Research map / Label-efficient and robust learning / Recognition and long-tail
Quantifying the Generalization Gap in Seizure Detection: A Large-Scale Empirical Benchmark via the SzCORE Challenge
Jonathan Dan, Amirhossein Shahbazinia, Christodoulos Kechris, David Atienza
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| evaluate on the independent Filadelfia set instead of developers' self-selected datasets | 28 submitted seizure detectors · Filadelfia | event-based F1 | Hurt |
| evaluate Sz Transformer on Filadelfia instead of SeizeIT1 | Sz Transformer · event-based F1 | event-based F1 (%): 36 -> 32 | Hurt |
| evaluate DeepSOZ-HEM on Filadelfia instead of SeizeIT1 | DeepSOZ-HEM · event-based F1 | event-based F1 (%): 24 -> 30 | Improved |
| evaluate SeizUnet on Filadelfia instead of SeizeIT1 | SeizUnet · event-based F1 | event-based F1 (%): 32 -> 11 | Hurt |
| evaluate eventNet on Filadelfia instead of Siena | eventNet · event-based F1 | event-based F1 (%): 53 -> 14 | Hurt |
| evaluate zhu-transformer on Filadelfia instead of Siena | zhu-transformer · event-based F1 | event-based F1 (%): 48 -> 19 | Hurt |
| raise classification threshold from S4Seizure v1 to v2 (ablation) | S4Seizure · Filadelfia | event-based F1 (%): 27 -> 27 | No clear effect |
| lower threshold from S4Seizure v2 to v3 (ablation) | S4Seizure · Filadelfia | event-based F1 (%): 27 -> 22 | Hurt |
| change HySEIZa architecture/threshold from v1 to v2 (ablation) | HySEIZa · Filadelfia | event-based F1 (%): 24 -> 14 | Hurt |
| use top mean-rank algorithm HySEIZa v1 instead of F1-leader Sz Transformer | Sz Transformer · Filadelfia | per-subject rank: Cliff's delta 0.01 | No clear effect |
| use top algorithms instead of classical Random Forest | Random Forest · Filadelfia | per-subject F1 rank: Cliff's delta > 0.6 | Improved |
| seizure shorter duration (median 48 s vs 118 s) | seizure detectors · Filadelfia | detection rate: 48 s vs 118 s | Hurt |
Closest papers by meaning
- Long-Term EEG Partitioning for Seizure Onset Detection (AAAI 2025 · similarity 0.90)
- DMNet: Self-comparison Driven Model for Subject-independent Seizure Detection (NeurIPS 2024 · similarity 0.89)
- NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces (NeurIPS 2026 · similarity 0.89)
- REST: Efficient and Accelerated EEG Seizure Analysis through Residual State Updates (ICML 2024 · similarity 0.88)
- Seizure-Semiology-Suite($S^3$): A Clinically Multimodal Dataset, Benchmark, and Models for Seizure Semiology Understanding (ICML 2026 · similarity 0.88)
- EvoBrain: Dynamic Multi-Channel EEG Graph Modeling for Time-Evolving Brain Networks (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
