Thesis topics
Bachelorarbeitsthema
Significance of the proposed Master thesis
The use of autonomous vehicles is vital in the manufacturing and distribution operations. The automated guided vehicles (AGVs) provide reliable and efficient product handling in several industrial applications.
Goals of the proposed Master thesis
In this Master thesis, we are aiming a deep learning-based solution for pallet detection and position tracking. The process of data augmentation will also be included in this Master thesis. On top of that is the integration of defect detection in the pallets, which is already available.
Planned steps with estimated time
- Literature review and programming language background (30 days)
- understanding of dataset (obtaining 2D images by converting the rangedata polar to cartesian coordinates and resizing the resultant images) (15 days)we will use the publicly available “2D Laser Rangefinder dataset”.
- Suitable possible solutions for the given problem (45 days)One possible solution is Faster R-CNN. In addition to Faster R-CNN, we will try other possible solutions.
- Program a Graphical user interface (GUI) (30days)Lastly, for the successful completion of the Master thesis a professional GUI with embedded solution is required.
- Writing Master thesis (60 days)Writing a master thesis report in English or German.
Prerequisites:
- Interest, motivation and knowledge in supervised machine learning methods.
- Programming skills in Python with following deep-learning libraries.o TensorFlow
- o Keras
- • Proficient use of scientific work
Begin and duration:
Immediately , 6 months
Co-supervisors:
Dr. Ing. Muhammad Saif-ur-Rehman
Tel.: 0208 / 88254806, muhammad.saif-ur-rehman@hs-ruhrwest.de
Supervisors:
Prof. Dr. Ioannis Iossifidis
Tel.: 0208 / 88254806, iossifidis@hs-ruhrwest.de
Masterarbeitsthema
Abstract:
Brain-computer interface (BCI), “the recipe of decoding intended actions from neural signals” is a way forward towards creating an intelligent neuroprosthetics solution. Deep learning (DL) algorithms provide many state-of-the-art results in the rapidly growing BCI applications. Despite this fact, DL algorithms are fragile against synthetic inputs called “adversarial inputs”. These inputs can be crafted by the slight perturbations of the original inputs. The perturbations are applied in the meaningful directions. However, the magnitude of the perturbation is so small that it keeps the original input and adversarial inputs indistinguishable. Nonetheless, these adversarial inputs can easily fool the DL classification algorithm. Hence forth, it is possible to hack the well-trained DL based BCI models.
The scope of this study is to investigate the existence and the importance of adversarial inputs in BCI applications. To the best of our knowledge, there is no other study that investigates the existence of adversarial inputs in the BCI applications.
Planned steps with estimated time
- Literature review and programming language background (30 days)
- Load Dataset (15 days)
we will use the publicly available “BCI competition IV 2b” dataset. - Feature extraction algorithm & DL algorithm (30 days)
We will use the short time Fourier transform (STFT) for feature vector extraction and convolutional neural network (CNNs) for the classification. - Investigations (60 days)
The following questions are required to be answered during the Master thesis.
- Crafting adversarial inputs with two different algorithms.
- Gradient signum algorithm
- Gradient norm method algorithm
- Comparison of adversarial inputs generated with two methods.
- Determine the pockets of adversarial inputs on the given dataset.
- Can we make DL algorithms robust against adversarial perturbations?
- Can we use adversarial inputs for data augmentation?
- Writing Task: Master thesis (45 days)
Prerequisites:
- Interest, motivation and knowledge in supervised machine learning methods.
- Programming skills in Python with following deep-learning libraries.
- TensorFlow
- Keras
- Proficient use of scientific work
Begin and duration:
Immediately , 6 months
Co-supervisors:
Muhammad Saif-ur-Rehman, M. Sc.
Tel.: 0208 / 88254806, muhammad.saif-ur-rehman@hs-ruhrwest.de
Supervisors:
Prof. Dr. Ioannis Iossifidis
Tel.: 0208 / 88254806, iossifidis@hs-ruhrwest.de
Abstract:
A spike sorting algorithm allows the identification of the activity of each neural source. We published two studies SpikeDeeptector and SpikeDeep-Classifier in the journal of the neural engineering. This study is based on our previously published studies. In this study, we aim to identify the neural activity of each source, online. More importantly, we aim to propose the first supervised learning-based solution to the online spike sorting problem.
This study is divided into two parts. Here, in part 1, we aim to label a large dataset. We aim to develop a pipeline that helps labeling the given dataset, conveniently. Our proposed pipeline does not require any experience in neuroscience for labeling.
