Long-tailed image recognition
BLT: Balancing Long-Tailed Datasets with Adversarially-Perturbed Images. 
Research
One-Shot One-Class image recognition
CLEAR: Cumulative LEARning for one-shot one-class image recognition (CVPR 2018) 
One-shot / few-shot image recognition
Studying problem of decreasing the performance gap between one-shot and few-shot testing scenarios.
CNNs training analysis
Analyzing the impact of specific training examples on the develompent of neural network weights. Studying ways of ensuring better quality weight-learning and generalization abilitites.
Self-supervised dataset explainability
Utilizing external knowledge bases to find semantically-sound subcategories in an existing dataset in order to achieve higher decision-making explainability.