Is this idea already taken?

This paper focuses on building foundation models to learn generic EEG representations, which is central to your goal of creating embeddings from EEG signals.

This paper explicitly discusses foundation models designed to extract unified representations (embeddings) from EEG data.

It addresses self-supervised learning for EEG, which is the standard technical approach for generating the generic embeddings you are interested in.

While it deals with EEG-based BCIs, its focus is on adversarial attacks rather than generating generic representation embeddings.

This evaluates foundation models for EEG, which inherently relies on the generation of the generic EEG embeddings you are researching.

It focuses on benchmarking foundation models for EEG, aligning closely with the goal of developing and utilizing generic EEG embeddings.

It uses time series foundation model embeddings but focuses on label transfer across devices rather than generic EEG signal representation.

This is a specific application paper about decoding tactile sensations, not about generic EEG representation learning.

This paper focuses on eliminating drowsiness effects in specific BCI tasks, which is unrelated to general-purpose embedding generation.

This paper is about TMS-EEG signal recording and artifact management, not embedding space construction.

This paper is about LLM caching and synthetic data, which is completely off-topic from EEG signals.

It uses contrastive learning for embeddings but is focused on clinical patient risk progression rather than generic EEG signals.

This is about monocular depth estimation and visual representation learning, not EEG signals.

This paper is about representation learning from dendrograms, which is not applicable to EEG signal embeddings.

This focuses on self-supervised representations in interactive agent environments, not EEG signals.

This paper performs a systematic evaluation of existing EEG foundation models, which aligns with the benchmarking task of the reference.

This is the source paper for the BrainWave model mentioned in the reference topic.

This paper focuses on learning robust representations for EEG, which directly aligns with the goal of creating useful EEG embeddings for downstream tasks.

Focuses on large-scale pre-trained EEG foundation models which aligns directly with the goal of creating LBMs for brainwave signal embeddings.

It directly evaluates continuous wavelet transform (CWT) for sleep staging, which is the specific alternative to STFT mentioned in the reference.

Uses recurrent neural networks on EEG data, though the focus is on event detection rather than general representation learning for embeddings.

Focuses on building generalizable EEG decoding models, which aligns with capturing signal characteristics for foundation models, though it is more decoding-specific.

Directly investigates the use of general time series foundation models for EEG signals, aligning perfectly with the goal of creating powerful EEG representations.

It explicitly fuses EEG and EMG for motor rehabilitation, aligning well with the reference's goal of processing diverse biosignals.

It addresses the reference topic of combining EEG and EMG for rehabilitation, though it focuses on classification rather than generalized embeddings.

It addresses the root topic of creating embeddings (via foundational models) for diverse physiological signals like EEG and ECG.