@phdthesis{Stojanovic2021,
  author    = {Stojanovic, Vladeta},
  title     = {Digital twins for indoor built environments},
  doi       = {10.25932/publishup-50913},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-509134},
  school      = {Universit{\"a}t Potsdam},
  pages     = {xxiii, 181},
  year      = {2021},
  abstract  = {One of the key challenges in modern Facility Management (FM) is to digitally reflect the current state of the built environment, referred to as-is or as-built versus as-designed representation. While the use of Building Information Modeling (BIM) can address the issue of digital representation, the generation and maintenance of BIM data requires a considerable amount of manual work and domain expertise. Another key challenge is being able to monitor the current state of the built environment, which is used to provide feedback and enhance decision making. The need for an integrated solution for all data associated with the operational life cycle of a building is becoming more pronounced as practices from Industry 4.0 are currently being evaluated and adopted for FM use. This research presents an approach for digital representation of indoor environments in their current state within the life cycle of a given building. Such an approach requires the fusion of various sources of digital data. The key to solving such a complex issue of digital data integration, processing and representation is with the use of a Digital Twin (DT). A DT is a digital duplicate of the physical environment, states, and processes. A DT fuses as-designed and as-built digital representations of built environment with as-is data, typically in the form of floorplans, point clouds and BIMs, with additional information layers pertaining to the current and predicted states of an indoor environment or a complete building (e.g., sensor data). The design, implementation and initial testing of prototypical DT software services for indoor environments is presented and described. These DT software services are implemented within a service-oriented paradigm, and their feasibility is presented through functioning and tested key software components within prototypical Service-Oriented System (SOS) implementations. The main outcome of this research shows that key data related to the built environment can be semantically enriched and combined to enable digital representations of indoor environments, based on the concept of a DT. Furthermore, the outcomes of this research show that digital data, related to FM and Architecture, Construction, Engineering, Owner and Occupant (AECOO) activity, can be combined, analyzed and visualized in real-time using a service-oriented approach. This has great potential to benefit decision making related to Operation and Maintenance (O\&M) procedures within the scope of the post-construction life cycle stages of typical office buildings.},
  language  = {en}
}
@phdthesis{Zieger2017,
  author    = {Zieger, Tobias},
  title     = {Self-adaptive data quality},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-410573},
  school      = {Universit{\"a}t Potsdam},
  pages     = {vii, 125},
  year      = {2017},
  abstract  = {Carrying out business processes successfully is closely linked to the quality of the data inventory in an organization. Lacks in data quality lead to problems: Incorrect address data prevents (timely) shipments to customers. Erroneous orders lead to returns and thus to unnecessary effort. Wrong pricing forces companies to miss out on revenues or to impair customer satisfaction. If orders or customer records cannot be retrieved, complaint management takes longer. Due to erroneous inventories, too few or too much supplies might be reordered. A special problem with data quality and the reason for many of the issues mentioned above are duplicates in databases. Duplicates are different representations of same real-world objects in a dataset. However, these representations differ from each other and are for that reason hard to match by a computer. Moreover, the number of required comparisons to find those duplicates grows with the square of the dataset size. To cleanse the data, these duplicates must be detected and removed. Duplicate detection is a very laborious process. To achieve satisfactory results, appropriate software must be created and configured (similarity measures, partitioning keys, thresholds, etc.). Both requires much manual effort and experience. This thesis addresses automation of parameter selection for duplicate detection and presents several novel approaches that eliminate the need for human experience in parts of the duplicate detection process. A pre-processing step is introduced that analyzes the datasets in question and classifies their attributes semantically. Not only do these annotations help understanding the respective datasets, but they also facilitate subsequent steps, for example, by selecting appropriate similarity measures or normalizing the data upfront. This approach works without schema information. Following that, we show a partitioning technique that strongly reduces the number of pair comparisons for the duplicate detection process. The approach automatically finds particularly suitable partitioning keys that simultaneously allow for effective and efficient duplicate retrieval. By means of a user study, we demonstrate that this technique finds partitioning keys that outperform expert suggestions and additionally does not need manual configuration. Furthermore, this approach can be applied independently of the attribute types. To measure the success of a duplicate detection process and to execute the described partitioning approach, a gold standard is required that provides information about the actual duplicates in a training dataset. This thesis presents a technique that uses existing duplicate detection results and crowdsourcing to create a near gold standard that can be used for the purposes above. Another part of the thesis describes and evaluates strategies how to reduce these crowdsourcing costs and to achieve a consensus with less effort.},
  language  = {en}
}
@phdthesis{Meier2017,
  author    = {Meier, Sebastian},
  title     = {Personal Big Data},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-406696},
