Open topics for theses and practical courses
Unless otherwise specified, all topics are available as practical course (P1/P2), bachelor or master thesis.
|
Work group:
|
Type:
|
Building self-explanatory transparent models
Explainable Artificial Intelligence (XAI) focuses on providing explanations for domain experts and data scientist. However, 'lay people', i.e., people without or only with limited technical knowledge, are often subject to the consequences of deploying algorithmic decision-making systems in public institutions, such as the COMPAS recidivism model. Finding new ways to render machine learning models comprehensible for lay people is therefore a current and highly relevant challenge. This project will focus on approaches for building machine learning models that are in themselves interpretable and explainable for lay people, soliciting slight or no external explanations. It will involve both programming and the conducting of user studies with lay people.
Contact: Timothée Schmude, Sebastian Tschiatschek
Causal Structure Learning for Questionnaires
Many research question are not only statistical but rather causal in nature. Two variables may be statistical dependent on each other, but do not not cause each other. For example, the relative number of Nobel laureates for a country strongly correlates with its annual chocolate consumption. However, increasing the national chocolate consumption, does obviously not produce more outstanding scientists. It is a key challenge for machine learning to move from learning purely statistical relations to causal ones. In the absence of randomized trials, causal structure learning focuses on identifying which variables in a given dataset are directly causally related and how these direct causal relations form a structure over which also indirect causal relations exist. These dependencies can be encoded by a Directed Acyclic Graph (DAG) and the first step in performing causal inference is to determine such underlying causal graph.
In this project you will make yourself familiar with causal structure learning in machine learning. You will benchmark recent algorithms to find underlying causal relations in datasets and apply them to questionnaire data from the social and political sciences.
Students working on this project need basic background knowledge in machine learning or probability, good programming skills in Python, and the desire to implement and test causal discovery algorithms.
Contact: Simon Rittel, Sebastian Tschiatschek
Attentive pixel prediction (Reinforcement Learning)
Auxiliary prediction tasks have been the source of many recent performance improvements for reinforcement learning agents, particularly when working with image inputs. These tasks are usually provided as expert knowledge by the algorithm designer and force the agent to pay attention to various aspects of the environment which might be helpful to solving the underlying task. In this project you will investigate whether such prediction tasks can also be learned by the agent itself. For this, you will first visualize what an RL agent actually "attends to" through investigating heatmaps of their first network layer. Then you will devise a method allowing the agent itself to decide which pixels of the state it wants to learn to reconstruct.
Required skills: Good programming skills, basic background in deep learning
Contact: Timo Klein, Sebastian Tschiatschek
Reinforcement Learning meets transfer learning (Reinforcement Learning)
Reinforcement learning as a field has seen tremendous recent success on benchmark environments such as Atari or Go. Current state of the art methods often perform well on a subset of these benchmarks but it is unclear how well these results generalize to other domains. In this project, you will investigate how well results obtained in one type of environment transfer to other types of environments. The first step consists of categorizing current RL benchmarks according to a number of abstract properties, e.g., their reward structure or the agent's point of view. In a second step, you will implement a number of algorithms on all of these benchmarks and evaluate how well results transfer among different categories of environments.
Required skills: Good Python coding skills, Interest in RL
Contact: Timo Klein, Sebastian Tschiatschek
Machine learning based climate model weighting
Climate models, global and regional, are based on numerical and physical equations and parametrizations and, thus, are computationally expensive especially when trying to simulate more than 10 years. Based on the interpretation of equations (e.g. topography following atmosphere) and the parametrizations for e.g. clouds results of climate models vary in their skills. This is especially the case in regions with complex topography.
In the past years the idea of mixing different climate model simulations based on their respective skills in e.g. reproducing history precipitation has emerged. Thus, generating a sort of mean climate simulation consisting of information of different climate models. This is called weighted model mixing (e.g., Brunner et al., 2020; Rana et al. 2013). The best performing model with respect to a specific observation type thus has the most influence on the resulting simulation.
With the emerging skill of different machine learning models the statistical weighting (Rana et al., 2013, Brunner et al., 2020, Merrifield et al., 2020, Hamill and Scheuerer (2018)) often has lower skills compared to e.g. a weighted Convolutional Neural Network (Bertrand et al., 2021). Other approaches can have shown good skills, too (Wong et al. 2021, Sun and Archibald (2021), Segupta et al. (2021), Amos et al. (2021).
The idea of the proposed thesis is to implement two methods, a classical statistical as baseline and a machine learning model, to combine selected GCM and RCMs with the weighted algorithm based on their respective skills.
