Search

Machine learning and argumentation can potentially greatly benefit from each other. Combining deep classifiers with knowledge expressed in the form of rules and constraints allows one to leverage different forms of abstractions within argumentation mining. Argumentation for machine learning can yield argumentation-based learning methods where the machine and the user argue about the learned model with the common goal of providing results of maximum utility to the user. Unfortunately, both directions are currently rather challenging. For instance, combining deep neural models with logic typically only yields deterministic results, while combining probabilistic models with logic often results in intractable inference. Therefore, we review a novel deep but tractable model for conditional probability distributions that can harness the expressive power of universal function approximators such as neural networks while still maintaining a wide range of tractable inference routines. While this new model has shown appealing performance in classification tasks, humans cannot easily understand the reasons for its decision. Therefore, we also review our recent efforts on how to “argue” with deep models. On synthetic and real data we illustrate how “arguing” with a deep model about its explanations can actually help to revise the model, if it is right for the wrong reasons.

Deep neural networks have demonstrated excellent performances in many real-world applications. Unfortunately, they may show Clever Hans-like behaviour (making use of confounding factors within datasets) to achieve high performance. In this work we introduce the novel learning setting of explanatory interactive learning and illustrate its benefits on a plant phenotyping research task. Explanatory interactive learning adds the scientist into the training loop, who interactively revises the original model by providing feedback on its explanations. Our experimental results demonstrate that explanatory interactive learning can help to avoid Clever Hans moments in machine learning and encourages (or discourages, if appropriate) trust in the underlying model. Deep learning approaches can show excellent performance but still have limited practical use if they learn to predict based on confounding factors in a dataset, for instance text labels in the corner of images. By using an explanatory interactive learning approach, with a human expert in the loop during training, it becomes possible to avoid predictions based on confounding factors.

While deep networks have been enormously successful, they rely on flat-feature vector representations. Using them in structured domains requires significant feature engineering. Such domains rely on relational representations to capture complex relationships between entities and their attributes. Thus, we consider the problem of learning neural networks for relational data. We distinguish ourselves from current approaches that rely on expert hand-coded rules by learning higher-order random-walk features to capture local structural interactions and the resulting network architecture. We further exploit parameter tying, where instances of the same rule share parameters. Our experimental results demonstrate the effectiveness of the proposed approach over multiple neural net baselines as well as state-of-the-art relational models.

With data becoming more and more complex, the standard tabular data format often does not suffice to represent datasets. Richer representations, such as relational ones, are needed. However, a relational representation opens a much larger space of possible descriptors (features) of the examples that are to be classified. Consequently, it is important to assess which features are relevant (and to what extent) for predicting the target. In this work, we propose a novel relational feature ranking method that is based on our novel version of gradient-boosted relational trees and extends the Genie3 score towards relational data. By running the algorithm on six well-known benchmark problems, we show that it yields meaningful feature rankings, provided that the underlying classifier can learn the target concept successfully.

The typical approach for learned DBMS components is to capture the behavior by running a representative set of queries and use the observations to train a machine learning model. This workload-driven approach, however, has two major downsides. First, collecting the training data can be very expensive, since all queries need to be executed on potentially large databases. Second, training data has to be recollected when the workload and the data changes. To overcome these limitations, we take a different route: we propose to learn a pure data-driven model that can be used for different tasks such as query answering or cardinality estimation. This data-driven model also supports ad-hoc queries and updates of the data without the need of full retraining when the workload or data changes. Indeed, one may now expect that this comes at a price of lower accuracy since workload-driven models can make use of more information. However, this is not the case. The results of our empirical evaluation demonstrate that our data-driven approach not only provides better accuracy than state-of-the-art learned components but also generalizes better to unseen queries.

Although interactive learning puts the user into the loop, the learner remains mostly a black box for the user. Understanding the reasons behind predictions and queries is important when assessing how the learner works and, in turn, trust. Consequently, we propose the novel framework of explanatory interactive learning where, in each step, the learner explains its query to the user, and the user interacts by both answering the query and correcting the explanation. We demonstrate that this can boost the predictive and explanatory powers of, and the trust into, the learned model, using text (e.g. SVMs) and image classification (e.g. neural networks) experiments as well as a user study.

