Decoding the Creative Ability of Subjects from Aesthetic Quality Assessment Using Dual Convolution Induced Capsule Network
- Resource Type
- Conference
- Authors
- Ghosh, Sayantani; Bose, Shirsha; Chowdhury, Ritesh Sur; Konar, Amit; Nagar, Atulya K.
- Source
- 2022 IEEE Symposium Series on Computational Intelligence (SSCI) Computational Intelligence (SSCI), 2022 IEEE Symposium Series on. :1319-1326 Dec, 2022
- Subject
- Computing and Processing
General Topics for Engineers
Robotics and Control Systems
Signal Processing and Analysis
Visualization
Convolution
Heuristic algorithms
Feature extraction
Brain modeling
Routing
Electroencephalography
scientific creativity
aesthetic sensibility
capsule network
- Language
Scientific creativity refers to the exhibition of original ideas in various domains of science that promotes technological progress. Aesthetic sensibility, the ability to comprehend the visual appeal of any object, plays a significant role in guiding the cognitive process of a scientific creation. The present study attempts to detect creative individuals in scientific field by analyzing their aesthetic quality judgment using Electroencephalographic (EEG) system. The major aim of the current study is accomplished by first acquiring brain signals from the subjects who were instructed to mentally assess the beauty of the presented stimuli. The acquired signals are pre-processed and transferred to a dual feature extraction module that abstracts spectrogram and brain connectivity based features as a strategy to enhance the diversity of the feature set. Brain connectivity pattern analysis confirms the active involvement of the right dorsolateral prefrontal and left pars orbitalis brain regions when the presented stimuli have been considered as beautiful. Moreover, spectrogram features reveal the activation of theta and alpha frequency bands for the present cognitive task. The dual features abstracted are fed to a novel classifier module referred as Dual Convolution Induced Capsule Network (DC-CapsNet) that categorizes the subjective creative ability from aesthetic sensibility into two classes: high and low. The novelty in the design of the proposed classifier consists of a dual convolution layer working in parallel to handle two types of features, the introduction of a new cross attention mechanism that highlights the relevant portions of the dual features and the addition of a Mish function based dynamic routing algorithm to enhance the prediction capacity of the capsule network. Comparative study performed with baseline models reveals the superior performance of the DC-CapsNet model.