Elsevier

Neural Networks

Volume 41, May 2013, Pages 3-14
Neural Networks

2013 Special Issue
Cognitive memory

https://doi.org/10.1016/j.neunet.2013.01.016Get rights and content

Abstract

Regarding the workings of the human mind, memory and pattern recognition seem to be intertwined. You generally do not have one without the other. Taking inspiration from life experience, a new form of computer memory has been devised. Certain conjectures about human memory are keys to the central idea. The design of a practical and useful “cognitive” memory system is contemplated, a memory system that may also serve as a model for many aspects of human memory. The new memory does not function like a computer memory where specific data is stored in specific numbered registers and retrieval is done by reading the contents of the specified memory register, or done by matching key words as with a document search. Incoming sensory data would be stored at the next available empty memory location, and indeed could be stored redundantly at several empty locations. The stored sensory data would neither have key words nor would it be located in known or specified memory locations. Sensory inputs concerning a single object or subject are stored together as patterns in a single “file folder” or “memory folder”. When the contents of the folder are retrieved, sights, sounds, tactile feel, smell, etc., are obtained all at the same time. Retrieval would be initiated by a query or a prompt signal from a current set of sensory inputs or patterns. A search through the memory would be made to locate stored data that correlates with or relates to the prompt input. The search would be done by a retrieval system whose first stage makes use of autoassociative artificial neural networks and whose second stage relies on exhaustive search. Applications of cognitive memory systems have been made to visual aircraft identification, aircraft navigation, and human facial recognition.

Concerning human memory, reasons are given why it is unlikely that long-term memory is stored in the synapses of the brain’s neural networks. Reasons are given suggesting that long-term memory is stored in DNA or RNA. Neural networks are an important component of the human memory system, and their purpose is for information retrieval, not for information storage. The brain’s neural networks are analog devices, subject to drift and unplanned change. Only with constant training is reliable action possible. Good training time is during sleep and while awake and making use of one’s memory.

A cognitive memory is a learning system. Learning involves storage of patterns or data in a cognitive memory. The learning process for cognitive memory is unsupervised, i.e. autonomous.

Introduction

During the summer of 1956 a seminar was held at Dartmouth College on the subject of artificial intelligence. This was the first AI meeting, and it was open to all. The founders of the field were there. It was an exciting time, with great anticipation in the air. What was floating around was the idea of developing an artificial brain by means of software, hardware, or a combination of both. Today, after more than fifty years, the goal is still elusive. The problem is too big, and how the brain works is for the most part still largely unknown.

In this paper, we are not trying to explain how the brain works. We are only trying to understand how human memory works, and how to build a human-like memory for computers. Human memory with all its complexity is not yet understood. This paper approaches the problem and is able to yield insight into some of memory’s processes. An artificial memory can now be built taking advantage of what can be learned from nature. This kind of memory connected to a conventional digital computer facilitates solutions to problems in pattern recognition, control systems, and learning systems.

Fundamentally, human and animal pattern recognition involves matching an unknown incoming pattern with a pattern seen before and currently stored in memory. This does not fit all of the existing pattern recognition paradigms and not everyone will accept this, but this is what we believe. When a visitor appears and an interesting subject is being discussed, the “mental tape recorder” is recording the sights, sounds, etc. of the visit. In a half hour of discussion, perhaps 100,000 images of the visitor’s face are recorded. These images are retinal views that capture the visitor’s face with different translations, rotations, scale, light levels, perspectives, etc. These images are stored permanently in memory, in sequential locations wherever there is an empty place. The subject of the conversation is also recorded.

A digital computer has numbered memory registers. Data storage locations are program controlled, and data is retrieved when needed by calling for it by register number. Human memory has no numbered registers. Data is stored in human memory wherever there is an empty place and once stored, the memory has no idea where the data has been stored.

