CLAIRE lab- Currently working on understanding the collapse of PPO. Implementing modifications (LayerNorm, InFeR and others) to make a comprehensive study of their effect on PPO training- Reproduced MuJoCo environment results with a different codebase for the No Representation, No Trust paper rebuttals (accepted to NeurIPS 2024)

Ad Quality department- Increased the revenue of dynamic ads by adding categorical features, tuning the model for the preliminary stage of banner selection and configuring the pipeline of regular model retraining- Worked on fast banner text generation with templates, which could be used to handle hundreds of millions of banners. Improved correctness of banner texts (pointwise score) by adding extracted templates from texts using YaLM LLM

International Laboratory of Algebraic Topology and Its Applications- Applied topological data analysis methods for Alzheimer’s diseaseclassification by MRI- Studied the behaviour of number of endpoints of a random walkon graphs, as a result published [1] (see the Publications section)

Backend of Yandex’s smart home