Pinar Öztürk

A context model for knowledge-intensive case-based reasoning

Decision-support systems that help solving problems in open and weak theory domains, i.e. hard problems, need improved methods to ground their models in real-world situations. Models that attempt to capture domain knowledge in terms of, e.g. rules or deeper relational networks, tend either to become too abstract to be efficient or too brittle to handle new problems. In our research, we study how the incorporation of case-specific, episodic, knowledge enables such systems to become more robust and to adapt to a changing environment by continuously retaining new problem-solving cases as they occur during normal system operation. The research reported in this paper describes an extension that incorporates additional knowledge of the problem-solving context into the architecture. The components of this context model is described, and related to the roles the components play in an abductive diagnostic process. Background studies are summarized, the context model is explained and an example shows its integration into an existing knowledge-intensive CBR system.

Analogical Trust Reasoning

Trust is situation-specific and the trust judgment problem with which the truster is confronted might be, in some ways, similar but not identical to some problems the truster has previously encountered. The truster then may draw information from these past experiences useful for the current situation. We present a knowledge-intensive and model-based case-based reasoning framework that supports the truster to infer such information. The suggested method augments the typically sparse trust information by inferring the missing information from other situational conditions, and can better support situation-aware trust management. Our framework can be coupled with existing trust management models to make them situation-aware. It uses the underlying model of trust management to transfer trust information between situations. We validate the proposed framework for Subjective Logic trust management model and evaluate it by conducting experiments on a large real dataset.

Short-Term Load Forecasting With Seasonal Decomposition Using Evolution for Parameter Tuning

This paper studies data-driven short-term load forecasting, where historic data are used to predict the expected load for the next 24 h. Our focus is to simplify and automate the estimation and analysis of various forecasting models. We propose a three-stage approach to load forecasting, consisting of preprocessing, forecasting, and postprocessing, where the forecasting stage uses evolution to automatically set the parameters for each model. In our implementation, the preprocessing stage includes removal of daily and weekly seasonality by a nonparametric method. This seasonal pattern is added in the postprocessing stage. The system allows for easy exploration of several forecasting models, without the need to have in-depth knowledge of how to obtain the best performance for each model. We apply the method to several forecasting algorithms and on three datasets: (1) distribution substation; (2) GEFCom 2012; and (3) a transmission level dataset. We find that the forecasting algorithms considered produce significantly more accurate forecasts when combined with our proposed preprocessing stage compared with applying the same algorithms directly on the raw data. We also find that the parameter values chosen by evolution often provide insights into the interplay between the different datasets and forecast models. Software is available online.

Self-organizing multiple models for imitation: teaching a robot to dance the YMCA

The traditional approach to implement motor behaviour in a robot required a programmer to carefully decide the joint velocities at each timestep. By using the principle of learning by imitation, the robot can instead be taught simply by showing it what to do. This paper investigates the self-organization of a connectionist modular architecture for motor learning and control that is used to imitate human dancing. We have observed that the internal representation of a motion behaviour tends to be captured by more than one module. This supports the hypothesis that a modular architecture for motor learning is capable of self-organizing the decomposition of a movement.

Acquisition and Reuse of Reasoning Knowledge from Textual Cases for Automated Analysis

Analysis is essential for solving complex problems such as diagnosing a patient, investigating an accident or predicting the outcome of a legal case. It is a non-trivial process even for human experts. To assist experts in this process we propose a CBR-based approach for automated problem analysis. In this approach a new problem is analysed by reusing reasoning knowledge from the analysis of a similar problem. To avoid the laborious process of manual case acquisition, the reasoning knowledge is extracted automatically from text and captured in a graph-based representation, which we dubbed Text Reasoning Graph (TRG), that consists of causal, entailment and paraphrase relations. The reuse procedure involves adaptation of a similar past analysis to a new problem by finding paths in TRG that connect the evidence in the new problem to conclusions of the past analysis. The objective is to generate the best explanation of how the new evidence connects to the conclusion. For evaluation, we built a system for analysing aircraft accidents based on the collection of aviation investigation reports. The evaluation results show that our reuse method increases the precision of the retrieved conclusions.

