Reference no: EM134037186 , Length: Word Count:2500
Question: "What exactly are you implementing?" We are adapting the authors' task/subject disentanglement framework to NinaPro DB2.
We retain the separate task-specific and subject-specific encoders and the reconstruction, task-triplet, subject-triplet, and cross-reconstruction objectives. Because DB2 has 12 sEMG channels rather than the 256-channel HD-sEMG spatial arrangement used by the original implementation, the input and encoder/decoder architecture need to be adapted while preserving the disentanglement objective. We will then evaluate gesture recognition from the task representation and, importantly, measure subject-identification leakage from that same representation. We will also perform the reciprocal evaluation on the subject representation and compare both against the original DB2 features. This implementation also relates to Artificial Intelligence methods for learning useful representations from complex data.
Objective: Can the method from your assigned paper separate gesture information from person/identity information when applied to the NinaPro DB2 dataset? The evaluation can also be considered alongside Data Mining approaches for extracting and evaluating patterns in datasets.
Paper - Understanding of Task-specific and Subject-specific Components in Surface EMG by Yangyang Yuan, Jionghui Liu, Xinyu Jiang, ChihHong Chou, Chenyun Dai, Jiahao Fan. The work is situated within Electrical Engineering, particularly in the context of signal-based systems and analysis.