Mikayel is available for hire
Hire MikayelMikayel Samvelyan
Verified Expert in Engineering
Deep Learning Developer
London, United Kingdom
Toptal member since May 1, 2018
Bio
With a master's degree from the University of Oxford, Mikayel is a machine learning expert and data scientist specializing in natural language processing, computer vision, and reinforcement learning. He has substantial experience in research and development of deep learning solutions and has worked at such companies as Reddit, Mentor, and USC. Mikayel has co-authored several scientific publications at leading machine learning conferences.
Portfolio
USC Information Sciences Institute (via Toptal)
OpenCV, Python
Reddit (via Toptal)
Amazon SageMaker, Custom BERT, PyTorch, Python
Highlander Technology, Inc. (via Toptal)
Amazon SageMaker, PyTorch, Python
Experience
- Python - 6 years
- Data Science - 5 years
- Machine Learning - 5 years
- C++ - 5 years
- Deep Learning - 5 years
- Generative Pre-trained Transformers (GPT) - 4 years
- Natural Language Processing (NLP) - 4 years
- PyTorch - 3 years
Availability
Part-time
Preferred Environment
PyCharm, Vi
The most amazing...
...project I've worked on is developing a new AI algorithm that masters the game of StarCraft II.
Work Experience
Research Engineer (Computer Vision)
2020 - 2020
USC Information Sciences Institute (via Toptal)
- Built a scalable pipeline for a video prediction models that runs parallel on multiple remote GPU instances.
- Performed a pose estimation and projected 3D objects to their corresponding 2D projections in the videos.
- Developed a motion detection software using Python and OpenCV.
- Constructed a head pose estimation library using OpenCV and Dlib.
- Developed a real-time object detection library based on YOLO.
- The work has been presented to DARPA during the on-site visit to USC ISI.
Technologies: OpenCV, Python
Machine Learning Expert
2019 - 2019
Reddit (via Toptal)
- Developed deep-learning solutions for large-scale natural language processing tasks using PyTorch.
- Fine-tuned pre-trained NLP models, such as BERT, XLNet, and RoBERTa.
- Conducted topic modeling experiments using methods such as LDA, NMF, and so on.
- Built a modular pipeline for large-scale natural language classification tasks.
- Analyzed large-scale datasets using BigQuery.
Technologies: Amazon SageMaker, Custom BERT, PyTorch, Python
Data Scientist
2018 - 2019
Highlander Technology, Inc. (via Toptal)
- Developed and optimized deep learning solutions for large scale natural language processing tasks using PyTorch.
- Implemented noise and novelty detection on a large corpus of data.
- Performed rigorous data mining for raw and noisy data.
- Created an API endpoint to use the model for multiclass classification.
- Built a machine learning pipeline for large-scale multiclass classification.
Technologies: Amazon SageMaker, PyTorch, Python
Machine Learning Consultant
2018 - 2018
Mission Ready Marketing, LLC (via Toptal)
- Designed a pipeline for machine learning methods to be used in a recommendation engine.
- Examined the methods for building a recommendation system with unsupervised learning techniques for settings without historical data.
- Analyzed item-based, user-based, matrix factorization-based, and hybrid recommender systems.
- Created a full technical approach documentation for connecting the learning component of the project with the database and back-end.
Technologies: Python
Machine-learning Researcher
2018 - 2018
XIX.ai
- Searched and identified objects on satellite, drone, and ground-based images.
- Designed and implemented a deep reinforcement learning algorithm for large-scale fleet management.
- Created a simulator for a ride-hailing service using Python.
- Built a machine learning framework for predicting a user's intentions.
Technologies: Google Vision API, Sacred, PyTorch, Python
Research and Development Engineer
2014 - 2016
Mentor Graphics
- Conducted research about algorithms for logic optimization, partitioning, placement, and routing.
- Developed a unified environment for design capturing, simulation setup, verification, and analysis for a custom integrated circuit design platform.
- Integrated the environment with third-party IDE software.
- Created various kinds of reusable compiler-compilers using Lex and Yacc.
Technologies: Amazon Lex, Yacc, C++
Software Engineer
2012 - 2013
Instigate Robotics
- Built firmware and software for robotic applications.
- Developed 3D printing technologies.
- Created an application's graphical user interface (GUI) with C++ and Qt.
- Designed and implemented various embedded applications on STM32 and Arduino MCUs.
- Built an educational development environment for robotics.
Technologies: Qt, Smalltalk, C++, C
Software Engineering Intern
2012 - 2012
Instigate Design
- Developed a hard/software independent environment for parallel computing on multiprocessor computers.
- Supported the development of a compiler's front end.
Technologies: Clang, C++
Experience
QMIX: Monotonic Value Function Factorization for Deep Multi-agent Reinforcement Learning
http://proceedings.mlr.press/v80/rashid18a.htmlQMIX is a state-of-the-art value-based algorithm for collaborative deep multi-agent reinforcement learning for the settings of centralized training with decentralized execution.
QMIX employs a neural network that estimates joint action-values as a complex nonlinear combination of per-agent values that condition only on local observations. We structurally enforce that the joint-action value is monotonic in the per-agent values, which allows tractable maximization of the joint action-value in off-policy learning, and guarantees consistency between the centralized and decentralized policies.
