EEG visual decoding aims to recover visual semantics from non‑invasive neural time‑series signals, and robust alignment between noisy EEG responses and stable semantic representations is essential. Existing contrastive methods usually rely on fixed visual or textual anchors; their semantic relations can become misaligned with EEG representations across trials, subjects, and learning stages, causing unstable alignment. Empirical results show that this instability appears in standard EEG decoding protocols as well as in more demanding robustness scenarios such as cross‑subject transfer and personalized continual adaptation. A formal analysis reveals that fixed semantic supervision biases optimization when EEG‑specific relations evolve, and structure‑agnostic perturbations may distort semantically important EEG components. To address these issues we propose Progressive Contrastive Alignment (ProCA), a unified model‑agnostic framework for adaptive neural‑semantic alignment. ProCA progressively refines class‑level contrastive supervision from frozen vision‑language priors to EEG‑aware semantic relations, and introduces structure‑consistent interpolation that constrains feature mixing according to channel‑wise and temporal importance. Across subject‑dependent, subject‑independent, strict cross‑subject transfer, and continual adaptation settings, ProCA achieves average relative Top‑1/Top‑5 gains of 7.4%/3.9%, 10.0%/4.6%, 28.1%/17.8% and 16.8%/11.6% respectively.
Review