Quantitative Data Scientist

Описание

Dmitrii Masnyi
Email :dmasny@yandex.ru
Github:dmasny99 Mobile : +7 977 878 12 28
Machine Learning Engineer/Data Science Researcher Linkedin :Dmitrii Masnyi Telegram:@dmasny
Skills •
Programming languages : Python, Bash
• Technologies : PyTorch, Docker, K8s, Git, TensorRT, AirFlow, SQL, L A
T
E X,
• Knowledge : Deep Learning (mostly ASR and CV, NLP), Classical ML, Calculus, Probability Theory, Linear Algebra
• Languages : Russian (native), English (C1)
Experience •
VK group
Moscow, Russia
Deep Learning Engineer, AI VK Feb. 2023 - PresentPython Docker ASR Computer Vision Pytorch Bash K8s
◦
Significantly contributed to ASR for VK Video Subtitles (core technology): improved the quality of subtitles: private bеnсhmаrk 1
(PB1) from 12.3 WER to 5.5 WER (55% relative improvement) ,PB2 from 9.7 WER to 6.7 WER (31% relative
improvement) ,PB3 from 11.3 WER to 8.5 WER (25% relative improvement) ,PB4 from 10.3 WER to 7.7 WER
(25% relative improvement) . This model is currently being prepared for integration into all video platforms in VK (VK Videos,
Odnoklassniki, Dzen).
◦ I developed a fully automated service in Kubeflow Pipelines for CLIP model evaluation using FAISS. It allowed for the computation
of T@I and I@T retrieval scores for a 12.5 million objects indexin no longer than15 minutes each . Additionally, index
sharding and half precision enabled the computation on a machine with 8 1080ti GPUs, utilizing only 40% of its VRAM.
Thus, the index size could be easily doubled without any performance degradation.
◦ I conducted research dedicated to improving search quality for short videos in VK Videos. Specifically, I proposed adding an audio
modality (subtitles, to be specific) to the CLIP model. This approach improved HitRate@12 from 0.0154 to 0.0535on the
private bеnсhmаrk.
◦ Participated in the improvement of the internal crowd-sourcing service (SQ). Specifically, I proposed an enhancement to the
mechanism for assessing the quality of assessors, which is actively being used now.
• Central University
Moscow, Russia
Teaching Assistant, Machine Learning CourseMar. 2024 - Present Teaching Mentoring Machine Learning
I am working as TA on Machine Learning course organized by Alexander Dyakonov (Head of AI dept.) atCentral University. Namely
my work consists of:
◦ Refactoring/preparing practical materials fort he lessons. Conducting practical coding lessons for 1-st year master students,
checking their homeworks.
• Skolkovo Institute of Science and Technology (Skoltech)
Skolkovo, Russia
Data Science Researcher, Neuroimaging groupOct. 2021 - Feb. 2023 Python Docker Computer Vision Machine Learning DSP Time Series
Worked in Evgeny Burnaev’s Applied AI Center in Neuroimaging group led by prof. Alexander Bernstein and assoc. prof. Maxim
Sharaev. Namely contributed to these projects:
◦ AE with deformable convolutions for latent representation of EEG spectrograms in classification tasks. Published ICMV 2022
DOI:10.1117/12.2679806.
◦ Graph machine learning for multivariate time series classification: developed an end-to-end solution for the classification of
multivariate time series based on graph ML and proposed a node ”attention” mechanism that improved the model’s interpretability.
Namely, this mechanism assigned higher weights to the brain regions responsible for the particular disease (MDD) in a completely
learnable fashion.
◦ Joint project with IHNA RAS: I was developing embedders for multivariate time-series (NDA).
Education •
Skolkovo Institute of Science and Technology
Skolkovo, Russia
Master’s Degree in Information Science and Technology (IoT and Wireless Technologies track) GPA: 5.0Sept. 2021 - June 2023Machine Learning Deep Learning Information and Coding Theory Digital Signal Processing
•
VK Big Data Academy (ex. MADE)
Moscow, Russia
Data science track GPA: 4.93 Sept. 2022 - July 2023Algorithms and Data Structures Machine Learning Deep Learning Optimization for Deep Learning MLOps
•
Bauman Moscow State Technical University
Moscow, Russia
Bachelor of Electrical Engineering (Industrial Automation track) GPA: 4.89Sept. 2017 - Aug. 2021Calculus Linear Algebra Probability Theory Applied Mechanics Algorithms and Data Structures