Master Thesis “cnn Classifier Configurations for

Vor 4 Tagen

Wien, Österreich Austrian Institute of Technology Vollzeit

Master Thesis “CNN classifier configurations for severely imbalanced multi-class industrial segmentation tasks”
- Center for Vision, Automation & Control
- **Classification and segmentation of images** is an effective way to **monitor product features in industrial inspection** and has especially gained momentum with the rise of neural networks and deep learning over the past few years. Convolutional neural networks (CNNs) have demonstrated great success in semantic segmentation of images into multiple classes (e.g., differentiating distinct product defect classes). However, if during training of CNNs the different classes are presented to the network in imbalanced instead of equal proportions, their performance generally worsens. In industrial inspection, the natural occurrence of different product defects is rarely balanced and often one defect class dominates while other classes are rare.
- There are **several common ways to deal with imbalanced data**, such as adapting the loss function to adjust for the over
- The aim of this master’s thesis project is **a quantitative comparison of the performance of a state-of-the-art segmentation network **architecture and alternative ones that have potential to better represent minority classes in imbalanced datasets. Explicitly**, the decrease of performance of a CNN for a multi-class segmentation task will be investigated** on an industrial dataset in conditions when the dataset becomes increasingly imbalanced.
- During the master’s thesis project you will **investigate properties of CNNS in the context of image segmentation**. The goal is to **analyze the performance of a vanilla segmentation network** that is trained on an imbalanced dataset and compare it to the performance of an alternative network architecture that is expected to better represent minority classes. The specific research hypothesis is that the segmentation performance and generalization capabilities of binary-classifier-networks are better when compared to multi-class network if the data is severely imbalanced (see below).
- For the analysis, you will use **an industrial dataset**. The network architecture and training code will be written in Python using PyTorch and training will be performed on a designated GPU cluster. The activities will include **writing code for optimal data handling** (real industrial data provided), **for training and inference of the neural network for the respective experimental setups** and **evaluating the results**.
- **The specific aims and expected outcome of this project are**:

- (1) As a first step, under the guidance of our team, you will evaluate the performance of the segmentation network. For this purpose, code for a data handling routine will be written that can feed data to the network in a balanced way or with a predefined imbalance. In this case, one expects the network performance to decrease with increasing imbalance.
- (2) In the second step you will investigate how the use of alternative loss functions can counteract the decrease in performance for imbalanced data.
- (3) Third, you will compare the performance of the segmentation network from specific aim 1, which is a multi-class classifier with the performance of n binary-classifier-networks, where n is the number of classes predicted by the segmentation network in specific aim 1 and the n binary-classifier-networks are trained only on one class versus all other classes. The expected result is that the performance of the binary-classifier-networks is better when compared to the multi-class network if the data is severely imbalanced and that the performance gain during training is more monotone (we expect performance fluctuations during training of the multi-class network).
- (4) The final investigation will reduce the huge architecture of n binary-classifier-networks to a single network with a shared backbone and n classification heads, where each classification head is only trained on one of the n classes. This network architecture drastically reduces the number of parameters and, thus, memory consumption and training time with respect to n binary-classifier-networks with expectedly comparable segmentation performance.
- The results of this work will include an **in-depth analysis of the properties of neural networks and the boundaries of their capabilities in an industrial contex**t. The project will give insights into the possibilities of semantic image segmentation with convolutional neural networks in the case that the dataset is imbalanced, and where it is difficult or not possible to synthetically balance the data. A detailed understanding of the performance of different neural network architectures in this context can have a large impact on the success in dealing with industrial data.
- By engaging in this project, **you will gain expertise in data science** including the **evaluation of image-based data**.
- You **will dive deep into the potentials and pitfalls of deep learning*

  • Master Thesis “deep Learning for Scalable

    vor 2 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and difitalisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis “representation of Stochastic

    Vor 3 Tagen

    Wien, Österreich Austrian Institute of Technology Vollzeit

    Master Thesis “Representation of Stochastic Phenomena in Optimizing Energy Management Systems” Master Thesis “Representation of Stochastic Phenomena in Optimizing Energy Management Systems” - Center for Energy - Energy Management Systems (EMS) in the context of local energy communities manage local flexibility such as storage systems and charging...

  • Master Thesis „artificial Intelligence for

    Vor 4 Tagen

    Wien, Österreich Austrian Institute of Technology Vollzeit

    Master Thesis „Artificial Intelligence for Airborne Phenotyping” - Center for Vision, Automation & Control - Our team’s expertise is the development and deployment of assistive and autonomous systems, particularly unmanned aircrafts und land-based mobile machines. - We cooperate with national and international research organizations and companies. -...