Determination and characterization of resistance reactions of crops against fungal pathogens are essential to select resistant genotypes. In plant breeding, phenotyping of genotypes is realized by time consuming and expensive visual plant ratings. During resistance reactions and during pathogenesis plants initiate different structural and biochemical defence mechanisms, which partly affect the optical properties of plant organs. Recently, intensive research has been conducted to develop innovative optical methods for an assessment of compatible and incompatible plant pathogen interaction. These approaches, combining classical phytopathology or microbiology with technology driven methods — such as sensors, robotics, machine learning, and artificial intelligence — are summarized by the term digital phenotyping. In contrast to common visual rating, detection and assessment methods, optical sensors in combination with advanced data analysis methods are able to retrieve pathogen induced changes in the physiology of susceptible or resistant plants non-invasively and objectively. Phenotyping disease resistance aims different tasks. In an early breeding step, a qualitative assessment and characterization of specific resistance action is aimed to link it, for example, to a genetic marker. Later, during greenhouse and field screening, the assessment of the level of susceptibility of different genotypes is relevant. Within this review, recent advances of digital phenotyping technologies for the detection of subtle resistance reactions and resistance breeding are highlighted and methodological requirements are critically discussed.

The automatic interpretation of 3D point clouds for building reconstruction is a challenging task. The interpretation process requires highly structured models representing semantics. Formal grammars can describe structures as well as the parameters of buildings and their parts. We propose a novel approach for the automatic learning of weighted attributed context-free grammar rules for 3D building reconstruction, supporting the laborious manual design of rules. We separate structure from parameter learning. Specific Support Vector Machines (SVMs) are used to generate a weighted context-free grammar and predict structured outputs such as parse trees. The grammar is extended by parameters and constraints, which are learned based on a statistical relational learning method using Markov Logic Networks (MLNs). MLNs enforce the topological and geometric constraints. MLNs address uncertainty explicitly and provide probabilistic inference. They are able to deal with partial observations caused by occlusions. Uncertain projective geometry is used to deal with the uncertainty of the observations. Learning is based on a large building database covering different building styles and facade structures. In particular, a treebank that has been derived from the database is employed for structure learning.

We consider the problem of learning Relational Logistic Regression (RLR). Unlike standard logistic regression, the features of RLRs are first-order formulae with associated weight vectors instead of scalar weights. We turn the problem of learning RLR to learning these vector-weighted formulae and develop a learning algorithm based on the recently successful functional-gradient boosting methods for probabilistic logic models. We derive the functional gradients and show how weights can be learned simultaneously in an efficient manner. Our empirical evaluation on standard and novel data sets demonstrates the superiority of our approach over other methods for learning RLR.

Magnetic resonance imaging (MRI) has emerged as an important tool to identify intermediate biomarkers of Alzheimer's disease (AD) due to its ability to measure regional changes in the brain that are thought to reflect disease severity and progression. In this paper, we set out a novel pipeline that uses volumetric MRI data collected from different subjects as input and classifies them into one of three classes: AD, mild cognitive impairment (MCI) and cognitively normal (CN). Our pipeline consists of three stages-(1) a segmentation layer where brain MRI data is divided into clinically relevant regions; (2) a classification layer that uses relational learning algorithms to make pairwise predictions between the three classes; and (3) a combination layer that combines the results of the different classes to obtain the final classification. One of the key features of our proposed approach is that it allows for domain expert's knowledge to guide the learning in all the layers. We evaluate our pipeline on 397 patients acquired from the Alzheimer's Disease Neuroimaging Initiative and demonstrate that it obtains state-of-the-art performance with minimal feature engineering.

Over the past decade, exploiting relations and symmetries within probabilistic models has been proven to be surprisingly effective at solving large scale data mining problems. One of the key operations inside these lifted approaches is counting be it for parameter/structure learning or for efficient inference. Typically, however, they just count exploiting the logical structure using adhoc operators. This paper investigates whether ‘Compilation to Graph Databases’ could be a practical technique for scaling lifted probabilistic inference and learning methods. We demonstrate that the proposed approach achieves reasonable speed-ups for both inference and learning, without sacrificing performance.