We contemplate a memory of enormous, almost unbelievable capacity, enough to hold many lifetimes of stored visual, auditory, tactile, olfactory, vestibular, etc. patterns of interest. Data and patterns are retrieved in response to an input query, an input pattern, whether visual, auditory, tactile, etc. or a combination of these. The input pattern serves as a prompt to initiate retrieval of related data patterns, if they are stored in the memory. If data patterns are retrieved and if they contain an identification, the input pattern is thereby identified. It is surprising that many aspects of human mental activity can be explained by such a simple idea of memory. Some of these aspects will be described below.

On the engineering side, we will introduce new approaches to computer memory and pattern recognition. Pattern matching is complicated by the fact that unknown incoming prompt patterns may be different from stored patterns, different in perspective, translation, rotation, scale, etc. Simple pattern matching may not be adequate, but this will be addressed.

The memory capacity must be enormous, and it should be implemented with a parallel architecture so that search time would be independent of memory size. There are many ways to structure a content-addressable memory. The “cognitive memory” proposed herein is content addressable and is of a unique design that could be physically built to give a computer a “human-like” memory, and furthermore, it is intended to serve as a behavioral model for some aspects of human and animal memory.

Section snippets

A design for a cognitive memory

The various components of the proposed cognitive memory system are represented by the diagrams of Fig. 1, Fig. 2, Fig. 3, Fig. 4. Fig. 1 shows storing input patterns from various sensors such as eyes, ears, etc. These patterns are stored in memory folders. Autoassociative neural networks are used for memory retrieval. Fig. 2 illustrates the training method for these neural networks. Fig. 3 shows how the trained neural networks are used for retrieval when the memory system is presented with a

Pattern recognition by cognitive memory

One application of the cognitive memory is in the field of pattern recognition. We first record a database of a very large number of patterns and their variations (rotation, translations, etc.), plus their identifications, if known, and the autoassociative neural networks are trained with the data patterns. The training is done off line, when the cognitive memory is not performing data retrieval.

When we are given an unknown object to be recognized, its image serves as a prompt pattern. If a

Face recognition

Yet another application for the cognitive memory system is that of face detection and recognition. If a person’s face approximately fills a square window having 20×20 pixels, displaying the face at this low resolution with only 400 pixels allows one to determine that the image is that of a person’s face, but it is almost impossible to determine from this image who the person is. A higher resolution image of the face is necessary to do recognition. A 50×50 pixel image with 2500 pixels will allow

A model of human memory

The cognitive memory system described above is a mechanistic system. It is designed to function in a useful way, and it is also designed to behave in a way that resembles the behavior of human memory as well as we understand this. Between both of the authors of this paper, we have experienced more than 100 years of living with human memory. Observations from everyday life have given us insight in the workings and the behavior of human memory and have led to the cognitive memory model described

Conclusions

The design of a practical and useful cognitive memory system is described based on certain ideas of how long-term human memory works. The cognitive memory is content addressable. Sensory inputs concerning a single object or subject are stored together in a memory folder. When the contents of the folder are retrieved, sights, sounds, tactile feel, smell, etc., are all obtained at the same time. Retrieval would be initiated by a query or prompt signal from a current set of sensory input patterns.

References (4)

  • E. Kandel

    In search of memory: the emergence of a new science of mind

    (2006)
  • P. Maquet

    Experience-dependent changes in cerebral activation during human rem sleep

    Nature Neuroscience

    (2000)
There are more references available in the full text version of this article.

Cited by (14)

  • Nature's learning rule: The Hebbian-LMS algorithm

    2023, Artificial Intelligence in the Age of Neural Networks and Brain Computing, Second Edition
  • Nature’s learning rule: The hebbian-LMS algorithm

    2018, Artificial Intelligence in the Age of Neural Networks and Brain Computing
  • A Survey on Challenges in Designing Cognitive Engines

    2020, 2020 6th International Conference on Web Research, ICWR 2020
View all citing articles on Scopus
View full text