  school      = {Universit{\"a}t Potsdam},
  pages     = {xxiv, 133},
  year      = {2017},
  abstract  = {Many users of cloud-based services are concerned about questions of data privacy. At the same time, they want to benefit from smart data-driven services, which require insight into a person's individual behaviour. The modus operandi of user modelling is that data is sent to a remote server where the model is constructed and merged with other users' data. This thesis proposes selective cloud computing, an alternative approach, in which the user model is constructed on the client-side and only an abstracted generalised version of the model is shared with the remote services. In order to demonstrate the applicability of this approach, the thesis builds an exemplary client-side user modelling technique. As this thesis is carried out in the area of Geoinformatics and spatio-temporal data is particularly sensitive, the application domain for this experiment is the analysis and prediction of a user's spatio-temporal behaviour. The user modelling technique is grounded in an innovative conceptual model, which builds upon spatial network theory combined with time-geography. The spatio-temporal constraints of time-geography are applied to the network structure in order to create individual spatio-temporal action spaces. This concept is translated into a novel algorithmic user modelling approach which is solely driven by the user's own spatio-temporal trajectory data that is generated by the user's smartphone. While modern smartphones offer a rich variety of sensory data, this thesis only makes use of spatio-temporal trajectory data, enriched by activity classification, as the input and foundation for the algorithmic model. The algorithmic model consists of three basal components: locations (vertices), trips (edges), and clusters (neighbourhoods). After preprocessing the incoming trajectory data in order to identify locations, user feedback is used to train an artificial neural network to learn temporal patterns for certain location types (e.g. work, home, bus stop, etc.). This Artificial Neural Network (ANN) is used to automatically detect future location types by their spatio-temporal patterns. The same is done in order to predict the duration of stay at a certain location. Experiments revealed that neural nets were the most successful statistical and machine learning tool to detect those patterns. The location type identification algorithm reached an accuracy of 87.69\%, the duration prediction on binned data was less successful and deviated by an average of 0.69 bins. A challenge for the location type classification, as well as for the subsequent components, was the imbalance of trips and connections as well as the low accuracy of the trajectory data. The imbalance is grounded in the fact that most users exhibit strong habitual patterns (e.g. home > work), while other patterns are rather rare by comparison. The accuracy problem derives from the energy-saving location sampling mode, which creates less accurate results. Those locations are then used to build a network that represents the user's spatio-temporal behaviour. An initial untrained ANN to predict movement on the network only reached 46\% average accuracy. Only lowering the number of included edges, focusing on more common trips, increased the performance. In order to further improve the algorithm, the spatial trajectories were introduced into the predictions. To overcome the accuracy problem, trips between locations were clustered into so-called spatial corridors, which were intersected with the user's current trajectory. The resulting intersected trips were ranked through a k-nearest-neighbour algorithm. This increased the performance to 56\%. In a final step, a combination of a network and spatial clustering algorithm was built in order to create clusters, therein reducing the variety of possible trips. By only predicting the destination cluster instead of the exact location, it is possible to increase the performance to 75\% including all classes. A final set of components shows in two exemplary ways how to deduce additional inferences from the underlying spatio-temporal data. The first example presents a novel concept for predicting the 'potential memorisation index' for a certain location. The index is based on a cognitive model which derives the index from the user's activity data in that area. The second example embeds each location in its urban fabric and thereby enriches its cluster's metadata by further describing the temporal-semantic activity in an area (e.g. going to restaurants at noon). The success of the client-side classification and prediction approach, despite the challenges of inaccurate and imbalanced data, supports the claimed benefits of the client-side modelling concept. Since modern data-driven services at some point do need to receive user data, the thesis' computational model concludes with a concept for applying generalisation to semantic, temporal, and spatial data before sharing it with the remote service in order to comply with the overall goal to improve data privacy. In this context, the potentials of ensemble training (in regards to ANNs) are discussed in order to highlight the potential of only sharing the trained ANN instead of the raw input data. While the results of our evaluation support the assets of the proposed framework, there are two important downsides of our approach compared to server-side modelling. First, both of these server-side advantages are rooted in the server's access to multiple users' data. This allows a remote service to predict spatio-in the user-specific data, which represents the second downside. While minor classes will likely be minor classes in a bigger dataset as well, for each class, there will still be more variety than in the user-specific dataset. The author emphasises that the approach presented in this work holds the potential to change the privacy paradigm in modern data-driven services. Finding combinations of client- and server-side modelling could prove a promising new path for data-driven innovation. Beyond the technological perspective, throughout the thesis the author also offers a critical view on the data- and technology-driven development of this work. By introducing the client-side modelling with user-specific artificial neural networks, users generate their own algorithm. Those user-specific algorithms are influenced less by generalised biases or developers' prejudices. Therefore, the user develops a more diverse and individual perspective through his or her user model. This concept picks up the idea of critical cartography, which questions the status quo of how space is perceived and represented.},
