Search

Fuzzy Cognitive Maps (FCMs) are considered a soft computing technique combining elements of fuzzy logic and recurrent neural networks. They found multiple application in such domains as modeling of system behavior, prediction of time series, decision making and process control. Less attention, however, has been turned towards using them in pattern classification. In this work we propose an FCM based classifier with a fully connected map structure. In contrast to methods that expect reaching a steady system state during reasoning, we chose to execute a few FCM iterations (steps) before collecting output labels. Weights were learned with a gradient algorithm and logloss or cross-entropy were used as the cost function. Our primary goal was to verify, whether such design would result in a descent general purpose classifier, with performance comparable to off the shelf classical methods. As the preliminary results were promising, we investigated the hypothesis that the performance of d -step classifier can be attributed to a fact that in previous d − 1 steps it transforms the feature space by grouping observations belonging to a given class, so that they became more compact and separable. To verify this hypothesis we calculated three clustering scores for the transformed feature space. We also evaluated performance of pipelines built from FCM-based data transformer followed by a classification algorithm. The standard statistical analyzes confirmed both the performance of FCM based classifier and its capability to improve data. The supporting prototype software was implemented in Python using TensorFlow library.

Various country-wide guidelines for speed limits setting make use of diverse parameters specifying properties of roads, their environment, as well as traffic intensity, statistical distribution of measured vehicle speeds and history of crash accidents. We argue, that although such extensive data are not present in public geospatial datasets, they can be inferred as latent features in machine learning models trained to predict speed limits. To verify this hypothesis we performed an experiment, in which we extracted tag-based and geometrical features from Open Street Map for Poland and applied various classification methods to predict class labels corresponding to speed limits. In spite of the fact that the datasets were imbalanced (with majority classes corresponding to default speed limits) we obtained F1 scores ranging from 0.72 to 0.91 for lower speed roads. The neural network classifier implemented with Tensorflow framework turned out to be the most efficient.

Authorship attribution aims at identifying the author of an unseen text document based on text samples originating from different authors. In this paper we focus on authorship attribution of Polish texts using stylometric features based on part of speech (POS) tags. Polish language is characterized by high inflection level and in consequence over 1000 POS tags can be distinguished. This allows building a sufficiently large feature space by extracting POS information from documents and performing their classification with use of machine learning methods. We report results of experiments conducted with Weka workbench using combinations of the following features: POS tags, an approximation of their bigrams and simple document statistics.

Authorship attribution aims at distinguishing texts written by different authors using text features representing their styles. In this paper we investigate stylometric features for the Polish language based on Part of Speech (POS) tagging (including POS bigrams) and function words. Due to high inflection level of Polish language the feature space tends to be very large. This in particular concerns POS n-grams. Focusing on POS bigrams, we propose their simplified representation allowing to keep the feature space compact. We report experiments, in which authorship attribution was conducted for varying in lengths documents, with use of classifiers from the Weka library. We evaluate classification results for combinations of the following features: POS tags, POS bigrams, function words and simple document statistics. Experiments indicate that the developed features provide good classification performance.

This paper presents a system for extraction of concepts from unstructured Polish texts. Here concepts are understood as n-grams, whose words satisfy specific grammatical constraints. Detection and transformation of concepts to their normalized form are performed with rules defined in a language, which combines elements of colored and fuzzy Petri nets. We apply a user friendly method for specification of samples of transformation patterns that are further compiled to rules. To improve accuracy and performance, we recently introduced rule management mechanisms, which are based on two relations between rules: partial refinement and covering. The implemented methods include filtering with metarules and removal of redundant rules (i.e. these covered by other rules). We report results of experiments, which aimed at extracting specific concepts (actions) using a ruleset refactored with the developed rule management techniques.

