PredANNpp-NMEDT-SongID-EncoderOnly-Surprisal-ctx16-pt10000-ft3500-seed42

PredANNpp-NMEDT-SongID-EncoderOnly-Surprisal-ctx16-pt10000-ft3500-seed42

Model description

This repository contains a PredANN++ encoder-only PyTorch Lightning checkpoint for NMED-T Song ID classification from 3-second EEG segments.

• Repository: Shogo-Noguchi/PredANNpp-NMEDT-SongID-EncoderOnly-Surprisal-ctx16-pt10000-ft3500-seed42
• Checkpoint file: predannpp_nmedt_songid_encoderonly_surprisal_ctx16_pt10000_ft3500_seed42.ckpt
• Stage: pretraining + finetuning
• Target / teacher representation: MusicGen Surprisal, 16 s context
• Architecture: Encoder-only Song ID classifier
• Pretraining: 10k epochs with 50% masking
• Finetuning: 3.5k epochs for Song ID classification
• Random seed: 42
• SHA256: 5a647c49765ad19565cf0c6dcf199687acd40fb0c4cff24c4d56c512910f123e

This checkpoint is a task-specific Song ID classifier. It directly maps a 3-second EEG segment to 10-class Song ID logits.

Capabilities

• Input: 128-channel EEG, 125 Hz, 3-second segments.
• Output: 10-class Song ID logits.
• Intended module type: encoder-only finetuning module.

For masked teacher-token prediction, use a PredANNpp-Pretrain-* checkpoint instead.

Training data

• Dataset: NMED-T (Naturalistic Music EEG Dataset – Tempo), 10 songs, 20 subjects, trial=1.
• Teacher / target source: MusicGen Surprisal token sequences with 16-second context.

Training procedure

1. Multitask pretraining with masked teacher-token prediction and auxiliary Song ID learning.
2. Encoder-only finetuning for NMED-T Song ID classification.

Intended use

• Reproducing PredANN++ Song ID classification experiments.
• Evaluating EEG-based music identification from 3-second EEG segments.
• Comparing acoustic and expectation-related teacher representations.

Not intended use

• Medical diagnosis, clinical decision making, or biometric identification.
• Commercial use without checking the PredANN++ code license, NMED-T terms, and upstream model/feature licenses.
• Masked-token pretraining evaluation; this repository contains a finetuned encoder-only classifier.

License and upstream dependencies

This checkpoint uses MusicGen / AudioCraft-derived features. The checkpoint is released under CC-BY-NC-4.0 for compatibility with NMED-T-derived training artifacts and upstream feature dependencies.

Reproducibility notes

• metadata.json stores standardized release metadata.
• SHA256SUMS stores the checkpoint checksum.
• Use the PredANN++ GitHub repository for model definitions and evaluation scripts.

Links

• Project page: https://shogonoguchi.github.io/PredANNpp/
• GitHub: https://github.com/ShogoNoguchi/PredANNpp
• Hugging Face collection: https://huggingface.co/collections/Shogo-Noguchi/predann-models
• Paper: https://arxiv.org/abs/2603.03190

Citation

If you use this checkpoint, please cite:

@misc{noguchi2026expectationacousticneuralnetwork,
  title={Expectation and Acoustic Neural Network Representations Enhance Music Identification from Brain Activity},
  author={Shogo Noguchi and Taketo Akama and Tai Nakamura and Shun Minamikawa and Natalia Polouliakh},
  year={2026},
  eprint={2603.03190},
  archivePrefix={arXiv},
  primaryClass={cs.AI},
  url={https://arxiv.org/abs/2603.03190}
}