Unsupervised Cyclic Siamese Networks Automating Cell Imagery Analysis

Novel neural network models that can handle complex tasks with fewer examples than before are being developed for a wide range of applications. In some fields, even the creation few labels is laborious task and impractical, especially data require more seconds to generate each label. biotechnological domain, cell cultivation experiments usually done by varying circumstances experiments, seldom in such way hand-labeled one experiment cannot be used others. this field, exact counts required analysis, modern standards, semi-supervised typically need hundreds achieve acceptable accuracy on task, while classical image processing yields unsatisfactory results. We research whether an unsupervised learning scheme able accomplish without manual labeling given data. present VAE-based Siamese architecture expanded cyclic fashion allow use labeled synthetic particular, we focus generating pseudo-natural images from which target variable known mimic existence natural show provides reliable estimates multiple microscopy technologies unseen sets labeling. provide source code as well use. The package open free (MIT licensed).