Researching problem of identifying Named Entities using few-shot and zero-shot approaches for Multilingual setting in less-researched Slavic languages with varying character sets (eg. Polish, Czech, Ukrainian) and recognizing coreference relations between them. Gathering final solutions from other team members and preparing final submission.
Co-authoring PolEval competition task for punctuation restoration including data acquisition, annotation and design of evaluation metrics. Contributing to the development of an evaluation platform. Evaluating contestants’ solutions and source codes.
Development of a hybrid ASR improvement system consisting of 2 main parts: lattice extender and transformer-based utterance rescorer. Researching new, more efficient, approaches for Language Modelling eg. Electra. Coordinating meetings, data transfer protocols and preparing final submission to the task.
Researching problem of identifying Named Entities using few-shot and zero-shot approaches for Multilingual setting in less-researched Slavic languages with varying character sets (eg. Polish, Czech, Ukrainian) and recognizing coreference relations between them.
Leading a team on Nested Named Entity Recognition task for Polish, proposing and researching SoTA methods for the problem, gathering results from the team and preparing final submission.
Conducting research and developing a system for identifying phrases in audio recordings.
Developing a Named Entity Recognition and Coreference Resolution Toolkit
Inforex is a web system for text corpora construction. Inforex allows parallel access and sharing resources among many users. The system assists semantic annotation of texts on several levels, such as marking text references, creating new references, or marking word senses.
Automatic tracking of cells in time-lapse microscopy is required to investigate a multitude of biological questions. To limit manipulations during cell line preparation and phototoxicity during imaging, brightfield imaging is often considered. Since the segmentation and tracking of cells in brightfield images is considered to be a difficult and complex task, a number of software solutions have been already developed. CellStar is one of such algorithms. It is optimized to segment and track images of budding yeast cells growing in monolayer (e.g. images from microfluidic chambers), however the algorithm can be also used to track other round objects (in brightfield as well as fluorescent images).
