Neuro-symbolic integration of symbolic and subsymbolic techniques represents a fast-growing AI trend aimed at mitigating the issues of neural networks in terms of decision processes, reasoning, and interpretability. Several state-of-the-art neuro-symbolic approaches aim at improving performance, most of them focusing on proving their effectiveness in terms of raw predictive performance and/or reasoning capabilities. Meanwhile, few efforts have been devoted to increasing model trustworthiness, interpretability, and efficiency - mostly due to the complexity of measuring effectively improvements in terms of trustworthiness and interpretability. This is why here we analyse and discuss the need for ad-hoc trustworthiness metrics for neuro-symbolic techniques. We focus on two popular paradigms mixing subsymbolic computation and symbolic knowledge, namely: (i) symbolic knowledge extraction (SKE), aimed at mapping subsymbolic models into human-interpretable knowledge bases; and (ii) symbolic knowledge injection (SKI), aimed at forcing subsymbolic models to adhere to a given symbolic knowledge. We first emphasise the need for assessing neuro-symbolic approaches from a trustworthiness perspective, highlighting the research challenges linked with this evaluation and the need for ad-hoc trust definitions. Then we summarise recent developments in SKE and SKI metrics focusing specifically on several trustworthiness pillars such as interpretability, efficiency, and robustness of neuro-symbolic methods. Finally, we highlight open research opportunities towards reliable and flexible trustworthiness metrics for neuro-symbolic integration.
2
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Real-time strategy games are currently very popular as a testbed for AI research and education. StarCraft: Brood War (SC:BW) is one of such games. Recently, a new large, unlabeled human versus human SC:BW game replay dataset called STARDATA was published. This paper aims to prove that the player strategy diversity requirement of the dataset is met, i.e., that the diversity of player strategies in STARDATA replays is of sufficient quality. To this end, we built a competitive SC:BW agent from scratch and trained its strategic decision making process on STARDATA. The results show that in the current state of the competitive environment the agent is capable of keeping a stable rating and a decent win rate over a longer period of time. It also performs better than our other, simple rule-based agent. Therefore, we conclude that the strategy diversity requirement of STARDATA is met.
3
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Highly structured knowledge bases such as lexical semantic networks contain various connectivity patterns that can be learned as node features using dedicated frameworks. However, semantic relations are often unequally distributed over such knowledge resources. Some of the language partitions may benefit from integrating structured resources which are more easily available for resource-rich languages. In the present paper, we propose a simple endogenous method for enhancing a multilingual knowledge base through the cross-lingual semantic relation inference. It can be run on multilingual resources prior to semantic representation learning. Multilingual knowledge bases may integrate preexisting structured resources available for resource-rich languages. We aim at performing cross-lingual inference on them to improve the low resource language by creating semantic relationships.
4
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Context is widely considered for NLP and knowledge discovery since it highly influences the exact meaning of natural language. The scientific challenge is not only to extract such context data, but also to store this data for further NLP approaches. Here, we propose a multiple step knowledge graph based approach to utilize context data for NLP and knowledge expression and extraction. We introduce the graph-theoretic foundation for a general context concept within semantic networks and show a proof-of-concept based on biomedical literature and text mining. We discuss the impact of this novel approach on text analysis, various forms of text recognition and knowledge extraction and retrieval.
The paper presents problems of technical product preparation tasks aimed at product selection and redesign for particular application supported by Al methods. The identification of customer needs was discussed, as well as the issues of product decomposition were presented. The QFD method was suggested to be applied as the product and process data integration tool, where the engineering characteristic of a product was combined with its trade characteristic. In the industrial product characteristic, the delivery time, which can be fixed with the use of graph based scheduling method, is particularly important. The paper shows that thanks to the application of KBS methods in production preparation it is possible to support decision making connected with product selection for particular use.
PL
W pracy przedstawiono analizę zadań technicznego przygotowania produkcji (TPP) ukierunkowanych na dobór i adaptację wyrobu do danego zastosowania wspomaganą metodami Al. Analizie poddano zagadnienia identyfikacji potrzeb klienta. Omówiono zagadnienia dekompozycji wyrobu. Zastosowano metodę QFD jako narzędzia integracji danych z zakresu TPP. Charakterystykę techniczną wyrobu powiązano z charakterystyką handlową obejmującą m.in. termin realizacji. Wykazano, że zastosowanie metod KBS w przygotowaniu produkcji umożliwia wspomaganie decyzji w zakresie doboru wyrobu do danego zastosowania.
6
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
The Reconfigurable Manufacturing System (RMS) offers a flexible, changeable and dynamic manufacturing platform which is complex. The RMS evolved from the necessity to satisfy the markets highly customised demands. As the requirements of the customers vary the manufacturing system will need continuous and timely reconfigurations. A Knowledge Based System (KBS) is therefore the critical link in the module selection from a database to configure a machine. Modules will be selected to be timeously assembled into machine tools to manufacture customised goods with minimal disruptions to satisfy customer delivery times. This paper will review the issues surrounding KBS software and their characteristics, as the appropriateness and requisite for establishing an RMS knowledge based system application is explored. Basic machine tree structures for the KBS application are also presented and discussed.