  language  = {en}
}
@phdthesis{Dick2016,
  author    = {Dick, Uwe},
  title     = {Discriminative Classification Models for Internet Security},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-102593},
  school      = {Universit{\"a}t Potsdam},
  pages     = {x, 57},
  year      = {2016},
  abstract  = {Services that operate over the Internet are under constant threat of being exposed to fraudulent use. Maintaining good user experience for legitimate users often requires the classification of entities as malicious or legitimate in order to initiate countermeasures. As an example, inbound email spam filters decide for spam or non-spam. They can base their decision on both the content of each email as well as on features that summarize prior emails received from the sending server. In general, discriminative classification methods learn to distinguish positive from negative entities. Each decision for a label may be based on features of the entity and related entities. When labels of related entities have strong interdependencies---as can be assumed e.g. for emails being delivered by the same user---classification decisions should not be made independently and dependencies should be modeled in the decision function. This thesis addresses the formulation of discriminative classification problems that are tailored for the specific demands of the following three Internet security applications. Theoretical and algorithmic solutions are devised to protect an email service against flooding of user inboxes, to mitigate abusive usage of outbound email servers, and to protect web servers against distributed denial of service attacks. In the application of filtering an inbound email stream for unsolicited emails, utilizing features that go beyond each individual email's content can be valuable. Information about each sending mail server can be aggregated over time and may help in identifying unwanted emails. However, while this information will be available to the deployed email filter, some parts of the training data that are compiled by third party providers may not contain this information. The missing features have to be estimated at training time in order to learn a classification model. In this thesis an algorithm is derived that learns a decision function that integrates over a distribution of values for each missing entry. The distribution of missing values is a free parameter that is optimized to learn an optimal decision function. The outbound stream of emails of an email service provider can be separated by the customer IDs that ask for delivery. All emails that are sent by the same ID in the same period of time are related, both in content and in label. Hijacked customer accounts may send batches of unsolicited emails to other email providers, which in turn might blacklist the sender's email servers after detection of incoming spam emails. The risk of being blocked from further delivery depends on the rate of outgoing unwanted emails and the duration of high spam sending rates. An optimization problem is developed that minimizes the expected cost for the email provider by learning a decision function that assigns a limit on the sending rate to customers based on the each customer's email stream. Identifying attacking IPs during HTTP-level DDoS attacks allows to block those IPs from further accessing the web servers. DDoS attacks are usually carried out by infected clients that are members of the same botnet and show similar traffic patterns. HTTP-level attacks aim at exhausting one or more resources of the web server infrastructure, such as CPU time. If the joint set of attackers cannot increase resource usage close to the maximum capacity, no effect will be experienced by legitimate users of hosted web sites. However, if the additional load raises the computational burden towards the critical range, user experience will degrade until service may be unavailable altogether. As the loss of missing one attacker depends on block decisions for other attackers---if most other attackers are detected, not blocking one client will likely not be harmful---a structured output model has to be learned. In this thesis an algorithm is developed that learns a structured prediction decoder that searches the space of label assignments, guided by a policy. Each model is evaluated on real-world data and is compared to reference methods. The results show that modeling each classification problem according to the specific demands of the task improves performance over solutions that do not consider the constraints inherent to an application.},
  language  = {en}
}
@phdthesis{Sawade2012,
  author    = {Sawade, Christoph},
  title     = {Active evaluation of predictive models},
  publisher = {Universit{\"a}tsverlag Potsdam},
  address   = {Potsdam},
  isbn      = {978-3-86956-255-1},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-65583},
  school      = {Universit{\"a}t Potsdam},
  pages     = {ix, 157},
  year      = {2012},
