There are several real-world situations where obtaining clean training data is difficult. For instance, low-light photography – astronomical imaging, for example – physically-based image synhesis and magnetic resonance imaging are such cases.
Aalto University and FCAI professor Jaakko Lehtinen with his team from NVIDIA and MIT postdoctoral researcher Miika Aittala show in their paper accepted to the International Conference on Machine Learning ICML 2018 that it is possible to recover signals under complex corruptions without observing clean signals, at performance levels equal or close to using clean target data.
They have applied basic statistical reasoning to signal reconstruction by machine learning — learning to map corrupted observations to clean signals — with a simple and powerful conclusion: under certain common circumstances, it is possible to learn to restore signals without ever observing clean ones, at performance close or equal to training using clean exemplars.
The team applies their methods to photographic noise removal, denoising of synthetic Monte Carlo images, and reconstruction of MRI scans from under-sampled inputs. All cases are based on only observing corrupted data.