A review of long-term memory in natural and synthetic systems

Abstract

Memory may be broadly regarded as information gained from past experience that is available in the service of ongoing and future adaptive behavior. The biological implementation of memory shares little with memory in synthetic cognitive systems where it is typically regarded as a passive storage structure. Neurophysiological evidence indicates that memory is neither passive nor centralized. A review of the relevant literature in the biological and computer sciences is conducted and a novel methodology is applied that incorporates neuroethological approaches with general biological inspiration in the design of synthetic cognitive systems: a case study regarding episodic memory provides an illustration of the utility of this methodology. As a consequence of applying this approach to the reinterpretation of the implementation of memory in synthetic systems, four fundamental functional principles are derived that are in accordance with neuroscientific theory, and which may be applied to the design of more adaptive and robust synthetic cognitive systems: priming, cross-modal associations, cross-modal coordination without semantic information transfer, and global system behavior resulting from activation dynamics within the memory system

Machines Learning - Towards a New Synthetic Autobiographical Memory

“…As noted in Wood et al [52], the cognitive science and neurobiology of memory has changed drastically over recent years, but synthetic approaches have not kept pace. In particular, models of episodic and autobiographical memory are not compatible with theories of active, distributed memory systems [43] or ideas such as the predictive brain [10].…”

“…Biological memory is highly distributed and tightly-woven into ongoing cognition [52]. The neural underpinnings of different kinds of memory and imagery cannot be separated from the circuitry of perception and action.…”

“…These models are largely symbolic, similar in spirit to classic models of cognition like ACT-R [1] and others (reviewed in [52]), with hand-set higher level representations and action scripts coordinated into event memories [37]. As in early cognitive science, these models of memory have appealed to the computer metaphor for the mind, though thorough critiques of this approach have been articulated [9].…”

“…vision, audition, touch, LIDAR, depth) at high bandwidth. However, robots remain poor at extracting or retrieving task-relevant information when needed, either offline from their vast archives, or online during streaming [52].…”

Modelling Concept Prototype Competencies using a Developmental Memory Model

“…The DAIM model operates on a set of four functional principles derived from the operation of memory within biological systems, embedded within the context of a wider cognitive system [5,32]. These principles are [32]: (1) memory as being fundamentally associative; (2) memory, rather than being a passive storage device, is an active component in cognition through activation dynamics; (3) memory as having a distributed structure; and finally (4) activation-based priming as subserved by the first three points.…”