.png){kind=icon}
.png){kind=icon}
Return
Evaluating Visualization Techniques for Binary Populations in High-Dimensional Evolutionary Search
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2026.3671826.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The effectiveness of Interactive Evolutionary Computation (IEC) relies fundamentally on the synergy between algorithmic power and human intuition. However, for hard combinatorial problems encoded as high-dimensional binary vectors, such as the Maximum Clique Problem, visualizing the population state remains a critical bottleneck. Traditional dimensionality reduction or direct plotting techniques often fail to convey topological nuances without inducing excessive cognitive load. This paper presents an empirical study evaluating the perceptual effectiveness of five visualization techniques: two baseline approaches (binary heatmap and MDS scatterplot) and three novel experimental metaphors (CNN-generated imagery, glyph-based sketch, and random particles). Through a controlled user study, we quantify human performance across four cognitive operations essential for Human-in-the-Loop systems: similarity perception, short-term visual memory, loop detection, and evolutionary trend analysis. Results reveal a fundamental trade-off between global awareness and local precision. While the MDS scatterplot was the most efficient (although equally effective in terms of accuracy) for monitoring macroscopic evolutionary trends, it failed significantly in local inspection tasks. Conversely, the proposed generative metaphors (specifically Sketch and CNN) demonstrated superior accuracy for perceiving similarity, retaining short-term memory, and detecting cyclic anomalies. Consequently, we advocate for utilizing coordinated views that leverage spatial projections for global steering combined with generative metaphors for precise local intervention.
Keywords:
Evolutionary computation
human-in-the-loop
information visualization
high-dimensional data
interactive optimization
visual perception