A brain-computer interface (BCI) typically requires a calibration phase to train its decoding model using supervised data, which can be time-consuming and impractical in real-world scenarios. This study eliminates the need for such calibration by investigating two zero-training approaches using a P300 event-related potential (ERP) dataset. We first evaluate reconvolution canonical correlation analysis (CCA) on ERP data, which marks its first application beyond its original domain of code-modulated visual evoked potentials (c-VEP). Second, we examine unsupervised mean-difference maximization (UMM), a recently proposed method for calibration-free ERP decoding. For both approaches, we assess performance under two conditions: instantaneous classification and cumulative learning across previously classified trials. Surprisingly, despite CCA originating from the c-VEP domain, our results demonstrate that both CCA and UMM achieve comparably high classification accuracy (97 % and 100 %) in decoding ERP data without requiring a calibration session. We highlight the unique advantages that each method offers and discuss implications for future research. By enabling reliable, calibration-free decoding, this work supports the development of more practical and accessible BCIs across various stimulus paradigms and applications. The potential integration of CCA and UMM presents a promising avenue to further enhance BCI usability.

We developed task-optimized deep neural networks (DNNs) that achieved state-of-the-art performance in different evaluation scenarios for automatic music tagging. These DNNs were subsequently used to probe the neural representations of music. Representational similarity analysis revealed the existence of a representational gradient across the superior temporal gyrus (STG). Anterior STG was shown to be more sensitive to low-level stimulus features encoded in shallow DNN layers whereas posterior STG was shown to be more sensitive to high-level stimulus features encoded in deep DNN layers.