/ConformalPrediction

Conformal prediction suite for acoustic estimation tasks

Primary LanguagePythonMIT LicenseMIT

ConformalPrediction

DOI

Description: This repo contains starter code for the application of conformal prediction to acoustic estimation tasks (specifically DOA estimation).

Data simulation: Plane wave model is used to generate acoustic signals alongwith environmental sources of uncertainty.

Methods: At present 3 deep learning techniques are trained and tested:

  1. Deep ensemble
  2. Deep quantile regression
  3. DNN Monte-Carlo dropout

To be extended

Metrics: Conformal prediction is validated with coverage score and interval width.

Note: Works pertaining to this code have been published in the Journal of Acoustical Society of America (JASA) and IEEE Workshop on Machine Learning and Signal Processing (MLSP 2023). Public access links coming soon!