  abstract  = {The field of machine learning studies algorithms that infer predictive models from data. Predictive models are applicable for many practical tasks such as spam filtering, face and handwritten digit recognition, and personalized product recommendation. In general, they are used to predict a target label for a given data instance. In order to make an informed decision about the deployment of a predictive model, it is crucial to know the model's approximate performance. To evaluate performance, a set of labeled test instances is required that is drawn from the distribution the model will be exposed to at application time. In many practical scenarios, unlabeled test instances are readily available, but the process of labeling them can be a time- and cost-intensive task and may involve a human expert. This thesis addresses the problem of evaluating a given predictive model accurately with minimal labeling effort. We study an active model evaluation process that selects certain instances of the data according to an instrumental sampling distribution and queries their labels. We derive sampling distributions that minimize estimation error with respect to different performance measures such as error rate, mean squared error, and F-measures. An analysis of the distribution that governs the estimator leads to confidence intervals, which indicate how precise the error estimation is. Labeling costs may vary across different instances depending on certain characteristics of the data. For instance, documents differ in their length, comprehensibility, and technical requirements; these attributes affect the time a human labeler needs to judge relevance or to assign topics. To address this, the sampling distribution is extended to incorporate instance-specific costs. We empirically study conditions under which the active evaluation processes are more accurate than a standard estimate that draws equally many instances from the test distribution. We also address the problem of comparing the risks of two predictive models. The standard approach would be to draw instances according to the test distribution, label the selected instances, and apply statistical tests to identify significant differences. Drawing instances according to an instrumental distribution affects the power of a statistical test. We derive a sampling procedure that maximizes test power when used to select instances, and thereby minimizes the likelihood of choosing the inferior model. Furthermore, we investigate the task of comparing several alternative models; the objective of an evaluation could be to rank the models according to the risk that they incur or to identify the model with lowest risk. An experimental study shows that the active procedure leads to higher test power than the standard test in many application domains. Finally, we study the problem of evaluating the performance of ranking functions, which are used for example for web search. In practice, ranking performance is estimated by applying a given ranking model to a representative set of test queries and manually assessing the relevance of all retrieved items for each query. We apply the concepts of active evaluation and active comparison to ranking functions and derive optimal sampling distributions for the commonly used performance measures Discounted Cumulative Gain and Expected Reciprocal Rank. Experiments on web search engine data illustrate significant reductions in labeling costs.},
  language  = {en}
}
@phdthesis{Brueckner2012,
  author    = {Br{\"u}ckner, Michael},
  title     = {Prediction games : machine learning in the presence of an adversary},
  publisher = {Universit{\"a}tsverlag Potsdam},
  address   = {Potsdam},
  isbn      = {978-3-86956-203-2},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-60375},
  school      = {Universit{\"a}t Potsdam},
  pages     = {x, 121},
  year      = {2012},
  abstract  = {In many applications one is faced with the problem of inferring some functional relation between input and output variables from given data. Consider, for instance, the task of email spam filtering where one seeks to find a model which automatically assigns new, previously unseen emails to class spam or non-spam. Building such a predictive model based on observed training inputs (e.g., emails) with corresponding outputs (e.g., spam labels) is a major goal of machine learning. Many learning methods assume that these training data are governed by the same distribution as the test data which the predictive model will be exposed to at application time. That assumption is violated when the test data are generated in response to the presence of a predictive model. This becomes apparent, for instance, in the above example of email spam filtering. Here, email service providers employ spam filters and spam senders engineer campaign templates such as to achieve a high rate of successful deliveries despite any filters. Most of the existing work casts such situations as learning robust models which are unsusceptible against small changes of the data generation process. The models are constructed under the worst-case assumption that these changes are performed such to produce the highest possible adverse effect on the performance of the predictive model. However, this approach is not capable to realistically model the true dependency between the model-building process and the process of generating future data. We therefore establish the concept of prediction games: We model the interaction between a learner, who builds the predictive model, and a data generator, who controls the process of data generation, as an one-shot game. The game-theoretic framework enables us to explicitly model the players' interests, their possible actions, their level of knowledge about each other, and the order at which they decide for an action. We model the players' interests as minimizing their own cost function which both depend on both players' actions. The learner's action is to choose the model parameters and the data generator's action is to perturbate the training data which reflects the modification of the data generation process with respect to the past data. We extensively study three instances of prediction games which differ regarding the order in which the players decide for their action. We first assume that both player choose their actions simultaneously, that is, without the knowledge of their opponent's decision. We identify conditions under which this Nash prediction game has a meaningful solution, that is, a unique Nash equilibrium, and derive algorithms that find the equilibrial prediction model. As a second case, we consider a data generator who is potentially fully informed about the move of the learner. This setting establishes a Stackelberg competition. We derive a relaxed optimization criterion to determine the solution of this game and show that this Stackelberg prediction game generalizes existing prediction models. Finally, we study the setting where the learner observes the data generator's action, that is, the (unlabeled) test data, before building the predictive model. As the test data and the training data may be governed by differing probability distributions, this scenario reduces to learning under covariate shift. We derive a new integrated as well as a two-stage method to account for this data set shift. In case studies on email spam filtering we empirically explore properties of all derived models as well as several existing baseline methods. We show that spam filters resulting from the Nash prediction game as well as the Stackelberg prediction game in the majority of cases outperform other existing baseline methods.},
