- Kaggle: https://www.kaggle.com/strangerone
- LeetCode: https://leetcode.com/TheStrangerOne/
- e-mail: [email protected]
- Telegram: @TheStrangerOne
Fields of Interests: Classical Machine Learning and Deep Learning especially in the CV domain. Data Analysis and its practical application for decision making. Reinforcement Learning for complicated technical systems.
Open for the colloboration
A model was developed that determined by names, attributes and pictures whether two products were the same. This solution will help Ozon customers to improve the user experience, and companies to optimize resources and save on the purchase of server equipment.
Was involved in training final gradient boosting models, logging experiments in mlflow, building submission pipelines in a docker container and was helping in other ad-hoc tasks. Presented the solution to the case holder at pitching.
As part of a hackathon with the GNU MISIS team, a system was developed that predicts which segment (Segment_num) a given creative (Advertisement ID) belongs to.
I was involved in classifying video segments by their embeddings obtained using the XCLIP network, as well as developing theories and concepts for the generative neural network LLaVa. Presented the solution to the case holder at pitching.
As part of the hackathon with the "MISIS and Mr. Smith" team was presenting the idea of a project to use models of varying sensitivity to be able to flexibly adjust the ratio of expended resources and potentially retained clients.
I built a CatBoostClassifier model, the hyperparameters of which were selected using the Optuna library, thanks to which it was possible to achieve the value of the ROC_AUC metric = 0.77+, which allowed us to take 5-th place on the public and 8-th place on the private leaderboard according to the results of the hackathon.
A set of Jupyter notebooks for training various computer vision models to classify and recognize dice of different configurations and face values. Full support of the project was carried out, starting from setting the technical specifications, to comparing the results obtained from different models, validating the results and drawing conclusions.