QMIX was presented at ICML 2018, one of the leading conferences for machine learning research.
QMIX employs a neural network that estimates joint action-values as a complex nonlinear combination of per-agent values that condition only on local observations. We structurally enforce that the joint-action value is monotonic in the per-agent values, which allows tractable maximization of the joint action-value in off-policy learning, and guarantees consistency between the centralized and decentralized policies.
QMIX was presented at ICML 2018, one of the leading conferences for machine learning research.
PyMARL: Python Multi-agent Reinforcement Learning
https://github.com/oxwhirl/pymarlPyMARL is a research platform for deep multi-agent reinforcement learning which allows out-of-the-box experimentation and development.
Written in PyTorch, PyMARL features implementations of several state-of-the-art methods such as QMIX, COMA, and independent Q-Learning.
In collaboration with the Berkeley AI Research lab, some of the above algorithms have also been successfully ported to the scalable RLlib framework.
Written in PyTorch, PyMARL features implementations of several state-of-the-art methods such as QMIX, COMA, and independent Q-Learning.
In collaboration with the Berkeley AI Research lab, some of the above algorithms have also been successfully ported to the scalable RLlib framework.
SMAC: The StarCraft Multi-agent Challenge
https://github.com/oxwhirl/smacThe StarCraft Multi-agent Challenge (SMAC) is a benchmark for cooperative multi-agent reinforcement learning (MARL) that provides elements of partial observability, challenging dynamics, and high-dimensional observation spaces. SMAC is built using the StarCraft II game engine, creating a testbed for research in cooperative MARL where each game unit is an independent RL agent.
Paper: https://arxiv.org/abs/1902.04043
Blogpost: http://whirl.cs.ox.ac.uk/blog/smac/
Blogpost by NVIDIA: https://nvda.ws/2I88F1I
SMAC was presented at AAMAS 2019, the largest and most influential conference in the area of agents and multiagent systems.
Paper: https://arxiv.org/abs/1902.04043
Blogpost: http://whirl.cs.ox.ac.uk/blog/smac/
Blogpost by NVIDIA: https://nvda.ws/2I88F1I
SMAC was presented at AAMAS 2019, the largest and most influential conference in the area of agents and multiagent systems.
ADAM Visual Perception
https://github.com/isi-vista/adam-visual-perceptionOur learning process depends on having a fairly rich (though developmentally plausible) input representation. This repository explores how two aspects of visual perception which are vital for early language learning can be captured by algorithms, namely motion cause detection and gaze object detection.
Education
2016 - 2017
Master of Science Degree in Computer Science
University of Oxford - Oxford, UK
2014 - 2016
Master of Science Degree in Computer Science and Applied Mathematics
Yerevan State University - Yerevan, Armenia
2010 - 2014
Bachelor of Science Degree in Computer Science and Applied Mathematics
Yerevan State University - Yerevan, Armenia
2013 - 2013
Undergraduate Exchange in Computer Science
Delta State University - Cleveland, MS, USA
Certifications
AUGUST 2015 - PRESENT
CS190.1x: Scalable Machine Learning
The University of California, Berkeley via edX
MAY 2015 - PRESENT
Algorithms: Design and Analysis, Part 2
Stanford University via Coursera
APRIL 2015 - PRESENT
Algorithms: Design and Analysis, Part 1
Stanford University via Coursera
Skills
Libraries/APIs
SciPy, NumPy, Pandas, PyTorch, TensorFlow, Scikit-learn, Keras, OpenCV, Spark ML, PySpark, Google Vision API, Google APIs
Tools
Jupyter, LaTeX, Makefile, Vim Text Editor, Subversion (SVN), Git, PyCharm, CVS, MATLAB, Tmux, Tableau, Amazon Lex, Sacred, BigQuery, Amazon SageMaker, AutoML
Languages
C++, C#, Python, Java, Lua, SQL, R, C, Objective-C, Smalltalk, Assembler x86, Prolog, Lisp, JavaScript, Octave
Frameworks
Flask, .NET, Qt, Spark
Platforms
Unix, Docker, MacOS, Linux, Amazon EC2, Amazon Web Services (AWS)
Paradigms
Requirements Analysis, Design Patterns
Storage
MongoDB, PostgreSQL, Redshift
Other
System Programming, Deep Learning, Reinforcement Learning, Natural Language Processing (NLP), Custom BERT, Statistics, Classification, Predictive Modeling, Machine Learning, Mathematics, Data Visualization, Generative Pre-trained Transformers (GPT), Robotics, Computer Vision, Data Science, Recommendation Systems, Big Data, Image Processing, Physics, Vi, Yacc, Clang, Time Series Analysis
Collaboration That Works
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
1
Share your needs
Discuss your requirements and refine your scope in a call with a Toptal domain expert.
2
Choose your talent
Get a short list of expertly matched talent within 24 hours to review, interview, and choose from.
3
Start your risk-free talent trial
Work with your chosen talent on a trial basis for up to two weeks. Pay only if you decide to hire them.
Top talent is in high demand.
Start hiring