Mikhail Teverovskiy, Ph.D.

ad4evx@r.postjobfree.com

I am a senior-level scientist/researcher specializing in mathematical modeling, machine and deep learning, computer vision, and image/signal processing, conducting R&D in biometrics, medicine, and engineering with solid experience leading complex multidisciplinary projects. My research includes quality control for lithium-ion battery manufacturing, patterns, and anomaly detection, signal-noise separation, time series, genome mutation recognition on tissue images and cancer outcome predictions by computer vision and machine learning methods, tumor quantification on cancerous images, stochastic processes, iris recognition, iris counter-spoofing methods; computer generated synthetic biometrics: fingerprints, iris and face, and image compression algorithms. I hold a PhD in applied math; have published in peer-reviewed journals, presented at major scientific conferences, and authored US patents and applications.

SKILLS HIGHLIGHTS

Artificial Intelligence, Computer Vision, Machine Learning, Deep Learning, Pattern Recognition

Computations and Analytics, Statistical Signal Processing

Image and Data Analysis, Clustering, Spectral Analysis, Time Series

Supervised and Unsupervised Learning, Multivariate Statistics

Segmentation, Classification, Regression, PCA, Hypothesis Testing

Modeling Dynamical System using Differential Equations.

TOOLS: MATLAB, PYTHON, SCIKIT-LEARN, PANDAS, TENSOR-FLOW, KERAS, SIMPY, GIT, PyVista, Trimesh

PROFESSIONAL EXPERIENCE

Fraunhofer-USA CMA, Riverdale MD 11.2021 – Present

Senior Research Scientist at the Data Science and AI group

Fraunhofer USA, Inc. is an R&D organization working with industry, universities, and state and federal governments on contract research projects. I worked on and contributed to the following projects (machine learning in Python):

AI-based optical microassembly: a) mathematics for an automated parts recognition (3D and 2D- mesh models)

Artificial Intelligence: Intelligent Control Systems research - approaches and trends.

3D Printing: Machine Learning models to predict mechanical properties of printed parts.

Laser Technologies: Machine Learning and Computer Vision modeling of laser cutting and welding processes; AI-based real-time multi-sensor process quality control.

Predictive Maintenance: statistical models describing mechanical loads in real-time maintenance for bearings monitoring.

Proposal writing for R&D projects.

CrowdDoing Organization 03.2020 – Present

Pro Bono Chief Data Science Officer

CrowdDoing is a joint non-profit initiative of Reframe It and Match4Action Foundation that is focused on helping to mitigate the effects of wildfire perils through selection of the fire preventive solutions which minimize regional losses from wildfires.

Led a team of data scientists and college students to build risk prevention machine learning models that choose optimal fire preventive solutions based on local climate, terrain, and economic data.

Member of the strategic planning team of the organization. Other activities include writing grants, and presentations, and working with pro bono teams from leading universities.

ImageSignal R&D Consulting Inc., Stamford, CT 04.2018 – 10.2021

Chief Scientific Officer. My recent research projects are:

Genome mutation prediction on lung cancer histopathology H&E images.

Concept of an early detection and wildfire tracking system: infrastructure, model, and algorithms. Python simulation (SimPy) of wildfire ignition and the fire spread through a region.

Mask RCNN training for object detection and instance segmentation under limited computing resources.

Implemented unsupervised learning algorithm for image clustering: k-means clustering has been integrated into the autoencoder framework when several clusters are unknown.

Siemens Corp., Research Group, Princeton NJ 04.2021 – 09.2021

Senior Key Expert

Computer Vision in manufacturing (auto and beverage production): auto-parts image registration, an algorithm for orientation of beverage cans randomly placed on a moving conveyor belt.

AI-based quality control system for manufacturing of Lithium-Ion batteries (concept, US Patent Application).

Computer Control System to monitor the quality of lithium-ion battery manufacturing operations (research project in collaboration with a major university).

Solid-State Batteries, state-of-the-art: perspectives, research & development, manufacturing, and modeling.

Genentech, Digital Pathology Group 10.2019 – 03.2020

Software Specialist IV

Scientific Research: Deep Learning modeling and Image Analysis of super large biomedical images (H&E and ImmunoFluorescence - IF), algorithms for IF signal vs. noise separation, and complex image registration, designed and created a labeling tool for deep learning model training – MATLAB AI.

Trove Predictive Data Science, Buffalo, NY 10.2018 – 04.2019

Senior Data Scientist

Data Analytics for energy demand-response project (Python, Pattern Recognition, Statistics)

Built data analytics for behavioral customers’ segmentation based on their energy consumption patterns via aggregation and classification of the consumption time series.

Proof-of-concept: a method for forecasting energy demands using customers' temporal consumption patterns.

EyeLock Corp., New York NY 09.2013 – 04.2018

Senior Research Scientist

R&D projects: mathematical and computer modeling for core methods of iris recognition; technology solutions for identity management; data analysis and performance measures for biometric systems; counter-spoofing solutions; filed two US Patent Applications. My major accomplishments are:

Deep Learning: designed and trained Convolutional Neural Networks for iris image classification; challenges: training with a limited amount of computing resources and labeled data, hyper-parameter selection.

Signal-noise separation in complex stochastic processes: created a method extracting signal-carrying components from non-stationary time series; the method has been applied for iris recognition applications. ARIMA models and Stochastic Spectral Analysis are used to analyze iris texture intensity profiles; signal and noise components were identified. Real matching experiments confirmed the efficiency of the method.