Modern phenotyping and plant disease detection methods, based on optical sensors and information technology, provide promising approaches to plant research and precision farming. In particular, hyperspectral imaging have been found to reveal physiological and structural characteristics in plants and to allow for tracking physiological dynamics due to environmental effects. In this work, we present an approach to plant phenotyping that integrates non-invasive sensors, computer vision, as well as data mining techniques and allows for monitoring how plants respond to stress. To uncover latent hyperspectral characteristics of diseased plants reliably and in an easy-to-understand way, we “wordify” the hyperspectral images, i.e., we turn the images into a corpus of text documents. Then, we apply probabilistic topic models, a well-established natural language processing technique that identifies content and topics of documents. Based on recent regularized topic models, we demonstrate that one can track automatically the development of three foliar diseases of barley. We also present a visualization of the topics that provides plant scientists an intuitive tool for hyperspectral imaging. In short, our analysis and visualization of characteristic topics found during symptom development and disease progress reveal the hyperspectral language of plant diseases.

Dependency networks approximate a joint probability distribution over multiple random variables as a product of conditional distributions. Relational Dependency Networks (RDNs) are graphical models that extend dependency networks to relational domains. This higher expressivity, however, comes at the expense of a more complex model-selection problem: an unbounded number of relational abstraction levels might need to be explored. Whereas current learning approaches for RDNs learn a single probability tree per random variable, we propose to turn the problem into a series of relational function-approximation problems using gradient-based boosting. In doing so, one can easily induce highly complex features over several iterations and in turn estimate quickly a very expressive model. Our experimental results in several different data sets show that this boosting method results in efficient learning of RDNs when compared to state-of-the-art statistical relational learning approaches. ispartof: Machine Learning vol:86 issue:1 pages:25-56 ispartof: location:Florence, Italy status: published

The prediction of both, vehicular traffic and communication connectivity are important research topics. In this paper, we propose the usage of innovative machine learning approaches for these objectives. For this purpose, Poisson Dependency Networks (PDNs) are introduced to enhance the prediction quality of vehicular traffic flows. The machine learning model is fitted based on empirical vehicular traffic data. The results show that PDNs enable a significantly better short-term prediction in comparison to a prediction based on the physics of traffic. To combine vehicular traffic with cellular communication networks, a correlation between connectivity indicators and vehicular traffic flow is shown based on measurement results. This relationship is leveraged by means of Poisson regression trees in both directions, and hence, enabling the prediction of both types of network utilization.

Understanding the response dynamics of plants to biotic stress is essential to improve management practices and breeding strategies of crops and thus to proceed towards a more sustainable agriculture in the coming decades. In this context, hyperspectral imaging offers a particularly promising approach since it provides non-destructive measurements of plants correlated with internal structure and biochemical compounds. In this paper, we present a cascade of data mining techniques for fast and reliable data-driven sketching of complex hyperspectral dynamics in plant science and plant phenotyping. To achieve this, we build on top of a recent linear time matrix factorization technique, called Simplex Volume Maximization, in order to automatically discover archetypal hyperspectral signatures that are characteristic for particular diseases. The methods were applied on a data set of barley leaves (Hordeum vulgare) diseased with foliar plant pathogens Pyrenophora teres, Puccinia hordei and Blumeria graminis hordei. Towards more intuitive visualizations of plant disease dynamics, we use the archetypal signatures to create structured summaries that are inspired by metro maps, i.e. schematic diagrams of public transport networks. Metro maps of plant disease dynamics produced on several real-world data sets conform to plant physiological knowledge and explicitly illustrate the interaction between diseases and plants. Most importantly, they provide an abstract and interpretable view on plant disease progression.

One of the challenges to information extraction is the requirement of human annotated examples, commonly called gold-standard examples. Many successful approaches alleviate this problem by employing some form of distant supervision, i.e., look into knowledge bases such as Freebase as a source of supervision to create more examples. While this is perfectly reasonable, most distant supervision methods rely on a hand-coded background knowledge that explicitly looks for patterns in text. For example, they assume all sentences containing Person X and Person Y are positive examples of the relation marriedX, Y. In this work, we take a different approach --- we infer weakly supervised examples for relations from models learned by using knowledge outside the natural language task. We argue that this method creates more robust examples that are particularly useful when learning the entire information-extraction model the structure and parameters. We demonstrate on three domains that this form of weak supervision yields superior results when learning structure compared to using distant supervision labels or a smaller set of gold-standard labels.