Contact: Irene Schicker (ZAMG), Katerina Schindlerova
The Complexity of Computing the Graph Edit Distance
The graph edit distance (GED) quantifies the dissimilarity of two graphs as the minimum cost of a sequence of edit operations turning one graph into the other. The problem complexity depends on the choice of the edit operation costs and the properties of the considered graphs. It has been shown that the classical NP-hard subgraph isomorphism problem (SI) and the maximum common subgraph problem (MCS) can be reduced to computing the GED. As a consequence, computing the GED also is NP-hard. However, for SI and MCS a fine distinction of the complexity depending on parameters such as treewidth, degree, and their combination is known. The goal of this project is to investigate the complexity of computing the graph edit distance in restricted graph classes identifying polynomial-time solvable and NP-hard cases.
Requirements: Strong interest in graph algorithms and theoretical computer science
Contact: Nils Kriege
Graph Rewiring Heuristics for Graph Neural Networks
Several methods of modifying input graphs before applying a learning algorithm have been proposed, e.g., adding additional virtual edges marked with a specific label. This can mitigate known weak points of learning methods regarding expressivity, over-smoothing, or the bottleneck problem. This project aims to formalize such weak points and to develop and experimentally evaluate heuristics for graph rewiring to overcome them.
Requirements: Strong interest in graph learning and neural networks
Contact: Nils Kriege
Reinforcement Learning for improving mental health treatments
In this project/thesis you will study the potential of using reinforcement learning (RL) for improving mental health treatments. In particular, you will investigate the potential of RL algorithms for optimizing exposure therapy protocols, which are usually used for the treatment of anxiety disorders, such as phobias.
Specifically, you will conduct simulation studies in which a computational associative learning model (such as the latent cause model) will be a stand-in for the patients, and where an RL algorithms will make modifications to the exposure protocol with the goal of robustly extinguishing negative associations, which play a crucial role in anxiety disorders. The investigation can start with standard model-free RL approaches, but is then planned to proceed with model-based augmentations, such as equipping the RL algorithm with "Theory-of-Mind"-type capabilities.
Students working on this project need basic background knowledge in machine learning, solid programming skills, a desire to work on reinforcement learning, and willingness to learn about models of human cognition.
Contact: Sebastian Tschiatschek
Graph Learning for molecules with bioisosteric replacements
The drug discovery process can be accelerated by machine learning methods. Of particular interest are graph learning methods operating on the molecular graphs, such as graph kernels and graph neural networks. The goal of this project is to incorporate knowledge on structural replacements that preserve the biological activity of molecules, so-called bioisosteres.
Requirements: Interest in machine learning, basic knowledge about the fundamentals of organic chemistry advantageous
Contact: Nils Kriege
Information-theoretic measures and the EEG time series of depression patients [Master Thesis]
Based on the EEG recordings database of depression patients, the goal is to utilize information-theoretic measures to assess the distances among the EEG signal of the patients under different treatment conditions with the intention to apply the results to a related classification problem. The interested student will be provided more details.
The work will be done within the international research project "Learning Synchronization Patterns in Multivariate Neural Signals for Prediction of Response to Antidepressants", which is a joint research project of University of Vienna and of the Czech Academy of Sciences
Contact: Katerina Schindlerova, Claudia Plant
Efficient algorithms for uncertain graphs [Master Thesis]
Uncertain graphs represent an important data model for real-world networks, where one can assign an independent probability of existence on every edge. Uncertainty, represented by the edge probability, may arise due to several reasons, e.g., errors in measurements, data integration from inconsistent and ambiguous sources, lack of precise information needs, inference and prediction models, and explicit manipulation for privacy purposes. Due to their rich expressiveness and utility in a wide range of applications, uncertain graphs have prompted a great deal of research by the data mining research community.
This work will mainly focus on designing efficient algorithms for selecting and modifying the probability of a given number of edges in an uncertain graphs. The developed algorithm will be evaluated against other existing methods. An experimental study will be conducted on real-world datasets. Students working on this project need basic knowledge in graph theory and algorithms, and solid programming skills in Python, C/C++ or Java.
Contact: Yllka Velaj, Claudia Plant
Common Subgraph Problems in Tree-Like Graphs
For two given graphs, G and H, the maximum common subgraph problem (MCS) asks for the largest graph contained in both G and H. An important application occurs in cheminformatics, where the similarity of molecular graphs needs to be quantified. Unfortunately, MCS is NP-hard unless additional constraints regarding the properties of G, H, and the subgraph are enforced. A polynomial-time solvable case requires that all these graphs are trees. Known negative complexity results for classes of tree-like graphs set narrow boundaries for generalization.
