Index

CLICK-SPOT: Detection and Classification of Cetacean Echolocation Clicks using Image-based Object Detection Methods applied to Advanced Wavelet-based Transformations

Detection of Alzheimer’s disease and depression in speech by Graph Neural Networks

Ensuring Quality of Bots Powered by Generative Artificial Intelligence with Automated AI-Persona-Based Testing

xLSTM: Extended Long Short-Term Memory for Enhanced Performance and Scalability

Automatic Data Augmentation for Multi organ Segmentation

Various medical image segmentation models have been created in the past few years for both supervised and semi-supervised tasks. Since variability and diversity of data play a huge role in the accuracy of such models, choosing the most appropriate data augmentation strategy for each organ or tumor is very important in improving the model outcomes. However, most of these data augmentation strategies used in UNet and its variants are simple, custom-made, and not fully optimal. This manual selection of data augmentation strategies limits the possibility for improvement of the accuracy of the medical image segmentation.
As a further improvement on this, Yang et al [1] proposed a technique for the automation data augmentation strategy selection using reinforcement learning. However, this technique requires a high computational time and only tells about the probability of each augmentation strategy. To overcome this issue, Xu et al. [2] proposed an automatic data augmentation strategy framework called ASNG, which searches for the most optimal augmentation strategy by formulating a bi-level optimization algorithm. This framework also designed a search space that includes the fixed magnitude of the operation and the interval of magnitude. Moreover, this work showcases the dynamic change of strategies during training as per requirement, achieving a state-of-the-art performance.
The thesis aims to research the following:
1. How different data augmentation strategies influence the segmentation performance in different data conditions. The data conditions are based on object shape/size, number of objects, mean and standard deviation of the intensity, and so on.
2. Perform analysis of the correlation of the data conditions and the augmentation methods. For this, first, a search space is to be defined which includes typical augmentation strategies with varying hyperparameters, similar to the work done by He et al [3] and Cubuk et al [4]. The performance of different augmentation methods is tested on a segmentation model built using the MONAI framework.
3. Implement an algorithm for searching strategy using machine learning regressor models like support vector regressor that will lead to automatic planning of the augmentation pipeline based on the target dataset. The dataset used for the algorithm development is AMOS [5]. AMOS is a large-scale, diverse, clinical dataset comprising CT and MRI scans for 15 abdominal organ segmentation.

Image2Tikz: Neural Network-Based TikZ Code Generation

Work description
This thesis aims to develop Image2Tikz, a tool that uses neural networks to transform images into TikZ code by segmenting objects, estimating distances between them, and then translating this information into TikZ. Starting with simple TikZ images and progressing to more complex, hand-drawn versions, the tool will be refined to handle increasing noise levels and complexities. The performance will be evaluated by comparing the generated TikZ code against the original images.

The following questions should be considered:

How can neural networks be effectively trained for tikz object segmentation and distance estimation in images?
What techniques can be used to translate segmented objects and their relative distances into TikZ code?
How does the introduction of noise and complexity in images affect the tool’s accuracy and reliability?
What strategies can improve the tool’s performance on more complex, hand-drawn images?

Prerequisites
Candidates should have a strong foundation in machine learning, particularly in neural networks, with practical experience in Python and familiarity with PyTorch. Skills in image processing and an understanding of LaTeX, especially TikZ, are desirable. The ability to work independently and creatively solve problems is essential.

Please include your transcript of record with your application.

Predictive Financial Modelling for Siemens Air Insulated Switchgears

In today’s financial forecasting, traditional methods like manual calculations and relying on expert
intuition often don’t meet the needs of complex industrial settings, such as Siemens’s Air-Insulated
Switchgear Business.
This thesis explores how machine learning algorithms can improve predictive financial modeling,
especially for budgeting and forecasting key financial metrics. Although only 3-4 years of historical
data are available, this research looks into machine learning techniques that can still make accurate
predictions by uncovering patterns in the data. The study focuses on time series models, regression
techniques, and ensemble methods that are effective for small datasets, and assesses their ability to
forecast financial KPIs. Additionally, the research examines which financial metrics most influence
forecasting accuracy, aiming to develop a more reliable, data-driven approach to financial planning
that can evolve and enhance organizational decision-making.
Siemens AG provides the dataset and contains tabular time series data.

Research Objectives
1. Assess the effectiveness of different algorithms in forecasting financial KPIs. Focus on determining
which algorithm provides the most accurate predictions and best captures the
non-linearity’s in the data.
2. Exploring different techniques to generate synthetic data points and various data augmentation
techniques to enhance the robustness of the predictive models.
3. Experiment with both unified and factory-specific models to identify whether a single model
can effectively forecast across all factories or if individual models for each factory yield better
results.
4. Implement and assess the impact of regularization, ensemble methods, and other advanced
techniques on the performance of the predictive models.

The thesis involves the following key steps:
• Step 1: Literature review and theoretical framework development.
• Step 2: Data pre-processing, and analysis.
• Step 3: Design and develop machine learning model architectures tailored to forecast financial
KPIs.
• Step 4: Evaluate and compare the results of the models.
• Step 5: Select the best performing model and refining it further.
• Step 6: Thesis writing and final presentation preparation.
Through an in-depth exploration of data analytics and machine learning, this thesis seeks to elevate
predictive financial modeling by investigating effective strategies and model architectures. The
theoretical framework, grounded in a comprehensive literature review, will guide the study’s key
steps, leading to actionable insights for improving budget planning and financial forecasting in
Siemens’ AIS-producing factories.

References
[1] Samuel A Assefa, Danial Dervovic, Mahmoud Mahfouz, Robert E Tillman, Prashant Reddy, and
Manuela Veloso. Generating synthetic data in finance: opportunities, challenges and pitfalls.
In Proceedings of the First ACM International Conference on AI in Finance, pages 1–8, 2020.
[2] Daniel Broby. The use of predictive analytics in finance. The Journal of Finance and Data
Science, 8:145–161, 2022.
[3] Odeyemi Olubusola, Noluthando Zamanjomane Mhlongo, Donald Obinna Daraojimba, Adeola
Olusola Ajayi-Nifise, and Titilola Falaiye. Machine learning in financial forecasting: A us
review: Exploring the advancements, challenges, and implications of ai-driven predictions in
financial markets. World Journal of Advanced Research and Reviews, 21(2):1969–1984, 2024.
[4] Meryem Ouahilal, Mohammed El Mohajir, Mohamed Chahhou, and Badr Eddine El Mohajir. A
comparative study of predictive algorithms for business analytics and decision support systems:
Finance as a case study. In 2016 International Conference on Information Technology for
Organizations Development (IT4OD), pages 1–6. IEEE, 2016.