  • Master Thesis “enhancing Experience in Chatbot

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation for applied research**, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis “mixed Reality Algorithms”

    Vor 4 Tagen

    Wien, Österreich Austrian Institute of Technology Vollzeit

    Master Thesis “Mixed Reality Algorithms” - Center for Digital Safety & Security - In the master thesis you will develop algorithms to scan and measure the real environment via MR glasses and to generate a virtual representation of this environment. Further, the generated mesh should be structured in elements like walls, doors, and windows; and visual...

  • Master Thesis “site-specific Radio Channel

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis deep Learning for Lidar-based Slam

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis development of a Fast Age-hardening

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis risk, Compliance and Cyber Security”

    vor 2 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    **JOIN OUR LIVE EVENT*** **When**:Wednesday, 20.11.2024 | 12 pm (CET) **Please subscribe **here**! (Participation is free) As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitalisation. To achieve our goals,...

  • Master Thesis „betriebsstrategien Für

    Vor 2 Tagen

    Wien, Österreich Austrian Institute of Technology Vollzeit

    Wir sind Österreichs größte Research and Technology Organisation und führend in der angewandten Forschung für innovative Infrastrukturlösungen. Das macht uns zum leistungsstarken Entwicklungspartner der Industrie und zum Top-Arbeitgeber in der internationalen Wissenschaftsszene. Unser **Center for Energy **sucht Ingenious Partner am Standort Wien: ...

  • Master Thesis tailoring Quasi-solid-state

    vor 3 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As Austria's largest research and technology organization for applied research, the **Austrian Institute of Technology (AIT)** aims to make significant contributions to addressing the major challenges of our time, such as climate change and digitalization. The **Center for Transport Technologies** offers a **master’s thesis** opportunity within the...

  • Master Thesis: Quasi-solid-state Electrolytes

    vor 3 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As Austria's largest research and technology organization for applied research, the **Austrian Institute of Technology (AIT)** aims to make significant contributions to addressing the major challenges of our time, such as climate change and digitalization. The **Center for Transport Technologies** offers a **master’s thesis** opportunity within the...

  • Master Thesis “towards Improved Lithium-ion

    vor 3 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As Austria's largest research and technology organization for applied research, the **Austrian Institute of Technology (AIT)** aims to make significant contributions to addressing the major challenges of our time, such as climate change and digitalization. The **Center for Transport Technologies** offers a **master’s thesis** opportunity within the...

  • Master Thesis autonomous Folding and Unfolding of a

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master Thesis „modeling of Human Perception and

    vor 21 Stunden

    Wien, Österreich Emprego AT Vollzeit

    Master Thesis „Modeling of human perception and motion behavior“ - Center for Energy -**IMAGINE** - The goal of the _perceiving:spaces_ project is to gain new insights into human movement behavior by integrating new sensor technologies in a Living Lab environment in real infrastructures. - Throughout this master thesis project, under the guidance of our...

  • Master Thesis zuverlässigkeitstool Für

    vor 7 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    Als **Österreichs größte Forschungs - und Technologieorganisation für angewandte Forschung** haben wir uns zum Ziel gesetzt, substanzielle Beiträge zur Lösung der großen Herausforderungen unserer Zeit, Klimawandel und Digitalisierung, zu leisten. Dabei setzen wir auf unsere spezifischen Forschungs-, Entwicklungs - und Technologiekompetenzen, die die...

  • Master Thesis high Speed Image Processing

    vor 4 Monaten

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organization** for applied research, we are dedicated to making substantial contributions to solving the major challenges of our time, climate change, and digitization. To achieve our goals, we rely on our specific research, development, and technology competencies, which are the basis of our commitment to...

  • Master Thesis or Professional Internship

    Vor 3 Tagen

    Wien, Österreich Semperit Vollzeit

    You like "out-of-the-box" thinking and enjoy developing innovative solutions? And you want to contribute to finding answers to tomorrow's challenges today? Then you've come to the right place! We offer motivated students (M/F/d) the opportunity to start their **MASTER THESIS or do a PROFESSIONAL INTERNSHIP** **in the fields of** **Production/Process...

  • Master Thesis “llm-supported Uav Swarm Mission Planner”

    vor 4 Wochen

    Wien, Österreich Austrian Institute of Technology Vollzeit

    As **Austria's largest research and technology organisation** for applied research, we are dedicated to make substantial contributions to solving the major challenges of our time, climate change and digitalisation. To achieve our goals, we rely on our specific research, development and technology competencies, which are the basis of our commitment to...

  • Master’s Thesis

    vor 4 Monaten

    Wien, Österreich Max Perutz Labs Vienna Vollzeit

    The goal of the Master project will be to reconstitute the functional signaling unit of TBK1 and mechanistically dissect its activation by upstream signals, such as damaged mitochondria or intracellular pathogens. The project will involve recombinant protein purification, quantitative biochemistry and structural biology.- Open since Sep 16, 2024**About the...