The project aims to investigate particular tree-like graph classes to design, implement, and experimentally evaluate new polynomial-time algorithms. New NP-hardness proofs for specific cases should refine the complexity status further.
Requirements: Strong interest in algorithms and graph theory
Contact: Nils Kriege
Dynamic Orbit Partition [Practical Course or Master Thesis]
A graph isomorphism is a mapping between the nodes of two graphs that preserves the edges. An automorphism is an isomorphism of a graph to itself. The (non-trivial) automorphisms intuitively represent the symmetries of a graph. Two vertices are considered equivalent if there is an automorphism that maps one vertex to the other. The corresponding equivalence classes are referred to as orbit partition. The orbit partition is of interest for symmetry breaking techniques for various combinatorial problems. Identifying and pruning symmetric branches in the search tree based on the symmetries of the input instance can lead to drastic speed-ups. This project aims to develop dynamic algorithms for maintaining the orbit partition when the input graph changes. The focus of the project can be on the development and theoretical analysis of algorithms or their implementation and experimental evaluation.
Requirements: Strong interest in algorithms
Contact: Nils Kriege, Christian Schulz, Kathrin Hanauer
Interpretable and Explainable Deep Learning
In this project you will study novel approaches for learning interpretable and explainable deep learning models with the goal of supporting users with different skill levels and knowledge. The models will be evaluated on applications with large societal impact, e.g., labour market data.
Contact: Sebastian Tschiatschek
Dynamic Information Acquisition in Questionnaires
In many areas of our life questionnaires/a collection of measurements are used to gather information for decision making, e.g., questionnaires are used to measure the success of marketing campaigns or customer satisfaction, or medical tests are conducted to decide on the best treatment for a patient. Common to those applications is that many questions need to be answered or many tests must be conducted. Often this is not because all the information is needed for deriving a decision but rather because the course of action is standardized (e.g., a survey always consists of the same questions).
In this project you will evaluate and develop methods for adaptive information acquisition that acquire the information necessary for decision making sequentially and adaptively, e.g., not all questions in a questionnaire are asked in all cases and the order of the questions can be changed. You will consider and evaluate recent machine learning models for this problem but also investigate how these methods can be extended to account for dependencies in the information that can be required, e.g., dependencies that only allow for a specific order of some questions of a questionnaire, the influence of how questions are asked on the answers, etc.
Contact: Sebastian Tschiatschek
Deep Probabilistic Clustering for Heterogeneous and Incomplete Data
Clustering is the task of finding groups in a data set. As an unsupervised learning approach it is important for exploratory data analysis, e.g., for finding commonalities between patients suffering from some disease. Often such data is high-dimensional and contains complicated non-linear relationships, which makes it difficult to apply standard clustering approaches. Deep clustering algorithms have been proposed recently to approach this kind of setting by combining clustering and deep learning in a common framework. An open problem that has not been sufficiently addressed in recent deep clustering research is incomplete and heterogeneous (e.g. survey data, blood tests, MRI images, etc.) data. Incomplete data is often an issue in practice, especially for health data where not everything about a patient’s history is known.
In this project you will study the clustering of incomplete data with heterogeneous data types with a particular focus on missingness and on determining which information is relevant for cluster assignments. You will develop an algorithm based on previous work and evaluate it on synthetic and real-world data. Some research questions of interest would be for instance: How do cluster assignments change when given new information? Or, given an uncertain cluster assignment (e.g. sick or healthy patient) which information would we need to query to make a decision?
Contact: Sebastian Tschiatschek
Multi-agent Reinforcement Learning under Mismatch
Enabling intelligent agents to collaborate effectively despite mismatch in perception or capabilities.
Contact: Sebastian Tschiatschek
Multi-agent Teaching Primitives
Enabling multiple interacting agents to convey their knowledge and skills through teaching.
Contact: Sebastian Tschiatschek
Abstraction in Reinforcement Learning
Learning abstractions of large state and action spaces in order to facilitate more efficient planning and exploration.
Contact: Sebastian Tschiatschek
Bivariate causal measures: Experiments with existing causal methods in python on synthetic and real data [Practical Course]
Bivariate causal measures: Experiments with existing causal methods in python on synthetic and real data.
Contact: Katerina Schindlerova
GPU based data mining on Android devices
Energy efficiency for high performance computing is a current hot topic in scientific research. Many Android(tm) devices have an integrated GPU and a vendor supplied OpenCL shared C library on the device. Although not directly supported by Android, these OpenCL libraries allow to address onboard GPUs by application developers.
