Entropic associative memory for real world images

TL;DR

It is shown that EAM appropriately stores, recognizes and retrieves complex and unconventional images of animals and vehicles and generates meaningful retrieval association chains for such complex images.

Abstract

The entropic associative memory (EAM) is a computational model of natural memory incorporating some of its putative properties of being associative, distributed, declarative, abstractive and constructive. Previous experiments satisfactorily tested the model on structured, homogeneous and conventional data: images of manuscripts digits and letters, images of clothing, and phone representations. In this work we show that EAM appropriately stores, recognizes and retrieves complex and unconventional images of animals and vehicles. Additionally, the memory system generates meaningful retrieval association chains for such complex images. The retrieved objects can be seen as proper memories, associated recollections or products of imagination.

Figures (11)

• Figure 1: Actions taking place in the $\lambda$-register (top) and $\eta$-recognition (bottom) operations.
• Figure 2: Actions taking place in the $\beta$-retrieval operation.
• Figure 3: Performance of the networks. The left side presents the mean RMSE of the autoencoder, this error is scaled to a percentage of the maximum value (255) for an RGB channel. The right side displays the mean accuracy values of the classifier.
• Figure 4: System architecture. The variable $p$ denotes the size of the Cifar10 images, $n$ indicates the number of extracted features, and $c$ is the size of the probability distribution of classifying a cue in the latent space as one the Cifar10 classes. Out of such distribution $argmax$ chooses the class number with the highest probability. Solid and dashed lines represent real and integer numbers, respectively.
• Figure 5: Memory performance for the AMRs of size $n\times m$. The large numbers located at the left indicate the number of columns. The Range Quantization Levels in the $X$-axes correspond to the number of rows. The line graphs show the precision and the recall for each AMR, with the associated entropy values at the bottom horizontal bar. The bar charts display the type and amount of responses for each register.