We consider the problem of learning probabilistic models from relational data. One of the key issues with relational data is class imbalance where the number of negative examples far outnumbers the number of positive examples. The common approach for dealing with this problem is the use of sub-sampling of negative examples. We, on the other hand, consider a soft margin approach that explicitly trades off between the false positives and false negatives. We apply this approach to the recently successful formalism of relational functional gradient boosting. Specifically, we modify the objective function of the learning problem to explicitly include the trade-off between false positives and negatives. We show empirically that this approach is more successful in handling the class imbalance problem than the original framework that weighed all the examples equally.

The past few years have witnessed an significant interest in probabilistic logic learning, i.e. in research lying at the intersection of probabilistic reasoning, logical representations, and machine learning. A rich variety of different formalisms and learning techniques have been developed. This paper provides an introductory survey and overview of the state-of-the-art in probabilistic logic learning through the identification of a number of important probabilistic, logical and learning concepts.

The learning approaches presented in the last two chapters employed the closed-world assumption i.e., whatever that is not observed in the data is assumed to be false. In this chapter, we relax this assumption and derive a boosting algorithm that can effectively work with missing data. The derivation is independent of the model and hence we will discuss about adapting it for RDNs and MLNs. As with other chapters, we will conclude with empirical evaluation on the SRL data sets.

The problem of learning to mimic a human expert/teacher from training trajectories is called imitation learning. To make the process of teaching easier in this setting, we propose to employ transfer learning (where one learns on a source problem and transfers the knowledge to potentially more complex target problems). We consider multi-relational environments such as real-time strategy games and use functional-gradient boosting to capture and transfer the models learned in these environments. Our experiments demonstrate that our learner learns a very good initial model from the simple scenario and effectively transfers the knowledge to the more complex scenario thus achieving a jump start, a steeper learning curve and a higher convergence in performance.

In the previous chapters, we discussed the structure learning algorithms for two SRL models and extended them to learn with missing data. In this chapter, we discuss how this algorithm can be adapted to learn to act in sequential domains. We then present three of our most successful applications in real health care data—two cardiovascular prediction problems and the third is prediction of onset of Alzheimer’s disease. We then conclude the chapter with a few NLP applications.

In this chapter, we show the use of functional gradient boosting for learning Relational Dependency Networks (RDNs). The use of several regression trees, instead of just one, results in an expressive model for the conditional distributions of RDNs. We then present a sample set of results that show superior performance when compared to state-of-the-art approaches.

The 2013 European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECMLPKDD) was held in Prague, September 2013. Following the long-standing tradition of the series, it brought together researchers in Machine Learning and Data Mining for an exciting program of cutting-edge research results, aiming in particular at a cross-fertilization between these two areas. The 2013 edition featured a recordbreaking 138 full presentations of novel research results, selected through a careful peer review process from 629 submissions. For the first time, the conference used a mixed submission model. Work could be submitted as a journal article to one of two journals participating in this model (Machine Learning and Data Mining and Knowledge Discovery), or it could be submitted for publication in the conference proceedings. A total of 86 original manuscripts were submitted to Machine Learning. Eight of these were accepted for inclusion in this issue, and for presentation at the conference. We briefly summarize them in alphabetical order of authors.

Magnetic resonance imaging (MRI) has emerged as an important tool to identify intermediate biomarkers of Alzheimer's disease (AD) due to its ability to measure regional changes in the brain that are thought to reflect disease severity and progression. In this paper, we set out a novel pipeline that uses volumetric MRI data collected from different subjects as input and classifies them into one of three classes: AD, mild cognitive impairment (MCI) and cognitively normal (CN). Our pipeline consists of three stages -- (1) a segmentation layer where brain MRI data is divided into clinically relevant regions, (2) a classification layer that uses relational learning algorithms to make pair wise predictions between the three classes, and (3)a combination layer that combines the results of the different classes to obtain the final classification. One of the key features of our proposed approach is that it allows for domain expert's knowledge to guide the learning in all the layers. We evaluate our pipeline on 397 patients acquired from the Alzheimer's Disease Neuroimaging Initiative and demonstrate that it obtains state-of the-art performance with minimal feature engineering.

