1612 Worcester Rd, Apt. 204A, Framingham, MA 01702
I am an AI scientist / architect / developer building AI systems.
- Design of Computational Products & Services. Intelligent Automation
- Design of Computational Models & Architectures. Distributed Computation
- NLP / Deep NLP & Text Mining
- Machine Learning, Deep Learning & Reinforcement Learning
- Probabilistic Programming, Causal Inference & Decision Making
- Artificial Intelligence / Cognitive Computing
- NLP / Deep NLP & text mining: tagging (based on a trigram HMM), syntactic parsing (based on a PCFG), feature engineering and dimensionality reduction, multi-label classification, word sense disambiguation, Twitter hashtag decomposition, relevance engine, topic modeling; sentiment analysis, contextual text mining;
- Data mining & machine learning. Recommendation & personalization (based on realtime reinforcement learning) systems;
- Deep learning in TensorFlow & Keras;
- Reinforcement learning: TD Lambda, Policy Gradients, DQN, and A3C in TensorFlow;
- Artificial intelligence: artificial neural networks, rules-based models, fuzzy logic controllers, genetic algorithm, and ant-colony optimization;
- Probabilistic programming in Figaro / Scala & deep probabilistic programming in Edward / TensorFlow
- Adaptive programming: Formans' reflection techniques for Java and .NET, dynamic object model, adaptive user interface, adaptive design patterns, and Viable System Model (a cybernetic approach);
- Distributed programming in MapReduce / Java and Python (mrjob), Spark / Scala and Python, Akka / Scala, and TensorFlow / Python;
- Open source libraries & toolkits: mrjob, Apache Mahout, MLlib, Apache Solr / Lucene, WEKA's API, NumPy, SciPy, PyMC, Scala Breeze, Figaro / Scala, Edward, NLTK, spaCy, MALLET, Mulan, Gensim + word2vec, scikit-learn + nolearn, milk, SVM-Light / PySVMLight, Pycluster, etc;
- Platforms: scientific computing (Spark, Keras & TensorFlow, WEKA / MEKA, IPython / Jupyter, OpenAI Gym, NetLogo), cloud computing (Google App Engine, Amazon EC2), Windows, Linux (Fedora, Ubuntu).
AI Scientist & Mathematician / Consultant, Boston, MA 04.2005 - current
- Conceptual design of products and services based on scientific publications;
- Design of computational frameworks to formulate and solve real-world problems adequately by combining ideas and techniques from different computational disciplines;
- Design and customization of computational models and algorithms from scratch and from academic code / scientific papers.
Principal AI Architect & Developer / Consultant, Boston, MA 04.2005 - current
- Research, evaluation, and design of computational architectures and systems around scalability / linear scalability, parallelism, streaming, and real-time processing;
- Development of algorithms and algorithmic systems in C / C++, C#, Java, Scala, Python, Octave, and Julia.
Partial Project List:
- Nonparametric inference for simulated trajectories with noise (Scientific Systems Company, Inc.): Researched and developed two solutions: 1) a parametric approximation for LSTM models based on a variational autoencoder and probabilistic classifier (TensorFlow Probability) and
2) non-parametric one based on the Gutmann's likelihood-free inference of the simulator-based statistical models.
- Advertising (ADΣXT Corp.): 1) Analyzed the data, found additional constraints and restored the meaning of the data. 2) Modified and prototyped a TD(0) algorithm for advertising (Reinforcement Learning). 3) Designed and prototyped a test framework to replicate and analyze the almost random behavior of users and our computational models as a response to them (multi-agent systems, probabilistic programming).
- Automatically generated responses for chatbots (a SF-based start-up): Worked with a few academic projects on adversarial learning as a proof of concept. Customized the best one for the client’s data (Torch7 / Lua). Researched Bayesian Deep Learning (Edward and GPflow) for reducing the training datasets.
- Automatic agent-as-text generation for API.AI (Earley Information Science): Researched, prototyped, and wrote a proposal for the Ford's VA project.
- The mortgage loan (Boston Consulting Group): Designed a cognitive approach to the mortgage loan automation.
- Algorithms for a virtual assistant (noHold): Developed several NLU / ML algorithms, including the extraction of subject-verb-object triples from the dependency parse (spaCy) and semantic operations on the text. Researched the Word Mover’s Distance and unsupervised deep embedding for text clustering (MXNet).
- Online dating (Likeli): Researched lots of scientific publications and competitors, and designed a conceptual model for the platform. Designed a personality similarity metric between users and implemented its computation with the Theano / GPU library. Designed an agent-based computational framework based on Scala, Breeze, Akka, and Redis to scale computations for many simultaneous users in real-time. Prototyped computing personality profiles for FB users based on the Big Five Personality Traits Model; computed their similarities. Prototyped a real-time personalization system based on users’ feedbacks for several types of bandit algorithms modified for a vector space: Epsilon-Greedy, Softmax, UCB1, Bayesian, and Bayes-UCB1. Improved Bayes-UCB1 to balance multi-factor learning.
- Algorithms for marketing (MutualMind): Developed various algorithms such as word sense disambiguation, Twitter hashtag decomposition (used the Norvig's algorithm), topic modeling, sentiment analysis on Twitter, etc.
- A relevance engine (SocialExtract): Prototyped a library of NLP and ML primitives and a framework for quick composition of computational pipelines based on Python functional and streaming libraries. Implemented a relevance engine in this framework. After the client was satisfied with its accuracy, I re-designed and re-developed it in a distributed environment (Hadoop / MapReduce, MangoDB, and Redis) for scalable production with mrjob / Python and Java. Implemented a distributed classifier, the proximal SVM.