Novel biometric system’s performance measures: Biometric-Signal-To-Noise-Ratio (BSNR) to characterize recognition performance on the entire working segment of the Detection-Error-Tradeoff curve – the BSNR value is analogous to the well-known DSP the Signal-To-Noise-Ratio.

Kalman filter statistical model to establish a sufficient and convenient set of biometric measures to minimize losses in financial transactions related to wrong user’s identities; to choose an appropriate measure for a successful transaction the model estimates the probability of a person’s identity based on a history of previous measures (e.g., iris recognition, fingerprints, geo-location patterns, etc.).

Iris counter-spoofing methods: supervised machine learning methods (statistical classifiers are trained and evaluated via cross-validation) are coupled with computer vision algorithms extracting data from iris images to successfully distinguish pictures acquired from the live irises vs. paper printed irises.

Placenta Analytics LLC, Larchmont NY 09.2012 – 06.2013

Senior Scientist

Quantitation of human placenta, extracting information that can be associated with the onset of children's autism:

Successfully applied the Fourier model to quantitate shapes of placentas presented on digital images and used spectral analysis of shape variations to find an optimal sampling of the placenta’s perimeter.

Created a probabilistic model of the placental vascular network. The model estimates spatial vessel density on the placenta’s surface. The network visualization and feature extraction were implemented in MATLAB.

Created algorithm for segmentation and quantitation of biomarker signals in the placental tissue digital images.

ImageSignal R&D Consulting Inc., White Plains NY 07.2010 – 02.2017

Chief Scientific Officer, contractual

ISRND is a scientific and engineering company that combines the design and integration of image/signal processing algorithms to life sciences, medical, and engineering applications:

MRI breast image analysis: tumor segmentation, quantitative study of texture features, and use fractal geometry of tumor boundaries to assess cancer aggressiveness - consulting contract with Memorial Sloan Kettering Cancer Center (New York, NY).

Survival of colon cancer patients: computer vision and machine learning algorithms for tumor segmentation, classification, and quantitation. The imaging features were combined with clinical records in a random forest model to predict the survival of colon cancer patients – a project with Cleveland Clinic (Cleveland OH).

Detection of authentic patterns on vein images: using a symmetric phase-only filter to detect authentic patterns on vein images that were acquired with a mobile phone camera.

IBG/Novetta Solutions, New York, NY 06.2011 – 09.2012

Senior Researcher and Developer, Manager of Computer Vision

Led development of a computer system capable of generating synthetic biometric images of fingerprints, irises, and faces which are a) model and approximate identities of real people; b) recognized by the existing biometric software; system's concept, R&D and supervised a team of engineers to implement the design:

Synthetic fingerprints: ridge patterns and noise mathematical models that produce realistically looking impostor fingerprints across all fingerprint classes and the NIST’s quality grades.

Synthetic irises: iris texture synthesis, anatomically truthful iris/pupil shape models, esthetic incorporation of a synthesized iris into a human eye; the modeling was done for cooperative and non-cooperative conditions.

Synthetic faces: computer and mathematical models of impostors, naturally looking human faces.

The project required both mathematical (using a mixture of probabilistic PCAs to generate parameters of face models, complex polynomials to compute ridge flows and singular points for digital fingerprints) and computer (use texture synthesis techniques for synthetic irises, various image analysis methods to simulate natural distortions on images) modeling. Machine learning and pattern recognition have been widely used to learn from large sets of data.

Definiens Inc., Parsippany, NJ 01.2009 – 01.2010

Senior Consultant, Life Science Unit

Developed custom imaging algorithms to quantitate images using Definiens' proprietary software: shape analysis; image segmentation; biomarker quantitation.

Aureon Laboratories Inc., Yonkers, NY 10.2002 – 01.2009

Manager, Machine Vision Department, and Senior Bio-Imaging Scientist

The company developed commercial tests that predict prostate cancer outcomes. The prediction models estimated the likelihood of cancer outcome (e.g., recurrence) using clinical and imaging (histopathological and multiplex molecular biomarker) data. The development was done in the framework of systems pathology. I was responsible for imaging components of the tests: machine vision algorithms and software development, technical documentation, and reports. My work also included R&D planning, inter-department collaboration, and scientific presentations/publications; major accomplishments are:

Led the development of machine vision technology, which was successfully implemented in cancer outcome prediction tests and pharmaceutical studies for drug response.

Statistical models and machine vision method to recognize and grade prostate tumors: supervised learning by color, texture, and morphometric features extracted from histological microscopic images.

Mathematical model for computer localization and quantification of biomarker signals in multispectral immunofluorescence images; created a method to discriminate signal from background.

Applied advanced mathematics (fractal geometry, wavelet decomposition, and minimum spanning trees) to quantitate digital tissue images, and segmentation of 3D nuclei objects from a stack of confocal images.

QuikCAT Inc., Cleveland, OH 01.2000 – 07.2002

Research Scientist, Image Compression Department

Image/data compression technology: real-time processing, image enhancement, compression (lossy & lossless) algorithms for general and medical (MRI, CT) images; C++ design and implementation, LEADTOOLS library.

Case Western Reserve University, Cleveland, OH 03.1998 – 10.2000

Research Scientist, Macromolecular Department

Modeled a nonlinear turbulent polymer flow through an extruder’s chamber by creating a spatial mesh emulating the 3D complex geometry of the device, using a super-computer for computations (FIDAP CFD simulation package).

EDUCATION

PhD - Computational Fluid & Gas Dynamics, Moscow State University, Moscow, Russia

MS - Applied Mathematics, Moscow Institute of Electronics and Mathematics, Moscow, Russia

Professional affiliation: senior member of IEEE.

Contact this candidate