Analyzing telemetry data of player behavior in computer games is a topic of increasing interest for industry and research, alike. When applied to game telemetry data, pattern recognition and statistical analysis provide valuable business intelligence tools for game development. An important problem in this area is to characterize how player engagement in a game evolves over time. Reliable models are of pivotal interest since they allow for assessing the long-term success of game products and can provide estimates of how long players may be expected to keep actively playing a game. In this paper, we introduce methods from random process theory into game data mining in order to draw inferences about player engagement. Given large samples (over 250,000 players) of behavioral telemetry data from five different action-adventure and shooter games, we extract information as to how long individual players have played these games and apply techniques from lifetime analysis to identify common patterns. In all five cases, we find that the Weibull distribution gives a good account of the statistics of total playing times. This implies that an average player’s interest in playing one of the games considered evolves according to a non-homogeneous Poisson process. Therefore, given data on the initial playtime behavior of the players of a game, it becomes possible to predict when they stop playing.

Multi-task and relational learning with Gaussian processes are two active but also orthogonal areas of research. So far, there has been few attempt at exploring relational information within multi-task Gaussian processes. While existing relational Gaussian process methods have focused on relations among entities and in turn could be employed within an individual task, we develop a class of Gaussian process models which incorporates relational information across multiple tasks. As we will show, inference and learning within the resulting class of models, called relational multi-task Gaussian processes, can be realized via a variational EM algorithm. Experimental results on synthetic and real-world datasets verify the usefulness of this approach: The observed relational knowledge at the level of tasks can indeed reveal additional pair wise correlations between tasks of interest and, in turn, improve prediction performance.

Recent years have seen a surge of interest in Statistical Relational Learning (SRL) models that combine logic with probabilities. One prominent example is Markov Logic Networks (MLNs). While MLNs are indeed highly expressive, this expressiveness comes at a cost. Learning MLNs is a hard problem and therefore has attracted much interest in the SRL community. Current methods for learning MLNs follow a two-step approach: first, perform a search through the space of possible clauses and then learn appropriate weights for these clauses. We propose to take a different approach, namely to learn both the weights and the structure of the MLN simultaneously. Our approach is based on functional gradient boosting where the problem of learning MLNs is turned into a series of relational functional approximation problems. We use two kinds of representations for the gradients: clause-based and tree-based. Our experimental evaluation on several benchmark data sets demonstrates that our new approach can learn MLNs as good or better than those found with state-of-the-art methods, but often in a fraction of the time.

There have recently been considerable advances in fast inference for (online) latent Dirichlet allocation (LDA). While it is widely recognized that the scheduling of documents in stochastic optimization and in turn in LDA may have significant consequences, this issue remains largely unexplored. Instead, practitioners schedule documents essentially uniformly at random, due perhaps to ease of implementation, and to the lack of clear guidelines on scheduling the documents.In this work, we address this issue and propose to schedule documents for an update that exert a disproportionately large influence on the topics of the corpus before less influential ones. More precisely, we justify to sample documents randomly biased towards those ones with higher norms to form mini-batches. On several real-world datasets, including 3M articles from Wikipedia and 8M from PubMed, we demonstrate that the resulting influence scheduled LDA can handily analyze massive document collections and find topic models as good or better than those found with online LDA, often at a fraction of time.

Recently, there have been considerable advances in fast inference for latent Dirichlet allocation (LDA). In particular, stochastic optimization of the variational Bayes (VB) objective function with a natural gradient step was proved to converge and able to process massive document collections. To reduce noise in the gradient estimation, it considers multiple documents chosen uniformly at random. While it is widely recognized that the scheduling of documents in stochastic optimization may have significant consequences, this issue remains largely unexplored. In this work, we address this issue. Specifically, we propose residual LDA, a novel, easy-to-implement, LDA approach that schedules documents in an informed way. Intuitively, in each iteration, residual LDA actively selects documents that exert a disproportionately large influence on the current residual to compute the next update. On several real-world datasets, including 3M articles from Wikipedia, we demonstrate that residual LDA can handily analyze massive document collections and find topic models as good or better than those found with batch VB and randomly scheduled VB, and significantly faster.

Learning the reward function of an agent by observing its behavior is termed inverse reinforcement learning and has applications in learning from demonstration or apprenticeship learning. We introduce the problem of multi-agent inverse reinforcement learning, where reward functions of multiple agents are learned by observing their uncoordinated behavior. A centralized controller then learns to coordinate their behavior by optimizing a weighted sum of reward functions of all the agents. We evaluate our approach on a traffic-routing domain, in which a controller coordinates actions of multiple traffic signals to regulate traffic density. We show that the learner is not only able to match but even significantly outperform the expert.

