Start here

Start at these documents if you want to get a general idea about the overall architecture or specific features of Xapagy:

L. Bölöni. Xapagy: the ongoing document

This is an overall introduction of the general ideas of the Xapagy architecture (as it was in March 2012). While some aspects of the architecture had been refined since then, this is the most complete description of the internal architecture.

 

L. Bölöni. Shadows and headless shadows: a worlds-based, autobiographical approach to reasoning, arXiv:1211.5643, 2012

 A write-up about the general principles of the shadowing and headless shadow mechanism.

 

Technical reports

These documents discuss specific issues in the Xapagy architecture, usually in a rather technical manner.

XTR-001: A cookbook of translating English to Xapi

Describes, through a series of examples the ways in which one can express English expressions in Xapi (focusing on the representations with scenes and identities).

XTR-002: Design patterns for domain engineering in Xapagy (TBD)

A Xapagy domain is a set of concepts, verbs and relations which provide the basis for representing stories of a certain type. 

A Xapagy agent will normally rely on a collection of domains, some of them being dependent on each other. For instance, an agent might load successively the domain of spatial relations, followed by domains representing animate objects, humans and the game of soccer.

This document describes the principles which have been found useful when creating domains through manual knowledge engineering. The focus of this document is on patterns used to express overlapping and contradictory attributes and relations as well as related verbs.

XTR-003: Building a synthetic autobiography for Xapagy agents (TBD)

This technical report describes plans for creating a synthetic autobiography for a Xapagy agent. We conjecture that the early part of autobiography must consist of a series of stages in which the agent learns to use increasingly abstract representations. These are the concrete sensing stage, anchored communication stage, abstract communication stage followed by the accumulation stage in which the agent will spend the reminder of its life-cycle. We describe the nature of the narratives encountered by the agent in each of these stages, and estimate their size and difficulty to build.  

XTR-004: An algebra of concepts, verbs and overlays

The Xapagy cognitive architecture relies on objects which are characterized with overlays of concepts and verbs. This report describes the reasoning behind the existence of this framework, by describing the affordances which the framework is intended to provide for the system. These affordances are discovery, change, matching, reference, learning and externalization.

The system of concepts, verbs and overlays implement these affordances by relying on a specific algebra which manipulates these entities. We describe the mathematical forms of the operations defined on these entities. Finally, through a series of examples relying on the artificial domain, we illustrate the practical results of these operators.

XTR-005: Shadow maintenance in Xapagy (TBD)

The shadowing architecture is the main reasoning technique of the Xapagy architecture. Both the instances and the verb instances (VIs) in the focus are shadowed by relevant entities from the autobiographical memory. These shadows, on their turn, are used for the various activities of the agent, such as the reasoning about the current narrative, prediction of the future and taking actions towards a goal.

This technical report describes the implementation of the shadow mechanism in the Xapagy architecture. We describe the ways in which shadows are measured through their energy, activation and relative activation levels. The shadows are maintained by a series of diffusion activities (DAs), which means that their values continuously change function of the current set of entities in the focus, but also the other shadows. We describe the operation of these DAs in detail.

Finally, we will describe the results of a number of experiments which trace the various values of DAs in time.

XTR-006: Maintenance of headless shadows in the Xapagy cognitive architecture (TBD)

The Xapagy cognitive architecture operates by maintaining in the focus the currently followed narrative in form of a collection of instances and verb instances (VIs). These entities are then shadowed by relevant entities from the autobiographical memory. The relations between shadows are then used to bring in entities from the autobiographical memory which are not directly mapped to entities in the focus. In general, we will refer to these entities as headless shadows – although in the process of their maintenance and use in subsequent reasoning, we will introduce a number of more specific types: SVRs, SVRIs, HLSs and choice objects.

This technical report describes the way in which the collection of these objects are maintained by the Xapagy agent.

XTR-007: Focus maintenance in Xapagy

In a Xapagy agent the focus contains instances and VIs which are currently referred by the agent. These entities are created, they spend a certain amount of time in the focus, acquire salience in the autobiographical memory, then leave the focus. The focus participation is affected both by sudden changes represented by spike activities (SAs) and gradual changes described by diffusion activities (DAs). This technical report describes the SAs and DAs affecting the focus and describes several experiments studying their impact.

Other readings

 

Some of these might be of interest, but most of this information was overwritten by more recent writeups.

 

L. Bölöni. An investigation into the utility of episodic memory for cognitive architectures. In AAAI Fall Symposium on Advances in Cognitive Systems, November 2011.

L. Bölöni. Modeling problems of identity in Little Red Riding Hood. arXiv:1211.5644, 2012

L. Bölöni. Shadows and headless shadows: an autobiographical approach to narrative reasoning. arXiv:1211-6097

L. Bölöni. Xapagy: a cognitive architecture for narrative reasoning. arXiv:1105.3486v1, 2011.