The Digital Synaptic Neural Substrate: Size and Quality Matters

📝 Abstract

We investigate the ‘Digital Synaptic Neural Substrate’ (DSNS) computational creativity approach further with respect to the size and quality of images that can be used to seed the process. In previous work we demonstrated how combining photographs of people and sequences taken from chess games between weak players can be used to generate chess problems or puzzles of higher aesthetic quality, on average, compared to alternative approaches. In this work we show experimentally that using larger images as opposed to smaller ones improves the output quality even further. The same is also true for using clearer or less corrupted images. The reasons why these things influence the DSNS process is presently not well-understood and debatable but the findings are nevertheless immediately applicable for obtaining better results.

💡 Analysis

We investigate the ‘Digital Synaptic Neural Substrate’ (DSNS) computational creativity approach further with respect to the size and quality of images that can be used to seed the process. In previous work we demonstrated how combining photographs of people and sequences taken from chess games between weak players can be used to generate chess problems or puzzles of higher aesthetic quality, on average, compared to alternative approaches. In this work we show experimentally that using larger images as opposed to smaller ones improves the output quality even further. The same is also true for using clearer or less corrupted images. The reasons why these things influence the DSNS process is presently not well-understood and debatable but the findings are nevertheless immediately applicable for obtaining better results.

📄 Content

The Digital Synaptic Neural Substrate:
Size and Quality Matters

Azlan Iqbal College of Computer Science and Information Technology, Universiti Tenaga Nasional
Putrajaya Campus, Jalan IKRAM-UNITEN, 43000 Kajang, Selangor, Malaysia azlan@uniten.edu.my

ABSTRACT

We investigate the ‘Digital Synaptic Neural Substrate’ (DSNS) computational creativity approach further with respect to the size and quality of images that can be used to seed the process. In previous work we demonstrated how combining photographs of people and sequences taken from chess games between weak players can be used to generate chess problems or puzzles of higher aesthetic quality, on average, compared to alternative approaches. In this work we show experimentally that using larger images as opposed to smaller ones improves the output quality even further. The same is also true for using clearer or less corrupted images. The reasons why these things influence the DSNS process is presently not well-understood and debatable but the findings are nevertheless immediately applicable for obtaining better results.

Keywords: artificial intelligence, creativity, brain, process, images, chess.

1 INTRODUCTION

In previous work we have explained and demonstrated in some detail the ‘Digital Synaptic Neural Substrate’ (DSNS) as a generic approach in computational creativity that can be used to generate a virtually unlimited number of creative objects, using chess problems or puzzles as the domain of investigation (Iqbal et al., 2016). We will therefore not repeat any material that has already been covered. For the benefit of readers, we can summarize that the DSNS approach involves automatically analyzing objects (e.g. a chess problem or photograph) for their attributes (e.g. number of pieces, piece value or material difference, image resolution, the number of colors used) that can be represented in numerical form. The ‘creative difference’ (i.e. ‘deviation value’) between two such objects is then calculated using a simple formula in order to pseudo-randomly generate the attributes for two new theoretical objects that produce the same or similar deviation value. These differing attribute values can then be used by an external object-generation system to produce the two new theoretical ‘child’ objects that have the potential of being similar in creative value to their ‘parents’. As a very simple example, if we start with two boxes that have only height, width and length measurements in inches as their attributes, i.e. say, 5 x 6 x 7 and 7 x 8 x 9, the DSNS process may produce two new sets of attributes such as 4.5 x 2.25 x 8 and 6 x 3 x 1 and these new measurements can be used to create or build two new boxes that may be as attractive as their parents.

In the aforementioned previous work (Iqbal et al., 2016) it was demonstrated that photographs of people (as opposed to say, paintings and computer-generated art) used in combination with sequences taken from chess games between weak players (as opposed to say, chess problems by experienced composers) produced the highest quality output via the DSNS approach. The reason for this remains an open question. The output referred to are computer-generated chess problems, as the output type can be from either one of the source domains used (in this case, photographs or chess problems). There are other open questions remaining as well such as why the DSNS approach should even work at all and we will not be attempting to address them here. Instead, we will focus on extending the work by demonstrating experimentally how the quality of the output can be further improved by taking into account the properties of the images used in the process. Section 2 explains the experimental design, section 3 presents a discussion of the results and section 4 concludes with some directions for further work.

2 EXPERIMENTAL DESIGN

The basic idea in this research was to test if using larger images and those of higher quality in the DSNS process as described in (Iqbal et al., 2016) could improve the quality of the output. For the experiments we used the Chesthetica (v9.99) computer program from earlier work (Iqbal et. al, 2016) to evaluate the aesthetics of all the chess problems it created. This was the criterion used in determining a better or higher quality object output from the system. Even though the difference between two aesthetic scores may be small, that difference nevertheless ranks one chess problem above the other. Chesthetica uses an experimentally-validated aesthetics model that is able to rank chess problems aesthetically in a way that correlates positively and well with domain- competent human assessment (Iqbal et al., 2012). It is the most cost-effective and reliable method of evaluating beauty in chess move sequences as opposed to say, using human experts whom are not only expensive but also physically and mentally incapable

This content is AI-processed based on ArXiv data.