This paper addresses the problem of modeling and automated recognition of video events. We propose to use Linear Temporal Logic as a language for events specification and Fuzzy Semantic Petri Nets (FSPN) as a tool for their recognition. FSPN are Petri nets coupled with an underlying fuzzy ontology. The ontology stores assertions (facts) concerning classification of objects and detected relations. Fuzzy predicates querying the ontology content are used as guards of transitions in FSPN. Tokens carry information on objects participating in a scenario and are equipped with weights indicating likelihood of their assignment to places. In turn, the places correspond to scenario steps. We describe a prototype detection system consisting of an FSPN interpreter, the fuzzy ontology, and a set of predicate evaluators. Initial tests yielding promising results are reported.

We present recently developed solution allowing extraction of concepts from unstructured Polish texts with special focus on correct morphological forms of obtained concept names. As Polish is a highly inflected language, detected names need to be transformed following Polish grammar rules. We propose a user-friendly method for specification of transformation patterns, which is based on a simple annotations language. Annotations prepared by a user are compiled into transformation rules. During the concept extraction process the input document is split into sentences and the rules are applied to sequences of words comprised in sentences. Recognized strings forming concept names are aggregated at various levels and assigned with scores. We report also results of initial experiments performed on a medical text.

The motivation for the presented work was the need for solution allowing efficient estimation of traffic parameters in based on sparse measurements covering a small fraction road segments within large urban road network. The solution was intended to be deployed within the Dynamic Map, an important component of the INSIGMA system used for route planning and traffic control. We have chosen to base the estimation model on Fuzzy Cognitive Maps (FCM) formalism and apply an Evolutionary Algorithm to learn the model parameters. The main argument in favor of FCM is their simplicity and the speed of calculations that can be required for real-time updates of traffic parameters. This paper discusses a novel evolutionary algorithm for FCM learning and demonstrates feasibility and efficiency of such approach by giving results of tests performed on road networks of various size.

Automated recognition of complex video events poses challenges related to: selection of formalisms for efficient event modeling and analysis, mapping semantic high-level concepts used in specifications on information extracted from video sequences, as well as managing uncertainty associated with this information. We propose Fuzzy Semantic Petri Nets (FSPN) as a tool aimed at solving the mentioned problems. FSPN are Petri nets coupled with an underlying fuzzy ontology. The ontology stores assertions (facts) concerning object classification and detected relations. Fuzzy predicates querying the ontology are used as transition guards. Places in FSPN represent scenario steps. Tokens carry information on objects participating in a scenario and have weights expressing likelihood of a step occurrence. FSPN enable detection of events occurring concurrently, analysis of various combinations of objects and reasoning about alternatives.

In this paper we address two problems. The first pertains to implementation of medical guidelines in an e-health system supporting self-management of chronic diseases. The system allows patients to enter observed symptoms and measured parameters, then makes assessment of disease state and informs about necessary actions. We propose to formalize guidelines as sets of fuzzy rules. Fuzziness is introduced to cope with uncertainty resulting from self-observations bias, low quality of sensors and limited patients skills. The second problem is more general. It concerns the reuse of knowledge gathered in ontologies and an application of Semantic Web technologies to perform fuzzy inference. We show that, despite the fact that commonly used ontology languages and supporting tools are not intended to handle vagueness and uncertainty, they can be successfully integrated to represent and execute a set of fuzzy rules. The proposed method consists in refactoring a domain ontology, then introducing additional relations expressing fuzzy properties, encoding Mamdani fuzzy rules in SWRL language and executing them with use of Pellet OWL reasoner. We describe a fuzzy reasoning engine applying this approach and discuss translation of fuzzy rules to SWRL constructs taking as example a complete set of rules formalizing a medical guideline for asthma control assessment.

For many classes of problems it is difficult to make decisions, assessments or develop plans based on precise quantitative models. Fuzzy Cognitive Maps (FCM) are a well-known tool for qualitative analysis of systems that uses a simple representation of knowledge in the form of a graph of concepts linked by causal relationships. Advantages of FCMs are the ease of gathering and representing knowledge and the simplicity of reasoning techniques, very close to neural networks. The paper gives results of experiments aiming at application of FCMs to analysis of development scenarios for research and teaching units. The analyzes were conducted for four representative classes of academic units: strong, medium, weak with a development potential and weak; their results are discussed and several observations related to reasoning with FCMs are made.