus not optimised when it comes to lift capacity and room for cell placement), the efficiency of the solar cells was not sufficient to extend the fight time by 25% (15.1% in summer, 5.6 in winter). Since these bottlenecks can easily be eliminated by replacing the drone and the cells, these results serve as a proof of concept and are an excellent starting point for future research.