The-KGP-Model-of-Agency-for-Global-Computing.pdf
《The-KGP-Model-of-Agency-for-Global-Computing.pdf》由会员分享,可在线阅读,更多相关《The-KGP-Model-of-Agency-for-Global-Computing.pdf(27页珍藏版)》请在三一文库上搜索。
1、The KGP Model of Agency for Global Computing: Computational Model and Prototype Implementation A. Bracciali1, N. Demetriou2, U. Endriss3, A. Kakas2, W. Lu4, P. Mancarella1, F. Sadri3, K. Stathis4,1, G. Terreni1, and F. Toni3,1 1 Dip. di Informatica, Universit a di Pisa braccia,paolo,terrenidi.unipi.
2、it 2 Dept of Computer Science, Cyprus University demetriou,antoniscs.ucy.ac.cy 3 Dept of Computing, Imperial College London ue,fs,ftdoc.ic.ac.uk 4 School of Informatics, City University London lue,kostassoi.city.ac.uk Abstract. We present the computational counterpart of the KGP (Kno- wledge, Goals,
3、 Plan) declarative model of agency for Global Computing. In this context, a computational entity is seen as an agent developed us- ing Computational Logic tools and techniques. We model a KGP agent by relying upon a collection of capabilities, which are then used to defi ne a collection of transitio
4、ns, to be used within logically specifi ed, context sensitive control theories, which we call cycle theories. In close relation- ship to the declarative model, the computational model mirrors the log- ical architecture by specifying appropriate computational counterparts for the capabilities and usi
5、ng these to give the computational models of the transitions. These computational models and the one specifi ed for the cycle theories are all based on, and are signifi cant extensions of, existing proof procedures for abductive logic programming and logic program- ming with priorities. We also disc
6、uss a prototype implementation of the overall computational model for KGP. 1Introduction Global Computing (GC) and its applications rely upon computing environments that are composed of autonomous computational entities whose activity is not centrally controlled but is decentralised instead. Decentr
7、alisation results either because global control is impossible or at times impractical, or because the en- tities are created or controlled by diff erent owners. The computational entities may be mobile, due to the movement of the physical platforms or by movement of the entities from one platform to
8、 another. In other words, the environment in which the entities are situated is open and evolves over time. For instance, in a typical GC application it might be required to allow for the introduction and deletion of computational entities. The internal structure and behaviour of these entities may
9、also be heterogeneous and may vary over time. Programming the behaviour of a computational entity that is situated in a GC environment is a non-trivial task. One of the problems is that such an en- tity should be in a position to operate with incomplete information about the environment. Incompleten
10、ess might arise from the entities having newly joined the environment of an application and having only a partial view over the status of that application. Incompleteness might also arise from the autonomy of the entities and their unwillingness to disclose information about themselves. More- over,
11、incompleteness might sometimes be caused by the fact that information in a GC environment becomes rapidly out of date. Thus, a GC entitity needs to be able to discover relevant information or other entities in the dynamically evolving environment. If the ultimate goal of GC research is to provide a
12、solid scientifi c foundation for the design of GC systems, we will need to lay the groundwork for achieving eff ective principles for building and analysing such systems. In trying to achieve this goal, within the GC project SOCS we interpret the GC vision as follows. Entities in GC systems are defi
13、 ned via Computational Logic (CL), as understood in 26,29,27, which is used to defi ne their internal organisation, reasoning and their mutual interactions. We call the entities computees, standing for agents in CL. 5 One important feature of computees is that they are able to reason by using CL too
14、ls and techniques. We call the systems composed of such entities societies (see 7) as they are characterised by “social rules” for computees to interact and operate in the presence of each other. In order to interact freely, computees can use high-level communication, as understood in multi-agent sy
15、stems. Computees may be heterogeneous as far as behaviour is concerned, provided by CL-based cycle theories allowing a highly modular and fl exible specifi cation of control. Cycle theories allow to render com- putees adaptable to dynamically changing environments and allow to charac- terise, via di
16、ff erent cycle theories, heterogeneously behaved computees. Computees also need to adapt their internal state as the environments in which they are situated evolve. A number of CL techniques have been developed for addressing tasks such as temporal reasoning in a changing environment, hy- pothetical
17、 reasoning for dealing with incomplete information, hypothetical rea- soning for planning, hypothetical reasoning to achieve communication, argumen- tation for decision-making, inductive logic programming for learning. However, in order to cope with the GC challenges, CL techniques in isolation are
18、inade- quate, as none serves all dimensions in the operation of computees. Our model for computees integrates (extensions of) a number of existing CL techniques, in order to achieve the enhanced performance which is required by the GC vision. We call our model KGP, since computees internal state con
19、sists of a knowl- edge base (K), from which they reason, goals (G) that they need to achieve, and plans (P) for their goals, consisting of actions that may be physical, sens- ing or communicative. Computees pursue their goals while being alert to the environment and adapt their goals and plan to any
20、 changes that they perceive. The paper is organised as follows. In section 2 we summarise the main fea- tures of the KGP model and give some of the technical details underlying it. In sections 35 we provide the computational models of some components of the KGP model and state their soundness wrt th
21、eir formal specifi cation. The overall 5 In this paper, we will use the terms computees and agents interchangeably. computational model is built bottom-up, mirroring the hierarchical and modular structure of the abstract model. Section 3 also gives some background on the CL techniques that we have e
22、mployed to defi ne the KGP model, namely Abductive Logic Programming (ALP) and Logic Programming with Priorities (LPP), as well as the proof procedures (for ALP and for LPP) from which we have built the computational counterpart of the KGP model, in a bottom-up fashion. In section 6 we describe the
23、prototype implementation of KGP agents, namely the SOCS-iC (for SOCS individual Computee) component of the PROSOCS plat- form 36. Section 7 concludes. 2KGP model: recap Here we briefl y summarise the KGP model for computees, see 18,17 for any additional details. This model relies upon an internal (o
24、r mental) state, a set of reasoning capabilities, supporting planning, temporal reasoning, identifi cation of preconditions of actions, reactivity and goal decision, a sensing capability, a set of transition rules, defi ning how the state of the computee changes, and defi ned in terms of the above c
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- The KGP Model of Agency for Global Computing
链接地址:https://www.31doc.com/p-3800839.html