In a former project a comfortable development environment for Android studio has been created. A C wrapper library serves as link between the Java/JNI/C part of an application and the OpenCL library on the device. Locking mechanisms allow to use the GPU in a safe manner and free resources timely after the application has been stopped externally (by the user or the operating system).
For this project the wrapper library already developed should be used to implement data mining algorithms with OpenCL and compare runtimes on the GPU with those achieved on the device (Java and C with SIMD instructions). Candidate algorithms are DBSCAN, KNN , Subclu and Optics. Students may implement further algorithms at their pleasure. An application stub with the OpenCL wrapper library will be provided.
Prerequisites:
- Good knowledge of Java, C and OpenCL (Kotlin only upon agreement)
- Knowledge of GPU design
- The student should be familiar with the IDE (Android Studio) for the application development.
- The student must have an Android device (tablet or mobile phone) with an integrated GPU and the OpenCL library must be available on the device (please check directories /system/lib, /system/lib64, /vendor/lib, /vendor/lib64 and technical documentation of the device). The application can NOT be developed with the AVD manager (GPU not supported)! If the Android OS version on the tablet is 7.0 or later (API>=24), please check if the name of the OpenCL library is listed in the files/vendor/etc/public.libraries.txt or /system/etc/public.libraries-COMPANYNAME.txt on the device, because if not, Android does not allow to dynamically load the library at runtime. This is a mandatory prerequisite for the use of the wrapper library.
- For the use of SIMD instructions, either knowledge of (inline) assembler (depending on the students tablet CPU architecture arm-v7, aarch64, x86 or x86_64) or knowledge of the use of SIMD intrinsics (NEON or SSE) in C.
Contact: Robert Fritze, Claudia Plant
Learning with Reduced Molecular Graphs
Drug discovery at its early stage can greatly benefit from machine learning methods. As molecules are structured objects consisting of atoms and chemical bonds, they cannot be represented by vectors in a straightforward way, but are adequately modeled as graphs. Recent advances in machine learning with graphs led to well engineered methods applicable to graphs annotated with node and edge attributes, e.g., graph neural networks. For molecules, different graph representations exists. The most natural approach is to represent atoms as nodes and bonds as edge. However, different so-called reduced graph models exist, where groups of atoms are represented by a single node and their properties by node attributes. The goal of this project is to compare the performance of the various graph learning techniques with different graph models. Based on the result tailored combinations of methods and representations should be developed.
Students wanting to work on this topic are expected to have experience in machine learning and basic knowledge on graph theory and algorithms. Solid programming skills, preferably in Python, are required.
Contact: Nils Kriege, Franka Bause
Oracle analysis of distant supervision errors
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Weakly supervised discourse relation prediction
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Incomplete Schema Relation Clustering
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Weakly supervised learning with latent class predictions
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Gradient matching for semi-supervised learning
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Threshold-finding for knowledge-base completion using Gaussian processes
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Path-based knowledge-base completion
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Better sentence representations based on BERT
Please see [Moodle] for a detailed description of this and other Natural Language Processing topics.
Contact: Benjamin Roth
Explainable Policies for Game Play [Master Thesis]
Reinforcement learning has shown remarkable successes in playing board games like Go but also in playing computer games like DOTA. While these results are impressive, the learned strategies (policies) are typically not directly interpretable by humans. However, interpretability is important for instance to enable better collaboration of humans with AI agents or to assess safety of the learned behavior.
In this project you will make such learned policies easier to understand by humans by incorporating constraints on their structure. You will develop and implement a novel policy training algorithm and evaluate it on simple computer games.
Students wanting to work on this topic are expected to have solid coding skills in Python, basic knowledge in PyTorch or TensorFlow, and be curious to learn about reinforcement learning and its applications.
Contact: Sebastian Tschiatschek
Learning Composable Policies [Master Thesis]
Reinforcement learning has shown remarkable successes in application domains like game play and robotics but suffers from large sample complexity, limiting its broader usage. A promising approach for reducing the sample complexity is to learn skills that can be reused when solving a series of related tasks. In the context of robotics, such skills could for instance correspond to lifting a leg or grepping an object.
In this project you will develop and implement a learning algorithm algorithm for such composable policies based on recently proposed Successor Options. You will evaluate this algorithm on several benchmark tasks and compare it to other state of the art algorithms.
Students wanting to work on this topic are expected to have solid coding skills in Python, basic knowledge in PyTorch or TensorFlow, and be curious to learn about reinforcement learning and its applications.