  language  = {en}
}
@phdthesis{Bickel2008,
  author    = {Bickel, Steffen},
  title     = {Learning under differing training and test distributions},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-33331},
  school      = {Universit{\"a}t Potsdam},
  year      = {2008},
  abstract  = {One of the main problems in machine learning is to train a predictive model from training data and to make predictions on test data. Most predictive models are constructed under the assumption that the training data is governed by the exact same distribution which the model will later be exposed to. In practice, control over the data collection process is often imperfect. A typical scenario is when labels are collected by questionnaires and one does not have access to the test population. For example, parts of the test population are underrepresented in the survey, out of reach, or do not return the questionnaire. In many applications training data from the test distribution are scarce because they are difficult to obtain or very expensive. Data from auxiliary sources drawn from similar distributions are often cheaply available. This thesis centers around learning under differing training and test distributions and covers several problem settings with different assumptions on the relationship between training and test distributions-including multi-task learning and learning under covariate shift and sample selection bias. Several new models are derived that directly characterize the divergence between training and test distributions, without the intermediate step of estimating training and test distributions separately. The integral part of these models are rescaling weights that match the rescaled or resampled training distribution to the test distribution. Integrated models are studied where only one optimization problem needs to be solved for learning under differing distributions. With a two-step approximation to the integrated models almost any supervised learning algorithm can be adopted to biased training data. In case studies on spam filtering, HIV therapy screening, targeted advertising, and other applications the performance of the new models is compared to state-of-the-art reference methods.},
  language  = {en}
}
@phdthesis{Dornhege2006,
  author    = {Dornhege, Guido},
  title     = {Increasing information transfer rates for brain-computer interfacing},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-7690},
  school      = {Universit{\"a}t Potsdam},
  year      = {2006},
  abstract  = {The goal of a Brain-Computer Interface (BCI) consists of the development of a unidirectional interface between a human and a computer to allow control of a device only via brain signals. While the BCI systems of almost all other groups require the user to be trained over several weeks or even months, the group of Prof. Dr. Klaus-Robert M{\"u}ller in Berlin and Potsdam, which I belong to, was one of the first research groups in this field which used machine learning techniques on a large scale. The adaptivity of the processing system to the individual brain patterns of the subject confers huge advantages for the user. Thus BCI research is considered a hot topic in machine learning and computer science. It requires interdisciplinary cooperation between disparate fields such as neuroscience, since only by combining machine learning and signal processing techniques based on neurophysiological knowledge will the largest progress be made. In this work I particularly deal with my part of this project, which lies mainly in the area of computer science. I have considered the following three main points: <b>Establishing a performance measure based on information theory:</b> I have critically illuminated the assumptions of Shannon's information transfer rate for application in a BCI context. By establishing suitable coding strategies I was able to show that this theoretical measure approximates quite well to what is practically achieveable. <b>Transfer and development of suitable signal processing and machine learning techniques:</b> One substantial component of my work was to develop several machine learning and signal processing algorithms to improve the efficiency of a BCI. Based on the neurophysiological knowledge that several independent EEG features can be observed for some mental states, I have developed a method for combining different and maybe independent features which improved performance. In some cases the performance of the combination algorithm outperforms the best single performance by more than 50 \%. Furthermore, I have theoretically and practically addressed via the development of suitable algorithms the question of the optimal number of classes which should be used for a BCI. It transpired that with BCI performances reported so far, three or four different mental states are optimal. For another extension I have combined ideas from signal processing with those of machine learning since a high gain can be achieved if the temporal filtering, i.e., the choice of frequency bands, is automatically adapted to each subject individually. <b>Implementation of the Berlin brain computer interface and realization of suitable experiments:</b> Finally a further substantial component of my work was to realize an online BCI system which includes the developed methods, but is also flexible enough to allow the simple realization of new algorithms and ideas. So far, bitrates of up to 40 bits per minute have been achieved with this system by absolutely untrained users which, compared to results of other groups, is highly successful.},
  subject      = {Kybernetik},
  language  = {en}
}