Models

This section documents the foundation models shipped with NeuralBench and the quirks (channel-mapping assumptions, pretraining-data overlap, dropout overrides) that need to be considered when evaluating them. Task-specific architectures (eegnet, deep4net, eegconformer, atcnet, bdtcn, ctnet, shallow_fbcsp_net, simpleconv_time_agg) are off-the-shelf braindecode models with no NeuralBench-specific configuration; see their upstream documentation or models/<name>.yaml for hyperparameters.