Contact: Sebastian Tschiatschek
The Cost of Feedback
Reinforcement learning has shown remarkable successes in application domains like game play and robotics. These successes where achieved by performing large numbers of interactions of a learning agent with the environment. In these interactions, the agent constantly receives information about rewarding behavior. However, in many applications involving human users (e.g., personal assistants), the reward information is explicitly or implicitly provided by these human users. Therefore, this information must be treated as a scarce and valuable resource, and the (cognitive) cost of providing this information should be accounted for.
In this project you will study the cost of providing different types of feedback used for reinforcement learning in user studies. To this end, you will implement a simple reinforcement learning environment for which a human user can provide reward information in various forms (e.g., comparisons, sorted lists, etc.) and compare the cost of providing this feedback as well as the usefulness of the feedback for training the reinforcement learning agent. Depending on whether you are doing a P1/P2/Bachelor Thesis/Master Thesis, the scope of the project will be adjusted.
Students wanting to work on this topic are expected to have solid coding skills in Python, basic knowledge about the development of interactive web pages, and be curious to learn about reinforcement learning and its applications.
Contact: Sebastian Tschiatschek
Reinforcement Learning for Optimizing Electronic Circuit Design [Practical Course]
Electronic circuits are parts of all electric devices we use on a daily basis. When developing a new electric device, tailored electronic circuits have to be constructed for the specific use case, an often time-consuming process. Therefore it is appealing to aim to automate this process. Reinforcement learning is a promising approach therefore and has shown remarkable successes in application domains like game play and robotics but is so far rarely used for designing and optimizing electronic circuits.
In this project you will build the basis for using reinforcement learning for electronic circuit design and optimization. In particular, you will wrap an existing electronic circuit simulator in a reinforcement learning environment such that it can be easily interacted with by standard reinforcement learning algorithms. You will then implement and evaluate standard reinforcement learning algorithms for designing and optimizing simple electronic circuits, e.g., current sources.
Students wanting to work on this topic are expected to have solid coding skills in Python and be curious to learn about reinforcement learning and apply it in a real-world application. Some background in electronic circuit design is advantageous but not a prerequisite.
Contact: Sebastian Tschiatschek
Algorithms for Finding Matched Molecular Pairs
The concept of matched molecular pairs (MMP) refers to two molecules, which have a large common substructure and differ only by a single local modification. Finding such pairs and analyzing their physical properties such as biological activity allows to attribute any differences to the specific structural change.
The problem can be formalized in terms of graph theory and is then closely related to variants of the classical graph isomorphism problem such as the maximum common subgraph problem and graph canonization. In this project, you will implement, develop and experimentally evaluate graph algorithms for finding MMPs. Several algorithmic challenges can be studied in more detail, e.g., the efficient computation of canonical forms in a dynamic setting or the use of index data structures and hashing techniques for acceleration.
Requirements: C++ or Java (no experience in chemistry necessary!)
Contact: Nils Kriege, Steffen Hirte
2D pharmacophore descriptors for machine learning and similarity searches
Pharmacophores are an abstract description of molecular features that are necessary for the molecular recognition of drugs by a biological macromolecule (usually proteins). The molecular features constituting a pharmacophore can be divided into different classes according to the nature of their non-bonding interaction with complementary parts of the macromolecular target receptor. Every molecule has a characteristic pharmacophoric feature pattern in terms of number, type and mutual distance of its features. Different molecules that are recognized by the same biological target usually show a high similarity regarding their feature pattern. A database search for molecules which have a pharmacophoric profile that is similar to a given query pharmacophore can thus reveal novel molecules that have a high chance to bind to the same receptor for which the query was created. To be able to search millions of molecules in acceptable time, a suitable in-memory pharmacophore representation is required which allows for a fast similarity assessment. In cheminformatics, an often employed representation of molecular structures which fulfills these requirements are fixed-size binary fingerprints (bit strings) where similarity calculations essentially boil down to fast logical bit-set operations.
The goal of this project is to implement such a binary encoding also for pharmacophores that can be used for fast pharmacophore similarity searches and for the development of machine learning models. The intended algorithm for converting an arbitrary pharmacophore into a fixed-size bit representation is exemplified in the following figure:
The implementation will be carried out on top of the open source cheminformatics toolkit CDPKit (https://github.com/aglanger/CDPKit) which already provides all required functionality for the I/O of molecular structures and pharmacophore generation.
Requirements: C++ (preferable) or Python, basic knowledge about the fundamentals of organic chemistry advantageous
Contact: Thomas Seidel, Nils Kriege
Causal inference for wind turbine extreme events [Bachelor or Master thesis]
This work will focus mainly on programme implementation of algorithms for causal detection among extreme event processes and their testing on real data. The causal inference will be tested among the wind turbine time series as well as other relevant meteorological time series, as provided by Austrian ZAMG (Die Zentralanstalt für Meteorologie und Geodynamik). More details to the interested student will be provided on request.