Planned steps with estimated time
- Literature review and programming language background (15 days)
- Load Dataset (15 days)
we will use the Utah array datasets, microwire datasets, and several publicly available datasets. - Feature extraction algorithm & sorting algorithm (15 days)
We will use the principal components analysis (PCA) for feature vector extraction and gaussian mixture model (GMM) for spike sorting. - Programming user friendly framework (90 days)
A user-friendly (user interface (UI)) framework is required to program. - Writing Task: Master thesis (45 days)
Prerequisites:
- Interest, motivation and knowledge in supervised machine learning methods.
- Programming skills in MATLAB (MathWorks Inc).
- Proficient use of scientific work
Begin and duration:
Immediately , 6 months
Co-supervisors:
Muhammad Saif-ur-Rehman, M. Sc.
Tel.: 0208 / 88254806, muhammad.saif-ur-rehman@hs-ruhrwest.de
Supervisors:
Prof. Dr. Ioannis Iossifidis
Tel.: 0208 / 88254806, iossifidis@hs-ruhrwest.de
| Everyone has experienced muscle fatigue during daily life, but it is more common during physical training or after an illness. In our research project, the muscle fatigue estimator is used as a control unit that regulates the amount of support provided by an exoskeletal system. To be more precise, the estimator is intended to predict muscle fatigue during dynamic arm contraction. For this purpose, muscle activity is measured by Electromyography (EMG), and arm movements are recorded by additional inertial measurement units (IMU). This project is divided into three parts. First, the task environment must be developed, preferable as a small game (like Jump and Run), which is controlled by arm movements. Second, the data acquisition, here up to 8 participants have to perform the previously developed task. Third, the data analysis, includes state-of-the-art feature extraction and estimator development. Depending on individual interest and time-scale the different parts can be adapted. Due to the current corona situation, most of the work should be done from home, with communication via WebEx, etc. However, the final data acquisition must be done in the laboratory under controlled hygienic conditions. |
| Planned steps with estimated time: Literature review (30 days)Task development (30 days)Data acquisition (15 days)Data analysis(60 days)Writing Task: Master Thesis (45 days) |
| Prerequisites: Interest, motivation and knowledge in mathematics and informatics. Programming skills in Matlab® or Python, Unity, Java Proficient use of scientific work |
| Begin and duration: Immediately, 3-6 months |
| Co-supervisors: Marie Schmidt, M. Sc. marie.schmidt@hs-ruhrwest.de Supervisors: Prof. Dr. Ioannis Iossifidis, iossifidis@hs-ruhrwest.de |
| Brain-computer interface (BCI) systems are a rapidly growing technology that controlsexternal devices, e.g. a neuroprosthetic limb, by directly decoding intended movements from the recorded neural activities and bypassing the spinal cord. Decoding neural activity is a two-step process, feature vector extraction and classification/regression. In online BCI applications, non-stationary behavior of neural signals makes the process of meaningful feature extraction non-trivial. Conventionally, band-pass filtering along with simple threshold crossings are often used to extract feature vectors. These extracted feature vectors include unnecessary noise/artifacts. Thus, makes the task of a classifier/regressor hard. To overcome this problem, we want to evaluate a novel feature extraction method based on our two previously introduced separate methods (Spike-deeptector and artifact rejector). The scope of this study is to investigate the role of novel feature extraction method in online invasive BCI applications. We will use previously recorded data from a tetraplegic patient of multiple recording sessions. We will evaluate the trained model of classifier/regressor along with trained model of novel feature extractor in online mode. |
| Planned steps with estimated time· Literature review (30 days).Understand the behavioral task and its connection to neural data (30 days)Programing Task (45 days).Online evaluation of designed algorithm (15 days)Writing Task: Master Thesis (45 days) |
| Prerequisites:Interest, motivation and knowledge in supervised machine learning methods.Programming skills in Matlab® (or Python) Proficient use of scientific work |
| Begin and duration:Immediately , 6 months |
| Co-supervisors: Muhammad Saif-ur-Rehman, M. Sc.Tel.: +49 208 88254806, Muhammad.Saif-ur-Rehmann@hs-ruhrwest.de Supervisors: Prof. Dr. Ioannis Iossifidis, Tel.: +49 208 88254806, iossifidis@hs-ruhrwest.de |
contact: Muhammad Ayaz Hussain, M. Sc.,
Tel: +49 208 88254806
muhammad.hussain@hs-ruhrwest.de
| A brushless dc (BLDC) motor drive is characterized by higher efficiency, lower maintenance, and higher cost. In a market driven by profit margins, the appliance industry is reluctant to replace the conventional motor drives with the advanced motor drives (BLDC) due to their higher cost. Therefore, it is necessary to have a low-cost but effective BLDC motor controller. In this Thesis, Student has to perform position, velocity and torque control using basic microcontroller, Simulink and BLDC driver circuitry. |
