Loss functions and neural networks: Comparing different loss functions for NLP neural networks

Neural network is an active research field which involves many different (unsolved) issues, for example, different types of configuration of the network architectures, training strategies, etc. Amongst these active issues, the choice of loss (or cost) functions plays an important role in how a neural network model is to be optimized (trained)...

Depth Light Field Training (DeLFT): NeRF as a rendering primitive

Neural radiance fields (NeRF) based solutions for novel view synthesis can achieve state of the art results. Recent work proposes models that take less time to render, need less training data or take up less space. However, few papers explore the use of NeRFs in classic rendering scenarios such as rasterization, which could contribute to wider...

Empirical study of GANITE’s robustness to hidden confounders

An empirical study is performed exploring the sensitivity to hidden confounders of GANITE, a method for Individualized Treatment Effect (ITE) estimation. Most real world datasets do not measure all confounders and thus it is important to know how crucial this is in order to obtain comparable predictions. This is explored through the removal of...

Signal-processing of audio for speech-recognition

The transcription of voice using neural networks is a technique that deserves attention, as speech assistants are becoming increasingly popular. Neural networks have often difficulty with determining the differences between a talking person and noise. Humans have a much better understanding of this and could possibly apply their knowledge of the...

Nerve fiber tracing in bright-field images of human skin using deep learning

The goal of this thesis is to find an automated method that can trace all nerve fibers in bright-field images of skin tissue. This is an important step towards the automated quantification of intra-epidermal nerve fiber density, an important biomarker in the diagnosis of small-fiber neuropathy.<br/>Deep learning is a popular new field of...

Statistical analysis to depict the influence of news headlines on stock volatility

A statistical approach to depict stock volatility based on general news headlines by exploiting Machine Learning methods such as sentiment analysis and neural networks. The results from the analysis are then used in the financial market with a particular trading strategy, called a bottom straddle. We conclude that not enough information can be...

Using a Physics-Informed Neural Network to solve the Ideal Magnetohydrodynamic Equations

In this work we investigate neural networks and subsequently physics-informed neural networks. Physicsinformed neural networks are away to solve physical models that are based on differential equations by using a neural network. The wave equation, Burgers’ equation, Euler’s equation, and the ideal magnetohydrodynamic equations are introduced and...

Automated Pricing Suggestions

As Machine Learning is becoming more accessible to small businesses, thanks to the rapid advance in computing power, smaller start-ups such as Sjauf (a ride sharing start-up) are starting to get interested in implementing Machine Learning solutions in their product. Sjauf needed a system that could automatically tell its customers how much a...

Exploring Convolutional Neural Networks

In this thesis we have looked into the complexity of neural networks. Especially convolutional neural networks (CNNs), which are useful for image recognition, are looked into. In order to better understand the process in the neural networks, in the first half of this report a mathematical foundation for neural networks and CNNs is constructed....

GAN Driven Audio Synthesis: On using adversarial training for data driven audio generation

In this study, we investigate the usage of generative adversarial networks for modelling a collection of sounds. The proposed method incites an interpretation of musical sound synthesis based on audio collections rather than synthesizer component controls. This promises the generation of arbitrarily complex sounds without the restrictions of...

AI on Low-Cost Hardware: FPGA subgroup

In the past decades, much progress has been made in the field of AI, and now many different algorithms exist that reach very high accuracies. Unfortunately, many of these algorithms are quite resource intensive, which makes them unavailable on low-cost devices. <br/>The aim of this thesis is to explore algorithms and neural network techniques...

Evaluating the performance of TDNN-BLSTM on Mandarin read and spontaneous speech

A limitation of current ASR systems is the so-called out-of-vocabulary words. The solution to overcome this limitation is to use APR systems. Previous research on Dutch APR systems identified Time Delayed Bidirectional Long-Short Term Memory Neural Network (TDNN-BLSTM) as one of best performing state-of-the-art NN architecture for PR. The goal...

Learning Reduced Order Mappings of Navier-Stokes: An Investigation of Generalization on the Viscosity Parameter

Solving Partial Differential Equations (PDEs) in engineering such as Navier-Stokes is incredibly computationally expensive and complex. Without analytical solutions, numerical solutions can take ages to simulate at great expense. In order to reduce this cost, neural networks may be used to compute approximations of the solution for use during...