Contact: Katerina Schindlerova
Clustering of physiological, behavioral and climatic data of wild boars (Sus scrofa) [Bachelor or Master thesis]
This thesis will mainly focus on developing a data base combining data retrieved from bio-loggers (body temperature, heart rate, acceleration data, location data), reproductive success and climate data collected during a research project on wild boars at the Research Institute of Wildlife Ecology, University of Veterinary Medicine. Aim of this work is to get a functional data base which would allow researchers to access and evaluate data of different aspects of climate change on wild boar ecology in the long term.
If the student is interested, this basic approach can be broadened by using and adapting e.g. a machine learning process to allow the interpretation of acceleration data. We already used the approach of machine learning to classify video-taped behaviors (laying, walking, trotting, lactating) using acceleration data (collected via ear-tags). However, it would be very interesting whether the approach of machine learning could help us to detect daily and annual patterns of activity, affected by e.g. high ambient temperatures, in these data.
We would be happy to find an interested student to support the ecological evaluation of these data.
Contact: Claudia Bieber, Claudia Plant
Clustering of spatio-temporal climatological data [Master thesis]
This work will focus on the integration of massive amounts of data of heterogeneous spatial and temporal resolution of spatio-temporal climatological data provided by ZAMG. An effective information-theoretic objective function supporting different length of time series and different spatial resolutions will be applied. A joint low-dimensional vector space embedding of all data will be learned, which will be then used as a model for data compression by clustering. Measurement sites with similar characteristics are in this space close to each other and measurement sites with different characteristics will be far apart from each other in this vector space. Deep neural nets and greedy optimization techniques such as iterative least square methods will be considered to learn this space. A clustering algorithm will be developed, which works with this representation of the data and iteratively refines it during the clustering process.
Contact: Katerina Schindlerova, Claudia Plant
Reinforcement Learning from Implicit and Explicit Feedback
Recent advances in reinforcement learning have enabled super-human performance of AI agents in many applications, e.g., game play. In this project you will work on building theory and/or models for reinforcement learning from implicit and explicit feedback. In particular, you will consider the setting in which no rewards are observed during execution of the agent but only cumulative reward information is obtained at the end of the execution. As a practical example, assume playing a computer game in which you cannot observe the score while you play the game but only see your achieved score at the very end. This setting has important applications in human-in-the-loop settings in which feedback can be expensive to obtain.
Students working on this project need basic background knowledge in machine learning, solid programming skills in Python, and the desire to work on reinforcement learning.
Contact: Sebastian Tschiatschek
Machine Learning for Personalized Education [Practical Course or Bachelor Thesis]
In this project you will work on predicting students' responses to mathematical questions using machine learning models. This is important to assess a student's skills and provide personalized educational resources. We use real-world data provided by the "Diagnostic Questions: Predicting Student Responses and Measuring Question Quality" challenge at NeurIPS'20. The challenge consists of four different tasks from which a subset can be selected for the project, depending on background knowledge and programming skills.
Students working on this project need basic background knowledge in machine learning, programming skills in Python, and the desire to develop and test machine learning models.
Contact: Sebastian Tschiatschek
Probabilistic Models of Multimodal Distributions
Many real-world phenomena can only be accurately described by multimodal distributions, e.g., the hourly power consumption of a household (typically) is bimodal, with peaks in the morning and the evening. However, learning deep generative models of these multimodal distributions, e.g., using variational auto encoders, still remains challenging. In this project you will develop theory and/or models for multi-modal distributions based on variational autoencoders. You will evaluate these models on benchmark and real-world datasets.
Students working on this project need basic background knowledge in machine learning, programming skills in Python, and the desire to develop and test machine learning models.
Contact: Sebastian Tschiatschek
Reward Inference for Sequential Decision Making from Diverse and Implicit Feedback [Master Thesis]
Automated sequential decision making is an important application of machine learning systems in which such a system needs to select a sequence of actions step by step to optimize a reward/utility function. For instance, in autonomous driving, such a system needs to execute a sequence of steering, breaking and acceleration actions, or in a medical intensive care setting, such a system needs to execute a sequence of measurement and treatment actions.
One challenge in realizing such automated sequential decision making systems is the definition of the reward/utility function. For example, in autonomous driving it is hard to specify all the factors which define good driving behavior. In such settings, automatically inferring the reward/utility function from users’ feedback can be beneficial.