In preference learning, the algorithm observes pairwise relative judgments (preference) between items as training data for learning an ordering of all items. This is an important learning problem for applications where absolute feedback is difficult to elicit, but pairwise judgments are readily available (e.g., via implicit feedback [13]). While it was already shown that active learning can effectively reduce the number of training pairs needed, the most successful existing algorithms cannot generalize over items or queries. Considering web search as an example, they would need to learn a separate relevance score for each document-query pair from scratch. To overcome this inefficiency, we propose a link-based active preference learning method based on Gaussian Processes (GPs) that incorporates dependency information from both feature-vector representations as well as relations. Specifically, to meet the requirement on computational efficiency of active exploration, we introduce a novel incremental update method that scales as well as the nongeneralizing models. The proposed algorithm is evaluated on datasets for information retrieval, showing that it learns substantially faster than algorithms that cannot model dependencies.

Latent Dirichlet allocation is a fully generative statistical language model that has been proven to be successful in capturing both the content and the topics of a corpus of documents. Recently, it was even shown that relations among documents such as hyper-links or citations allow one to share information between documents and in turn to improve topic generation. Although fully generative, in many situations we are actually not interested in predicting relations among documents. In this paper, we therefore present a Dirichlet-multinomial nonparametric regression topic model that includes a Gaussian process prior on joint document and topic distributions that is a function of document relations. On networks of scientific abstracts and of Wikipedia documents we show that this approach meets or exceeds the performance of several baseline topic models.

Statistical relational learning (SRL) provides effective techniques to analyze social network data with rich collections of objects and complex networks. Infinite hidden relational models (IHRMs) introduce nonparametric mixture models into relational learning and have been successful in many relational applications. In this paper we explore the modeling and analysis of complex social networks with IHRMs for community detection, link prediction and product recommendation. In an IHRM-based social network model, each edge is associated with a random variable and the probabilistic dependencies between these random variables are specified by the model, based on the relational structure. The hidden variables, one for each object, are able to transport information such that non-local probabilistic dependencies can be obtained. The model can be used to predict entity attributes, to predict relationships between entities and it performs an interpretable cluster analysis. We demonstrate the performance of IHRMs with three social network applications. We perform community analysis on the Sampson's monastery data and perform link analysis on the Bernard & Killworth data. Finally we apply IHRMs to the MovieLens data for prediction of user preference on movies and for an analysis of user clusters and movie clusters.

Triggered by a market relevant application that involves making joint predictions of pedestrian and public transit flows in urban areas, we address the question of how to utilize hidden common cause relations among variables of interest in order to improve performance in the two related regression tasks. Specifically, we propose stacked Gaussian process learning, a meta-learning scheme in which a base Gaussian process is enhanced by adding the posterior covariance functions of other related tasks to its covariance function in a stage-wise optimization. The idea is that the stacked posterior covariances encode the hidden common causes among variables of interest that are shared across the related regression tasks. Stacked Gaussian process learning is efficient, capable of capturing shared common causes, and can be implemented with any kind of standard Gaussian process regression model such as sparse approximations and relational variants. Our experimental results on real-world data from the market relevant application show that stacked Gaussian processes learning can significantly improve prediction performance of a standard Gaussian process.

Gaussian processes have successfully been used to learn preferences among entities as they provide nonparametric Bayesian approaches for model selection and probabilistic inference. For many entities encountered in real-world applications, however, there are complex relations between them. In this paper, we present a preference model which incorporates information on relations among entities. Specifically, we propose a probabilistic relational kernel model for preference learning based on Silva et al.'s mixed graph Gaussian processes: a new prior distribution, enhanced with relational graph kernels, is proposed to capture the correlations between preferences. Empirical analysis on the LETOR datasets demonstrates that relational information can improve the performance of preference learning.

The task of aligning sequences arises in many applications. Classical dynamic programming approaches require the explicit state enumeration in the reward model. This is often impractical: the number of states grows very quickly with the number of domain objects and relations among these objects. Relational sequence alignment aims at exploiting symbolic structure to avoid the full enumeration. This comes at the expense of a more complex reward model selection problem: virtually infinitely many abstraction levels have to be explored. In this paper, we apply gradient-based boosting to leverage this problem. Specifically, we show how to reduce the learning problem to a series of relational regressions problems. The main benefit of this is that interactions between states variables are introduced only as needed, so that the potentially infinite search space is not explicitly considered. As our experimental results show, this boosting approach can significantly improve upon established results in challenging applications.