PL
Rekonfigurowalny System Wytwarzania (RMS) stanowi elastyczną, zmienną i dynamiczną platformę produkcyjną, będącą złożonym układem. Systemy RMS z potrzeby zaspokojenia wymagań rynków, dostosowanych do indywidualnych żądań klientów. Ponieważ potrzeby klientów się zmieniają, systemy wytwarzania będą wymagały nieustannej rekonfiguracji wykonywanej z zachowaniem określonych ograniczeń czasowych. System Baz Wiedzy (KBS) jest więc elementem krytycznym w sekcji modułów, łączącym bazę danych z procesem konfiguracji maszyny. Wybór modułów będzie dokonywany z uwzględnieniem czasu potrzebnego do zamontowania ich w narzędziach do obróbki skrawaniem i innych maszynach. W niniejszym artykule dokonano przeglądu zagadnień związanych ze środowiskiem programowym Systemów Baz Wiedzy (KBS) wraz z ich charakterystykami, a także ich przydatności i wymagań pozwalających na opracowanie Rekonfigurowalnego Systemu Wytwarzania (RMS) opartego na bazie wiedzy. Przedstawiono i omówiono także podstawowe struktury drzewa decyzyjnego dla wyboru maszyn i wyposażenia dla aplikacji systemów opartych na bazach wiedzy (KBS).
7
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
The task oriented reasoning represents a powerful design paradigm. This research starts from a representative research in the area, the Unified Problem-Solving Method Development Language (UPML), and proposes a multi-agent architecture to support this ap- proach. The paper first models an applicative program, solving non- linear equation systems, using UPML concepts and then instantiates the proposed generic multi-agent architecture in order to generate a multi-agent system to solve the proposed problem.
The paper concerns a class of knowledge-based control systems containing a controller with inputs and outputs assumed to be values of uncertain variables. The description of the controller then has the form of certainty distribution of these variables. Two general approaches to the knowledge-based design of such a controller are presented: in a prescriptive approach expert's knowledge concerns a controller and in a descriptive approach expert's knowledge concerns a plant. A method for evaluating control quality obtained with both approaches is proposed, and a concept of controller's adaptation is presented. Numerical example and results of simulations are included.
The symbol grounding problem is discussed for the case of simple language of formulas with modal operators and the cognitive agent. The language formulas are built from modal operators and logical connectives of conjunction, disjunction and exclusive disjunction. The cognitive agent carries out perceptions of an external world and stores their content in dedicated temporal database. The empirical experience contained in this database defines the scope of possible meaning that can ever be assigned to belief formulas. Two close but different approaches to implementing the idea of grounding in cognitive structures of the agent are presented.
In this paper, we present an expert network scheme designed to obtain discrete transfer functions for LTI systems under real sampling of finite duration rather than an instantaneous ideal one. For this purpose, the expert network handles two different identification methods to derive parametric discrete models techniques of reduced mathematical complexity from measured input-output data series. One of the methods is based on a typically used least-squares minimization, while the other one is based on the Leverrier's algorithm; that is, using a data series of the impulse response of the system to identify a parametric discrete model. These techniques are of particular practical interest when the continuous-time system is unknown or when dealing with discrete-time systems whose analytical expression becomes very complex due, for instance, to the use of finite duration real sampling. The expert network improves the discretization process implementing a biestimation mechanism that switches to the model that provides a better performance at each estimation instant considered for different values of the hold order.
11
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Celem pracy jest budowa samonastrajalnego modelu rozmytego zarządzania przedsięwzięciami informatycznymi, który będzie wykorzystywany do wspomagania wytwarzania systemów opartych na wiedzy. Realizacja takich systemów, traktowanych według COCOMO jako osadzone, przysparza kierownikom zespołów projektowych wielu problemów wynikających z ograniczonej znajomości dziedziny przedmiotowej, z braku narzędzi informatycznych do pozyskiwania i implementacji wiedzy oraz koordynacji współpracy ekspertów i inżynierów wiedzy. Z tych powodów poszukiwane są rozwiązania, które będą wspomagać procesy zarządzania przedsięwzięciami, a w szczególności zmianami, ryzykiem i czasem realizacji. Proponowany model jest odpowiedzią na takie zapotrzebowanie. Bazuje on na rozwiązaniach opartych na wiedzy oraz teorii systemów dynamicznych i zbiorów rozmytych. W szczegółowej konstrukcji wykorzystano wiedzę dotyczącą zarządzania rzeczywistymi przedsięwzięciami informatycznymi (PATRIC i ECOSIM), która pozwoliła na dostrojenie modelu w zakresie konstrukcji reguł baz wiedzy oraz funkcji przynależności. Wiedzę pozyskaną z realizacji trzeciego z projektów (SUTRA) wykorzystano do weryfikacji rozwiązania, m.in. do sprawdzenia mechanizmów jego samonastrajania. Zbadano stopień prawdziwości reguł oraz przeprowadzono ocenę reguł sprzecznych. Wykorzystano specyfikę przedsięwzięcia informatycznego polegającą na dokonywaniu zmian w metodach i narzędziach informatyki jedynie w początkowej, głównej i końcowej fazie przedsięwzięcia, co pozwoliło na budowę funkcji przynależności na podstawie metody klasteryzacyjnej. W metodzie tej bada się położenie środków ciężkości klastrów i stosownie do nich modyfikuje się odpowiednie parametry funkcji przynależności. Opracowany model rozmyty stanowi podstawę do konstrukcji systemu wspomagającego podejmowanie decyzji w przedsięwzięciach informatycznych wytwarzania systemów opartych na wiedzy.