This project investigates approaches for reward/utility inference from diverse and implicit feedback, building on ideas for inverse reinforcement learning, active learning, implicit feedback, etc.
Interested students are expected to have solid mathematical and machine learning skills, and have experience in Python and deep learning (using PyTorch or TensorFlow).
Contact: Sebastian Tschiatschek
Imitation Learning Under Domain Mismatch
Reinforcement learning has been successfully used to solve certain challenging sequential decision making problems in recent years. The employed techniques commonly require (i) huge amounts of interactions with the envirnoment and (ii) clearly specified reward signals to work well. In many applications however, one or both of these requirements are not met. In such cases, imitation learning can be an efficient approach to sequential decision making problems: an expert demonstrates near-optimal behavior and a learning agent attempts to mimic this behavior.
This project considers imitation learning in settings in which there is some form of mismatch between the expert demonstrator and the learning agent. The scope of the project is to study how existing algorithms perform in this setting and proposes modifications to existing algorithms to achieve better performance.
Students wanting to work on this topic are expected to have a basic understanding of machine learning techniques, solid knowledge of Python and basic knowledge of deep learning libraries (PyTorch or TensorFlow).
Contact: Sebastian Tschiatschek
Selecting Sequences of Items for Non-monotone functions [Bachelor or Master Thesis]
Many applications involve the selection of sequences of items, e.g., recommender systems and personalization in MOOCs. Common to many of these applications is that the order of the selected items is not random but that items selected early, influence which items are selected later. The problem of selecting sequences of items has been studied in various settings, including that in which dependencies between items are modeled using graphs and monotone submodular functions.
This project aims at extending these settings to cover the case of non-monotone submodular functions, by proposing new algorithms and analyzing their properties. The findings are validated by an implementation of the proposed algorithm(s) and comparison agains reasonable baselines.
Students wanting to work on this topic are expected to have solid mathematical skills, a basic understanding of machine learning techniques, solid knowledge of Python and deep learning libraries (PyTorch or TensorFlow).
Contact: Sebastian Tschiatschek
Posterior Consistency in Partial Variational Autoencoders
Variational Autoencoders (VAEs) are powerful deep generative models that have been successfully applied in a wide range of machine learning applications. Recently, the Partial VAE (PVAE), a variant of VAEs that can process partially observed inputs has been proposed and its effectiveness for data imputation has been demonstrated. Key to the fast training of VAEs and PVAEs is the amortized prediction of posterior distributions from observations. In PVAEs, these posterior distributions are predicted from partial observations.
This project aims at studying the consistency of these posterior distributions for different patterns of missing data. The insights are used to create/train better inference models and thereby improve the quality of PVAEs.
Students wanting to work on this topic are expected to have solid mathematical skills, a basic understanding of machine learning techniques and good programming skills in Python.
Contact: Sebastian Tschiatschek
Evaluation of different nowcasting techniques and data bases for temperature nowcasting
Nowcasting in meteorology refers to short-time prediction of timeseries of e.g. a few minutes to hours into the future. For this short-time predictions the most recent observations are of importance as well as spatial relationships of upstream observations.
Different methods for nowcasting exist in the meteorological world, often physics of physics-statistical driven. Only a few studies focused on applying machine learning and data mining techniques. For the latter nowadays also crowd-sourced information can contain important information. Here, data of NetAtmo stations could provide valuable information. For the first part a study investigated already different machine learning algorithms for nowcasting of temperature using the Austrian Met-Service (ZAMG) data. The idea of the proposed thesis is to combine the already existing algorithms with the NetAtmo data by first clustering the ZAMG data and NetAtmo data and then develop an algorithm which incorporates the NetAtmo data. Important to note is that NetAtmo sites are prone to errors thus need to be cleaned beforehand.
The developed algorithm will be evaluated against two statistical forecasting methods, an analogue search based method and a model output statistics method.
Contact: Irene Schicker, Claudia Plant
Causal inference among climatological time series with extreme events [Master thesis]
This work will focus mainly on implementation of an algorithm for causal detection among processes as well as its testing on probability distributions with heavy tailed distributions and on climatological data provided by ZAMG (Die Zentralanstalt für Meteorologie und Geodynamik) following these distributions. More details to the interested student will be provided on request.