| Planned steps with estimated time· Motor and Control Basics(depending on level of knowledge)Familiarization with basic Hardware setup (5 days)Familiarization with Simulink and Arduino (10 days) (prior Arduino knowledge not required)Implementation of control algorithm (40 days).Evaluation and comparison of implemented algorithms (15 days) (if we have anything to compare with) Writing of the thesis report (45 days) |
| Prerequisites:Interest, motivation to learn and do project in Motor Control. Basic Knowledge of Motors and Control Theory.Programming skills (preferably in Matlab) Proficient use of scientific workAvailability to work at Robotics Lab in Campus Bottrop 3 days a week |
| Begin and duration:Anytime, 6 months |
| Co-supervisors: Muhammad Ayaz Hussain, M. Sc., Tel: +49 208 88254806 muhammad.hussain@hs-ruhrwest.de Supervisors: Prof. Dr. Ioannis Iossifidis, Tel: +49 208 88254806 ioannis.iossifidis@hs-ruhrwest.de |
| Reinforcement Learning (RL) is a subdomain of machine learning that has developed rapidly in recent years and has become increasingly popular. In reinforcement learning an agent learns from experience using a scalar reward signal, in contrast to learning from examples of labelled data as it is done in supervised machine learning. The agent interacts with an environment by taking actions that influence the environment and thus the observations that are the agent’s input.Experience Replay is a technique that helps agents learn more from a given amount of transitions (i.e. experience). While classical incremental online algorithms use only the current transition in a learning step, experience replay stores transitions in a replay memory from where samples can be taken randomly. The idea of prioritized experience replay is that some transitions are more valuable to learn from than others, so they should be sampled with a higher probability. There are several ways to prioritize and multiple implementation choices.For this thesis you would search for such methods in existing literature, choose some of them (the number depends on the amount of work each one entails) and implement and compare them amongst each other and, if time allows it, with standard non-prioritized variants of the same underlying algorithms. Criteria will include learning efficiency and computational efficiency but you should add your own as well. |
| Planned steps with estimated time· Reinforcement Learning foundation (depending on level of knowledge)Literature review and understanding (45 days)Implementation of chosen algorithms (20 days).Evaluation and comparison of implemented algorithms (15 days)Writing of the thesis (45 days) |
| Prerequisites:Interest, motivation in machine learningKnowledge and experience with RL are desirable, but not strictly necessary. This will be reflected in the length of the initial learning phase.Programming skills (preferably in Python) Proficient use of scientific work |
| Begin and duration:Anytime, 6 months |
| Co-supervisors: Felix Grün, M. Sc., Tel.: +49 208 88254 875, felix.gruen@hs-ruhrwest.de Supervisors: Prof. Dr. Ioannis Iossifidis, Tel: +49 208 88254806 iossifidis@hs-ruhrwest.de |
| Reinforcement Learning (RL) is a subdomain of machine learning that has developed rapidly in recent years and has become increasingly popular. In reinforcement learning an agent learns from experience using a scalar reward signal, in contrast to learning from examples of labelled data as it is done in supervised machine learning. The agent interacts with an environment by taking actions that influence the environment and thus the observations that are the agent’s input.What is often learned by a RL agent is the expected return of taking a certain action in a certain state. There are, however, also algorithms that try to capture more information than just the expected value, i.e. higher order moments or even full distributions. Learning the variance for example has been used in designing risk-sensitive agents.For this thesis you would search for such methods in existing literature, choose some of them (the number depends on the amount of work each one entails) and implement and compare them amongst each other and, if time allows it, with non-distributional variants of the same underlying algorithms. Criteria will include learning efficiency and computational efficiency but you should add your own as well. |
| Planned steps with estimated time· Reinforcement Learning foundation (depending on level of knowledge)Literature review and understanding (45 days)Implementation of chosen algorithms (20 days).Evaluation and comparison of implemented algorithms (15 days)Writing of the thesis (45 days) |
| Prerequisites:Interest, motivation in machine learningKnowledge and experience with RL are desirable, but not strictly necessary. This will be reflected in the length of the initial learning phase.Programming skills (preferably in Python) Proficient use of scientific work |
| Begin and duration:Anytime, 6 months |
| Co-supervisors: Felix Grün, M. Sc.Tel.: +49 208 88254 875, felix.gruen@hs-ruhrwest.de Supervisors: Prof. Dr. Ioannis Iossifidis, +49 208 88254806 iossifidis@hs-ruhrwest.de |