Winners of the DSSC XS package

• Achieving robustness to distribution shifts in instance segmentation of time-lapse microscopy: A brain-inspired inhibition approach (Dr. George Azzopardi - BI; Dr. Andreas Milias Argeitis - GBB)
  

The aim of this project is to develop instance segmentation algorithms for the delineation of budding yeast cells in brightfield time-lapse microscopy images. This is important for the automatic extraction of cellular readouts in high volumes of imaging data. The main scientific challenge is the high variability in brightfield imaging, especially across labs and imaging platforms (microfluidic chips, agarose pads, glass slides). The proposed methodology will integrate a brain-inspired inhibition mechanism in instance segmentation convolutional neural networks and evaluate them in terms of generalization ability on various real brightfield imaging data sets from international labs. More specifically, we will develop a computational model of the push-pull inhibition phenomenon, which is known to contribute to generalization in the mammalian brain, and integrate with state-of-the-art segmentation networks (e.g. Mask R-CNN).

• Visual analysis of dis/misinformation and hateful meme (Dr. Jiapan Guo - BI; Dr. Kun He - Faculty of Arts, Media Studies)