Developing Deep Learning Models to Simulate Human Declarative Episodic Memory Storage

Abu Kamruzzaman,Charles C

doi:10.14569/ijacsa.2019.0100302

Abstract

Human like visual and auditory sensory devices became very popular in recent years through the work of deep learning models that incorporate aspects of brain processing such as edge and line detectors found in the visual cortex. However, very little work has been done on the human memory, and thus our aim is to model human long-term declarative episodic memory storage using deep learning methods. An innovative way of deep neural network was created on supervised feature learning dataset such as MNIST to achieve high accuracy as well as storing the models hidden layers for future extraction. Convolutional Neural Network (CNN) learning models with transfer learning models were trained to imitate the long-term declarative episodic memory storage of human. A Recurrent Neural Network (RNN) in the form of Long Short Term Memory (LSTM) model was assembled in layers and then trained and evaluated. A Variational Autoencoder was also used for training and evaluation to mimic the human memory model. Frameworks were constructed using TensorFlow for training and testing the deep learning models.

Highlights

The aim of this research is to construct a deep learning model to simulate the human brain long-term declarative episodic memory storage, focusing primarily on the computer science perspective of the Rosenblatt Model for experiential storage in neural networks [1]
Previous work showed that Convolutional Neural Networks (CNN) models work well for classification of spatial data while CNN was unable to store the hidden layers for future predictions [2]
Our new hypothesis is that an integrated framework of CNN, Long Short Term Memory (LSTM) and Variational Autoencoder (VAE) adequately stores images for future recall

Summary

Introduction

The aim of this research is to construct a deep learning model to simulate the human brain long-term declarative episodic memory storage, focusing primarily on the computer science perspective of the Rosenblatt Model for experiential storage in neural networks [1]. It is not known completely how human memory remembers past events. The dataset might not generate accurate results while used inaccurately and larger dataset increase the amount of inconsistency of generating errors [3] This issue can be resolved through additional training on the larger dataset

Objectives

Methods

Results

Conclusion