The Synthesis of ACT-R and Leabra (SAL) hybrid cognitive architecture is the integration of two theories of cognitive functioning, each itself a highly integrative theory of cognition, ACT-R being predominantly a symbolic production-rule based architecture and Leabra a neural modeling architecture. The combination of the two architectures allows for richer dynamics that take advantage of neural and symbolic aspects and provides mutual constraints that promote convergence towards models that are both neurophysiologically and psychologically valid. We present a hybrid model that makes use of multi-level and multi-system integration to allow an instructed assembly task to be carried out in way that is noise and error robust. Specifically, the model shows how higher-level error recovery routines can interface with lower-level sensory, motor, and error detection processes and result in a robustness to noise and noise-induced errors. Multiple systems and processes operating at multiple levels are recruited to provide a way around the limitations of simpler systems composed of isolated modules that do not allow information to be propagated as easily. The benefits of this approach provide motivation for the adoption of a generally integrated approach to cognitive systems.

We present a model of unsupervised learning in the hybrid SAL (Synthesis of ACT-R and Leabra) architecture. This model follows the hypothesis that higher evaluative cognitive mechanisms can serve to provide training signals for perceptual learning. This addresses the problem that supervised learning seems necessary for strong perceptual performance, but explicit feedback is rare in the real world and difficult to provide for artificial learning systems. The hybrid model couples the perceptual strengths of Leabra with ACT-R's cognitive mechanisms, specifically its declarative memory, to evolve its own symbolic representations of objects encountered in the world. This is accomplished by presenting the objects to the Leabra visual system and committing the resulting representation to ACT-R's declarative memory. Subsequent presentations are either recalled as instances of a previous object category, in which case the positive association with the representation is rehearsed by Leabra, or they cause ACT-R to generate new category labels, which are also subject to the same rehearsal. The rehearsals drive the network's representations to convergence for a given category; at the same time, rehearsals on the ACT-R side reinforce the chunks that encode the associations between representation and label. In this way, the hybrid model bootstraps itself into learning new categories and their associated features; this framework provides a potential approach to solving the symbol grounding problem. We outline the operations of the hybrid model, evaluate its performance on the CU3D-100 (cu3d.colorado.edu) image set, and discuss further potential improvements to the model, including the integration of motor functions as a way of providing an internal feedback signal to augment and guide a purely bottom-up unsupervised system.

The major limitation to standard classification techniques is that the classifiers have to be trained on objects for which the ground truth, ACT-R contains a robust declarative memory module, which in terms of either a pre-assigned label or an error signal, is stores information as "chunks." A chunk in ACT-R may contain known. This limitation prevents the classifiers from dynamically any number of slots and values for those slots; slot values developing their own categories of classification based may be other chunks, numbers, strings, lists, or generally on information obtained from the environment. Previous attempts any data type allowed in Lisp (the base language for to overcome these limitations have been based on ACT-R). Retrieval from declarative memory is handled by a classical machine learning algorithms (Modayil and Kuipers request to the retrieval module; the request specifies the conditions 2007) (Kuipers et al. 2006). Here we present an alternative to be met in order for a chunk to be retrieved from approach to this problem, and develop the beginnings of declarative memory, and the module either returns a chunk a framework within which a classifier can evolve its own matching those specifications or generates a failure signal if representations based on dynamical information from the a retrieval cannot be made.