EN
The aim of the paper is to construct the self adjusting fuzzy model of the software engineering management, which is going to aid the creation of knowledge based systems. Realization of such systems, which COCOMO treats as embedded, creates problems for managers of project teams. This, in turn, is connected with a limited knowledge on the subject matter, lack of IT tools for the acquisition and implementation of knowledge, as well as with difficulties in co-ordination of the cooperation between experts and engineers. Therefore new solutions that aid the project management processes, especially those related to the changes, risk and time of realization, are sought in this paper. Our suggested model, based on knowledge and fuzzy sets, offers an effective solution to the above problems. While building the system, the knowledge concerning a practical management of real software systems (PATRIC, ECOSIM) has been applied as real-world conditioning data in tuning of the model, building knowledge-base rules and membership functions. The knowledge obtained as a result of realization of another project (SUTRA) has been used in a procedure of verifying the obtained solution, checking suitability of its mechanisms, and tuning its parameters. The degrees of veracity of the rules have been studied, and conflicting rules have been assessed. Unique characteristics of software undertakings with respect to grouping the measurement data (of the changes in the management) into 3 project phases have been utilized in the design. Such an approach has enabled a simple application of clustering methods in constructing the membership function. In this procedure the locations of centers of gravity are studied and, according to their changes, the membership function is modified. The suggested fuzzy model treats software projects as a source of knowledge, which can be used for carrying out other projects. The model presents the mechanism of the acquisition, implementation and use of knowledge. It can be useful in designing new IT tools based on the experts' knowledge and in aiding the assessment of decision scenarios - in the case of undesirable changes in the realization process, taking into account the risk connected with the project completion in the assumed time horizon and at assumed financial costs.
Motion planning is the process of computing a path, i.e., a sequence of robot configurations, allowing it to move from one place to another. It is a central problem in the development of autonomous mobile robots. This is true indeed, but when the environment of the robot becomes complex, i.e., uncertain, partially known, with moving obstacles or other robots, it takes much more than global motion planning to achieve motion autonomy. In this case, the ability to detect unexpected events and react accordingly becomes essential. Reactivity provides the robot with an important mechanism to immediate respond to unpredicted environmental changes. This paper describes an intelligent path planning system for omnidirectional mobile robots. Our proposed solution to the dual need for global path planning and reactivity is to adopt a two-level model: at the upper level, a planner provides the system with a global path, based on the available knowledge; at the lower level, a reactive controller follows this given global path, while dealing with the environmental contingencies. The control architecture, presented in this paper, relies upon two main complementary modules: a global path planner, that computes a nominal path between the current configuration of the robot and its goal, and a reactive local planner, whose purpose is to generate the appropriate commands for the actuators of the robot, so as to follow the global path as close as possible, while reacting in realtime to unexpected events by locally adapting the robots movements, so as to avoid collisions with unpredicted or moving obstacles. This reactive local planner consists of two separate fuzzy controllers for path following and obstacle avoidance. The functioning of the proposed system with respect to omnidirectional mobile robots and results of simulated experiments will be presented.
In this paper, a partial model of a multiagent system is presented. This multiagent system is built of agents that use consensus methods to construct and update their knowledge of recognizable world states. Each agent in this system encapsulates a private database containing representations of empirically verified parts of knowledge of experienced world states (so-called encapsulated world profiles). In each agent and for each object of the world, this knowledge is computed by the agent as the consensus of encapsulated update resources. These update resources consist of the agent's perceptions of the object and similar perceptions communicated to the agent by other members of the same multiagent population. An example of encapsulated update resources is given and a list of related requirements for consensus computation is discussed.
Inconsistent information is one of main difficulties in the explanation and recommendation tasks of decision analysis. We distinguish two kinds of such information inconsistencies : the first is related to indiscernibility of objects described by attributes defined in nominal or ordinal scales, and the other follows from violation of the dominance principle among attributes defined on preference ordered ordinal or cardinal scales, i.e. among criteria. In this paper we discuss how these two kinds of inconsistencies are handled by a new approach based on the rough sets theory. Combination of this theory with inductive learning techniques leads to generation of decision rules from rough approximations of decision classes. Particular attention is paid to numerical attribute scales and preference-ordered scales of criteria, and their influence on the syntax of induced decision rules.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.