Contact: Katerina Schindlerova
Implementation of Data Mining Approach for Short-range Temperature Forecasts
Short-range forecasts of wind speeds (i.e., 1 - 72 hours into the future) and in particular nowcasting (i.e., very short-range forecasts with a time horizon of up to 6 hours) are vital for a wide range of applications. In contrast to wind, temperature typically changes gradually and may thus need a different setup than wind speed. Temperature involves daily fluctuations, which are well predictable but dependable on the daytime, which must be considered in the training dataset. Depending on the location, also rapid temperature changes may occur which are related, for example, to cold air pools or Föhn. Thus, also temperature highly depends on location (specific topography, prevailing weather conditions, and atmospheric dynamics). Depending on the application, we can either give point or area based predictions. Points refer to a particular location (e.g., a weather station) whereas spatial forecasts typically give a forecast for each grid cell over a region (i.e., each forecast is valid for the whole cell).
ZAMG (Zentralanstalt für Meteorologie und Geodynamik) employs (gridded) numerical weather prediction models in conjunction with observation data for short-range forecasting and a nowcasting system, INCA, for the prediction of meteorological parameters. Alternatively, machine learning methods are now being implemented. In particular, an artificial neural network (ANN) and a random forests (RF) are used in an experimental setup to show the skills of these methods and, possibly, be used as additional wind speed point forecasting method for the 10-meter wind. The existing methods can be used for temperature as well with the same training set, but the setup needs to be adapted.
The proposed student project shall address temperature forecasting (in two meters height) at meteorological observation sites by machine learning and data mining methods (e.g.: random forest, feed forward/BP artificial neural networks, kernel methods, etc.) and input feature selection of the training data set. It is possible to experiment with related meteorological parameters as well (e.g.: drew point temperature and relative humidity). Related work suggests to use back propagation neural network for predicting the 2-m drew point temperature and 2-m temperature. This can be used as a starting point to find a suitable selection of the training data for the current model and then extend the our current approaches by a back propagation neural network in order to set up a first prototype for temperature prediction by machine learning methods. The new model shall be tested on various scenarios (e.g.: different prevailing weather condition, locations, seasons) in order to compare the new data mining based model with the currently employed nowcasting system INCA.
The work is co-supervised by ZAMG's Section for numerical weather predictions (NWP) Applications. The developed method shall have a Python based frontend, C/C++ backend, and use csv or sqlite based meteorological data (provided by ZAMG) in order to align with other machine learning implementations running in our IT environment. Finally, the developed method will be set up in our development environment (Python 2.7/Linux 64-bit, multi-cored shared memory machine) to provide forecasts and validation of the method for selected test scenarios.
Contact: Petrina Papazek, Claudia Plant
Completed
- Wei Chen, Bachelor Thesis: "Mining Brain Networks", winter term 2022/2023
- Kejsi Hoxhallari, Bachelor Thesis: "Statistical validation and visualization of causal inference with extremes in wind-turbine data set", winter term 2022/2023
- Daan Scheepens, Master Thesis: "A deep convolutional RNN model for spatio-temporal prediction of wind speed extremes in the short-to-medium range for wind energy applications", winter term 2021/2022
- Yigit Berkay Bozkurt, Bachelor Thesis: "Anomaly Detection by Heterogenous Graphical Granger Causality and its Application to Climate Data", 2019
- Christina Pacher, Bachelor Thesis: "Clustering Weather Stations: A Clustering Application for Meteorological Data", summer term 2019
- Thomas Spendlhofer, Bachelor Thesis: "Evaluating the usage of Tensor Processing Units (TPUs) for unsupervised learning on the example of the k-means algorithm", summer term 2019
- Ernst Naschenweng, Bachelor Thesis: "A cache optimized implementation of the Floyd-Warshall Algorithm", summer term 2018
- Hermann Hinterhauser, Bachelor Thesis: "ITGC: Information-theoretic grid-based clustering", summer term 2018, accepted paper in EDBT 2019 (download available here)
- Mahmoud A. Ibrahim, Bachelor Thesis: "Parameter Free Mixed-Type Density-Based Clustering", winter term 2017/2018, accepted paper in DEXA 2018 (download available here)
- Markus Tschlatscher: "Space-Filling Curves for Cache Efficient LU Decomposition", winter term 2017/2018
- Theresa Fruhwuerth, Master Thesis: "Uncovering High Resolution Mass Spectrometry Patterns through Audio Fingerprinting and Periodicity Mining Algorithms: An Exploratory Analysis", summer term 2017
- Robert Fritze, PR1 "Combining spatial information and optimization for locating emergency medical service stations: A case study for Lower Austria", summer term 2017
- Alexander Pfundner, PR2 "Integration of Density-based and Partitioning-based Clustering Methods", summer term 2017
- Anton Kovác, Katerina Hlavácková-Schindler, Erasmus project, "Graphical Granger Causality for Detection Temporal Anomalies in EEG Data", winter term 2016/2017 (download available here)