Evidence-Driven Retrieval in Textual CBR: Bridging the Gap Between Retrieval and Reuse

The most similar case may not always be the most appropriate one to guide a problem-solving process. It is often important that a retrieved past case can be easily adapted to a target problem. The presented work deals with the retrieval and adaptation in textual case-based reasoning (TCBR) where cases are described textually. In TCBR, it is common to use similarity-based retrieval methods from information retrieval where adaptability of the retrieved cases is not considered. In this paper we introduce a novel case retrieval method called evidence-driven retrieval (EDR). It uses the notion of evidence to determine which parts of the new problem text have been useful in the past solutions and will be used in the adaptation to a new problem. This allows EDR to retrieve cases that are not only similar but also adaptable. We evaluated EDR as part of our TCBR approach that aims to support human experts in root cause analysis of transportation incidents. This approach relies on causal knowledge automatically extracted from incident reports from the Transportation Safety Board of Canada, which are used as textual cases in our experiments. The results for EDR are compared with information retrieval methods traditionally applied in TCBR.

Levels and Types of Action Selection: The Action Selection Soup

Action selection (AS) is defined as the process where an action is selected among a number of alternatives. This definition, however, does not sufficiently describe what an action is. What is the unit of selection in the first place? We maintain that the artificial intelligence (AI) accounts of AS typically mix and merge two AS situations that indeed are qualitatively different. Most of the accounts actually deal only with one type of AS but purport to cover both types of AS. We propose three dimensions along which the commonalities and the differences between various AS accounts can be analyzed, and use these for a preliminary conceptualization of what we call a two-system action selection account. In particular, we identify two qualitatively different AS situations whose architectures, we suggest, can be designed inspired by neuroscience models of the basal ganglia (BG) and the cerebellum, respectively.

Extraction and generalisation of variables from scientific publications

Scientific theories and models in Earth science typically involve changing variables and their complex interactions, including correlations, causal relations and chains of positive/negative feedback loops. Variables tend to be complex rather than atomic entities and expressed as noun phrases containing multiple modifiers, e.g. oxygen depletion in the upper 500 m of the ocean or timing and magnitude of surface temperature evolution in the Southern Hemisphere in deglacial proxy records. Text mining from Earth science literature is therefore significantly different from biomedical text mining and requires different approaches and methods. Our approach aims at automatically locating and extracting variables and their direction of variation: increasing, decreasing or just changing. Variables are initially extracted by matching tree patterns onto the syntax trees of the source texts. Next, variables are generalised in order to enhance their similarity, facilitating hierarchical search and inference. This generalisation is accomplished by progressive pruning of syntax trees using a set of tree transformation operations. Text mining results are presented as a browsable variable hierarchy which allows users to inspect all mentions of a particular variable type in the text as well as any generalisations or specialisations. The approach is demonstrated on a corpus of 10k abstracts of Nature publications in the field of Marine science. We discuss experiences with this early prototype and outline a number of possible improvements and directions for future re

Finding potential seeds through rank aggregation of web searches

This paper presents a potential seed selection algorithm for web crawlers using a gain - share scoring approach. Initially we consider a set of arbitrarily chosen tourism queries. Each query is given to the selected N commercial Search Engines (SEs); top msearch results for each SE are obtained, and each of these mresults is manually evaluated and assigned a relevance score. For each of m results, a gain - share score is computed using their hyperlinks structure across N ranked lists. Gain score of each link present in each of m results and a portion of the gain score is propagated to the share score of each of m results. This updated share scores of each of m results determine the potential set of seed URLs for web crawling. Experimental results on tourism related web data illustrate the effectiveness of the proposed seed selection algorithm.

A review of case-based reasoning in cognition–action continuum: a step toward bridging symbolic and non-symbolic artificial intelligence

In theories and models of computational intelligence, cognition and action have historically been investigated on separate grounds. We conjecture that the main mechanism of case-based reasoning (CBR) applies to cognitive tasks at various levels and of various granularity, and hence can represent a bridge - or a continuum - between the higher and lower levels of cognition. CBR is an artificial intelligence method that draws upon the idea of solving a new problem reusing similar past experiences. In this paper we re-formulate the notion of CBR to highlight the commonalities between higher level cognitive tasks such as diagnosis, and lower level control such as voluntary movements of an arm. In this view, CBR is envisaged as a generic process independent from the content and the detailed format of cases. Diagnostic cases and internal representations underlying motor control constitute two instantiations of the case representation. In order to claim such a generic mechanism, the account of CBR needs to be revised so that its position in non-symbolic AI becomes clearer. The paper reviews the CBR literature that targets lower levels of cognition to show how CBR may be considered as a step toward bridging the gap between symbolic and nonsymbolic AI.