Information Bottleneck as Optimisation Method for SSVEP-Based BCI.

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Brain-computer interfaces (BCIs) translate neural signals into commands.
• Steady-state visual evoked potentials (SSVEPs) are commonly used in BCIs.
• Current SSVEP BCIs often use suboptimal classification rules like the argmax classifier.

Purpose of the Study:

• To propose and evaluate the information bottleneck method as an optimization technique for SSVEP-based BCIs.
• To develop a novel classification method that optimizes BCI performance measures.
• To explore a new application of the information bottleneck in representation learning for BCIs.

Main Methods:

• Applied the information bottleneck method to optimize classification rules for SSVEP BCIs.
• Utilized standard feature extraction methods: power spectral density analysis (PSDA) and canonical correlation analysis (CCA).
• Employed task-related component analysis (TRCA) on a second dataset.

Main Results:

• The proposed information bottleneck approach demonstrated superior performance compared to standard methods on two public datasets.
• Outperformed related studies on one dataset using PSDA and CCA.
• Outperformed the argmax classifier in information transfer rate on the second dataset, especially with few classes.

Conclusions:

• The information bottleneck offers a novel and effective optimization strategy for SSVEP BCIs.
• This method optimizes information transfer and classification, potentially improving BCI accuracy and calibration.
• Represents the first known application of the information bottleneck in SSVEP-based BCIs.