Tomáš Chobola

MetaDL Challenge 2020 focused on image classification tasks in few-shot settings. This paper describes second best submission in the competition. Our meta learning approach modifies the distribution of classes in a latent space produced by a backbone network for each class in order to better follow the Gaussian distribution. After this operation which we call Latent Space Transform algorithm, centers of classes are further aligned in an iterative fashion of the Expectation Maximisation algorithm to utilize information in unlabeled data that are often provided on top of few labelled instances. For this task, we utilize optimal transport mapping using the Sinkhorn algorithm. Our experiments show that this approach outperforms previous works as well as other variants of the algorithm, using K-Nearest Neighbour algorithm, Gaussian Mixture Models, etc.

Bioluminescence is an attractive alternative to fluorescence for live-cell imaging; however, its low intensity has prevented widespread adoption. Specialized microscopes compensate by sacrificing spatial resolution, field of view and dynamic range-constraints imposed by the highest-sensitivity camera to date: the electron-multiplying charge-coupled device. Recently, quanta image sensor (QIS) technology has emerged for low-light imaging. Here, we show that a commercial QIS camera has exceptional sensitivity; however, its sensor dimensions necessitate a microscope designed to maximize its properties. We introduce a Keplerian-telescope-inspired microscope setup that, with the QIS, results in modestly improved signal-to-noise ratios at substantially higher spatial resolution, field of view and dynamic range, relative to the state of the art. The telescopic design also confers modularity, enabling multimodal imaging with epifluorescence. The 'QIScope' makes bioluminescence a viable tool for technically challenging live-cell experiments such as monitoring intracellular and extracellular vesicles simultaneously and the dynamics of low-abundance proteins.