Legged robots require accurate models of their environment in order to plan and execute paths. We present a probabilistic technique based on Gaussian processes that allows terrain models to be learned and updated efficiently using sparse approximation techniques. The major benefit of our terrain model is its ability to predict elevations at unseen locations more reliably than alternative approaches, while it also yields estimates of the uncertainty in the prediction. In particular, our nonstationary Gaussian process model adapts its covariance to the situation at hand, allowing more accurate inference of terrain height at points that have not been observed directly. We show how a conventional motion planner can use the learned terrain model to plan a path to a goal location, using a terrain-specific cost model to accept or reject candidate footholds. In experiments with a real quadruped robot equipped with a laser range finder, we demonstrate the usefulness of our approach and discuss its benefits compared to simpler terrain models such as elevations grids.

Hidden Markov Models (HMM) have been successfully used in applications such as speech recognition, activity recognition, bioinformatics etc. There have been previous attempts such as Hierarchical HMMs and Abstract HMMs to elegantly extend HMMs at multiple levels of temporal abstraction (for example to represent the user's activities). Similarly, there has been previous work such as Logical HMMs on extending HMMs to domains with relational structure. In this work we develop a representation that naturally combines the power of both relational and hierarchical models in the form of Logical Hierarchical Hidden Markov Models (LoHiHMMs). LoHiHMMs inherit the compactness of representation from Logical HMMs and the tractability of inference from Hierarchical HMMs. We outline two inference algorithms: one based on grounding the LoHiHMM to a propositional HMM and the other based on particle filtering adapted for this setting. We present the results of our experiments with the model in two simulated domains.

In probabilistic mobile robotics, the development of measurement models plays a crucial role as it directly influences the efficiency and the robustness of the robot’s performance in a great variety of tasks including localization, tracking, and map building. In this paper, we present a novel probabilistic measurement model for range finders, called Gaussian beam processes, which treats the measurement modeling task as a nonparametric Bayesian regression problem and solves it using Gaussian processes. The major benefit of our approach is its ability to generalize over entire range scans directly. This way, we can learn the distributions of range measurements for whole regions of the robot’s configuration space from only few recorded or simulated range scans. Especially in approximative approaches to state estimation like particle filtering or histogram filtering, this leads to a better approximation of the true likelihood function. Experiments on real world and synthetic data show that Gaussian beam processes combine the advantages of two popular measurement models.

Navigation is one of the fundamental tasks for a mobile robot. The majority of path planning approaches has been designed to entirely solve the given problem from scratch given the current and goal configurations of the robot Although these approaches yield highly efficient plans, the computed policies typically do not transfer to other, similar tasks. We propose to learn relational decision trees as abstract navigation strategies from example paths. Relational abstraction has several interesting and important properties. First, it allows a mobile robot to generalize navigation plans from specific examples provided by users or exploration. Second, the navigation policy learned in one environment can be transferred to unknown environments. In several experiments with real robots in a real environment and in simulated runs, we demonstrate the usefulness of our approach. © 2006 IEEE. ispartof: pages:2792-2797 ispartof: 2006 IEEE/RSJ International Conference on Intelligent Robots and Systems pages:2792-2797 ispartof: International Conference on Intelligent Robots and Systems (IROS) location:Beijing, China date:9 Oct - 15 Oct 2006 status: published

Combining statistical and relational learning receives currently a lot of attention. The majority of statistical relational learning approaches focus on density estimation. For classification, however, it is well-known that the performance of such generative models is often lower than that of discriminative classifiers. One approach to improve the performance of generative models is to combine them with discriminative algorithms. Fisher kernels were developed to combine them with kernel methods, and have shown promising results for the combinations of support vector machines with (logical) hidden Markov models and Bayesian networks. So far, however, Fisher kernels have not been considered for relational data, i.e., data consisting of a collection of objects and relational among these objects. In this paper, we develop Fisher kernels for relational data and empirically show that they can significantly improve over the results achieved without Fisher kernels.