Active Learning by Discrepancy Minimization: A Comparison of Active Learning Methods Motivated by Generalization Bounds

In many settings in practice it is expensive to obtain labeled data while unlabeled data is abundant. This is problematic if one wants to train accurate (supervised) predictive models. The main idea behind active learning is that models can perform better with less labeled data, if the model may choose the data from which it learns. Active...

Tabular Reinforcement Learning Aided by Generalisation Methods

Reinforcement learning is a machine learning paradigm that deals with optimisation and learns by interacting with its environment. Tabular reinforcement learning methods are popular because of their relative simplicity combined with good guarantees of finding an optimal solution. The downside is that they suffer from an exponentially growing...

Micro-Activity Recognition using Wearables for Human Augmentation

Wearable devices have paved way for several context-aware applications in the field of health-care and sports to improve the well-being of users and their performance for human augmentation. During rehabilitation patients need accurate feedback that can empower and improve the speed of recovery. On the other hand competitive athletes need a...

Predicting customer loyalty

The Investment Department of ING asked us to build a model that predicts customer loyalty using the Net Promoter Score (NPS) system and a dashboard that would show the results of the model. With this model and dashboard ING may be able to get an insight into which groups of customers are at risk to complain.

Where will you comment next? Exploiting comments for personalized recommendations

Since the advent of Web 2.0, users have not only increasingly created content, but also contributed reactions to content in the form of comments. Comments are challenging to analyze due to their short lengths and informal style, meaning that any individual comment provides very little data to work with and is highly variable. However, comments...

Enhancing Real-Time Twitter Filtering and Classification using a Semi-Automatic Dynamic Machine Learning setup approach

Twitter contains massive amounts of user generated content that also contains a lot of valuable information for various interested parties. Twitcident has been developed to process and filter this information in real-time for interested parties by monitoring a set of predefined topics, exploiting humans as sensors. An analysis of the relevant...

Supervised Learning for Measuring Hip Joint Distance in Digital X-ray Images

Osteoarthritis is a degenerative joint disease which is hard to diagnose objectively and may vary based on the surgeon. This disease is usually diagnosed by measuring several characteristic features of Hip X-rays mainly the joint distance between the femoral head and acetabular cup. Hip joint distance reduction is a clear symptom of...

High-Speed Object Detection: Design, Study and Implementation of a Detection Framework using Channel Features and Boosting

In this thesis we design, implement and study a high-speed object detection framework. Our baseline detector uses integral channel features as object representation and AdaBoost as supervised learning algorithm. We suggest the implementation of two approximation techniques for speeding up the baseline detector and show their effectiveness by...

Anomaly detection for internet banking using supervised learning on high dimensional data

Nowadays, a high number of transactions are performed via internet banking. Rabobank processes more than 10 million transactions per day. Most of these transactions are (part of) normal behaviour. On the other hand, some transactions are considered to be out of the ordinary. These anomalous events occur relatively infrequently (less than 10 per...

Classification of valence using facial expressions of TV-viewers

Emotion has been shown to have a large impact on our interactions with people and devices. In our daily lives, however, these emotions are not taken into account when working with our computers and other machines. If our devices could pick up on social cues, for instance in relation to disinterest, the usability of various systems could be...

Using the Multiple Instance Learning framework to address differential regulation

Cell differentiation is a natural process occurring in all higher organisms, since the early fetal stage of life. It is, also, a part of disease – such as cancer – as the cell cycle becomes deregulated and cells behave differently compared to healthy ones. Differentiation occurs although the genome of all cells is identical across all cell types...

Interpretable classification of tumours through multiple instance learning and somatic mutations

Next generation sequencing is brought into the clinic. Screening of disease associated genes will aid the diagnosis of disorders with a genetic component. The diagnosis of cancer is of particular interest due to its variety and prevalence. The obtained mutations provide clues about underlying biological properties that could be used for...

Sample reusability in importance-weighted active learning

Recent advances in importance-weighted active learning solve many of the problems of traditional active learning strategies. But does importance-weighted active learning also produce a reusable sample selection? This thesis explains why reusability can be a problem, how importance-weighted active learning removes some of the barriers to...

Evaluating the Quality of Opponent Models in Automated Bilateral Negotiations

Automated negotiation agents are agents that interact in an environment for the settlement of a mutual concern. An important factor influencing the performance of a negotiation agent is how it takes the opponent into account. The main challenge in this aspect, is that opponents typically hide private information to avoid exploitation. In such a...