PMC

Thresholding for warm start learning. Prior to beginning the recursive training portion of learning, a thresholding process is used to implement feature selection for a warm start to final learning. Here we show an example of deep learning with artificial neural networks to predict CBCL scores using regression. Plotting the R² metric against the number of features reveals that 14 features is optimal. A threshold value is then identified which constrains the available feature set for final recursive learning after the warm start to the 14 most important features identified during initial learning prior to the warm start.

Workflow response time on the cold start and the warm start of the original workflow, all functions fused workflow, and the workflow fused by our method (in real-word workflows).

Workflow response time on the cold start and the warm start of the original workflow, all functions fused workflow, and workflow fused by our method (in example workflows).

Time series of the slow-phase velocity (SPV) of caloric nystagmus to warm and cold external canal ear irrigation. Abscissa: time in seconds from the beginning of irrigation; Ordinates: SPV of the induced ocular nystagmus in degree per second; purple bar: duration of the ear irrigation (30 s). (A) Typical response of a control senior. The red vertical lines indicate the start and the end of the perception of rotation for warm irrigation. The blue vertical lines indicate the start and the end of the perception of rotation for warm irrigation. The perception appeared when the SPV reached a value for SPV close to 5°/s (start) and disappears when SPV decline to or below 10°/s (end). (B) Typical response of one of our senior complaining for postural instability. Note that despite the high SPV (exceeding 15°/s) to both warm and cold irrigations, the patient did not report any rotation perception.