EVENT DOSSIER
Hugging Face released the paper related to the 2026 PNPL (Non-Invasive Speech Decoding) competition on September 7, 2026. The competition aims to improve word classification performance and achieve efficient generalization across subjects, consisting of two complementary parts: the depth challenge uses data from approximately 80 hours per subject for optimization of decoding; the breadth challenge involves progressively reduced subject-specific data and generalization tests with all-zero samples. The competition is open worldwide, with a deadline of October 15, 2026, and a prize of $5,000. The organizers have made the LibriBrain100 dataset, reference models, PyTorch data loader (`pip install pnpl`), Colab tutorials, and real-time leaderboards available, allowing participants without a background in neuroscience to participate.
Hugging Face 发布 2026 PNPL 非侵入式语音解码竞赛论文，聚焦词分类与跨被试高效泛化。该竞赛包含两个互补挑战：深度挑战利用单被试约 80 小时数据提升解码性能；广度挑战实现跨被试泛化，涵盖逐步减少的被试特定数据及全零样本被试。竞赛面向所有人开放，截止日期为 10 月 15 日，奖金为 5,000 美元。主办方已发布 LibriBrain100 数据集、参考模型、PyTorch 数据加载器（pip install pnpl）、Colab 教程及实时排行榜，无需神经科学背景即可参与。