- NLP and ML tasks (IEEE): Researched novelty detection in research articles and Deep NLP as a framework for NLP problems. Prototyped the multi-label classification for large label sets (> 7000) in scikit-learn and Mulan / MEKA frameworks, and similarity metrics for research articles as a core for the recommender system.
- Intelligent real-time logistics (Weft): Researched, evaluated, and designed the conceptual model for the real-time logistics platform based on the DARPA's Cougaar architecture. Developed several heuristics based on ant-colony optimization for route optimization (Java).
- A courier management system (Deliverator): Researched and designed a solution based on ant-colony optimization for delivery scheduling and optimization, which led to a grant from a federal agency.
- A personalized cognitive assistant for mobile (Knowtle): Researched, evaluated, and designed a conceptual model based on the DARPA's CALO project.
- A social recommender system (PopCircle): Designed a composable approach to collaborative filtering. Used several social regularizations. Developed a first version in Octave.
- Computational marketing (FocusKPI): Designed a conceptual model for computational marketing based on a CF recommender.
- E-learning (Junyo): Designed a conceptual model similar to the Knewton's. Developed the concept map extraction from documents (Stanford NLP) and fuzzy matching based on partial textual data. Found patterns in the data for sales prediction (Weka).
Compuware Corp., Technology Department, Detroit, MI 03.2000 – 11.2004
Software Developer VII
Developed several advanced features for the QALoad...
- Discovering server scalability bottlenecks based on the performance counters. Used the Kruskal-Wallis test and modified Hodges-Lehmann estimators for statistical modeling;
- Pattern recognition for QALoad capture's files and 'restoration of the application logic' for c-scripts;
- SQL parser and the algorithm for dynamic variablization based on this parser.
Central Transport International, Inc., Sterling Heights, MI 03.1998 – 02.2000
- A the NT-based distributed enterprise architecture for 100+ terminals across the US, Canada, and Mexico using satellite communication;
- Two-way communications with the mainframe legacy system (DB2, CICS) based on SNA / COMTI;
- Multi-threaded asynchronous services for accessing ERP data on MS SQL Server from mainframe CICS clients (based on APPC communication with mainframe CICS) with the cluster support.
Resource Technologies, Troy, MI 08.1997 – 03.1998
Developed the scalable architecture for terminal operations based on MTS and MS SQL Server with up to 300 MS-DOS clients / hand-held computers using MS RPC. After several months of work, the client hired me as a Systems Architect to lead the technical reconstruction of his IT.
Advanced System & Designs, Inc., Troy, MI 01.1996 – 05.1997
Designed and developed the DoES for Windows - the primary product for the Shainin approach to Design of Experiment, and the ANOVA-TM 2.x for Windows - the primary product for the Taguchi approach to DoE.
DataNet Technologies, Inc., Troy, MI 02.1993 – 06.1994
Developed all statistics for the WinSPC including non-normal capability analysis for all types of Pearson's and Johnson's distributions.
Voronezh State University, the Department of Algebra and Topological Analysis Methods, Voronezh, Russia
Master of Science, Mathematics
Specialization: non-linear functional analysis. The master's thesis: diffeomorphisms in Banach spaces.
CONTINUING EDUCATION WITH CERTIFICATION:
Model Thinking, Coursera / University of Michigan - Ann Arbor, 2012
Introduction to Complexity, Santa Fe Institute, 2013
Introduction to Dynamical Systems and Chaos, Santa Fe Institute, 2014
Algorithmic Information Dynamics, Santa Fe Institute, 2018
Decision Making in a Complex & Uncertain World, FutureLearn / University of Groningen, 2014
Game Theory, Coursera / Stanford University, 2013
Big Data and Social Physics, edX / MIT, 2014
Data Mining with Weka. Parts 1 & 2, University of Waikato, 2014
Mining Massive Datasets, Coursera / Stanford University, 2014
Machine Learning, Coursera / Stanford University, 2012
Statistical Learning, Stanford University, 2014
Natural Language Processing, Coursera / Columbia University, 2013
Text Mining and Analytics, Coursera / University of Illinois at Urbana-Champaign, 2015
Algorithms: Design and Analysis. Parts 1 & 2, Coursera / Stanford University, 2013
Parallel Programming, Coursera / École Polytechnique Fédérale de Lausanne, 2017
CONTINUING EDUCATION WITHOUT CERTIFICATION:
Web Intelligence and Big Data, Coursera / Indian Institute of Technology - Delhi, 2012
Machine Learning: Reinforcement Learning, Udacity / Georgia Tech, 2014
Scalable Machine Learning (Spark / PySpark), edX / UC - Berkeley, 2015
Neural Networks for Machine Learning, Coursera / University of Toronto, 2015
Deep Learning for Natural Language Processing, Stanford University, 2015
Deep Natural Language Processing, University of Oxford & Google DeepMind, 2017
Deep Learning (TensorFlow), Udacity / Google, 2016
Probabilistic Graphical Models, Coursera / Stanford University, 2016
Bayesian Methods for Machine Learning, Coursera / National Research University Higher School of Economics, 2019
Artificial Intelligence. Part 1, edX / UC - Berkeley, 2014
Knowledge-Based AI: Cognitive Systems, Udacity / Georgia Tech, 2016
Approximation Algorithms. Parts 1 & 2, Coursera / École Normale Supérieure, 2016
Functional Programming Principles in Scala, Coursera / École Polytechnique Fédérale de Lausanne, 2015
Superhuman OS, Integral Institute, 2018
IEEE International Conference on Self-Adaptive and Self-Organizing Systems (SASO'07)
O'Reilly Artificial Intelligence Conference, 2016-2018
The International Conference on Probabilistic Programming, 2018
Cognitive Computing Consortium