Trifon
Triantafillidis
I'm a software engineer, researcher and inventor. My background spans business leadership, custom software development, machine learning, computer vision, linear algebra and mathematical optimization.
Engineering with a research mindset.
I combine long-term software development experience with a strong mathematical background, particularly in linear algebra and efficient algorithms. My technical work has included machine learning, neural networks, convolutional neural networks, computer vision, pattern recognition, face detection, optimization and business applications. Before turning to software, I spent years in executive and entrepreneurial roles in manufacturing and wholesale trade.
Professional path
My working life divides into two halves: two decades running businesses, then two decades writing software.
Lead Software Engineer — ISON Psychometrica
I design and build the company's psychometric testing platform in C#, from architecture through implementation, and guide the other developers working on it.
What began as a handful of modules is now a suite of web and desktop applications, together with the smaller tools used in day-to-day psychometric work.
External Partner — ISON Psychometrica
Working as an external collaborator, I developed the first modules of the psychometric platform.
Developer — Bluebit Software
Development of linear algebra software components, principally the Bluebit .NET Matrix Library.
Freelance Developer
Custom applications adapted to business needs and complementary to mainstream commercial packages.
Shareholder & CEO — Trimar
Imports and wholesale of household goods in Ionia, Thessaloniki.
Shareholder & CEO — Makedoniki Keramiki
Manufacturer of decorative and utilitarian ceramics in Thessaloniki.
Technical strengths
Programming, mathematics and algorithmic problem solving run through everything I do.
Programming
C++ — more than 10 years of experience, including native C++ combined with .NET managed code.
C# / .NET — extensive application-development experience using .NET libraries.
SQL — extensive practical experience.
Python, Java, HTML — working knowledge and development experience.
Applied Computing
Machine learning, neural networks, convolutional neural networks, real-time computer vision, pattern recognition and face detection.
Business software, asynchronous client/server updating and order-shipping optimization.
Mathematics
Strong mathematical background with particular emphasis on linear algebra, linear programming and minimax problems.
Algorithms
I focus on efficient algorithms for complex problems. My work on face-detection techniques in 2008 led to a new minimax-equilibrium / linear-programming solution method that I later patented in the United States.
Research and inventive work
The work I'm most attached to — patents, research and the projects behind them.
Bluebit .NET Matrix Library
My longest-running project: a commercial object-oriented linear algebra library for the .NET platform, developed under Bluebit Software. It solves systems of linear equations, least-squares problems, eigenvalue and singular-value problems, and provides the associated Eigen, LU, LQ, QR, Cholesky and SVD factorizations for both real and complex matrices.
I built it as a mixed-mode C++ project — native code and .NET managed code in one assembly — over optimized BLAS and LAPACK routines, to get native speed behind a managed API. It shipped through version 7.0.1 and is distributed on NuGet for x86 and x64. An earlier generation of the same work shipped as the Matrix ActiveX Component.
NuGet: x64 ↗ NuGet: x86 ↗Method for Solving Minimax and Linear Programming Problems
While working on face-detection techniques I found a new algorithm for solving minimax problems, particularly effective where the corresponding matrix is large, together with a technique for recasting linear programming problems as minimax problems so that the same method solves them. It converges at a rate of O(1/T), with O(n²) work per iteration. The patent text names it the Bluebit algorithm.
Filed January 2008 with a November 2007 priority date, granted as U.S. Patent 7,991,713 B2 in August 2011, followed by continuation 8,200,590 B2 in June 2012. Both are held individually rather than assigned to a company.
US7991713B2 ↗ US8200590B2 ↗Deep-learning filters for digital images
My academic work investigates denoising, super-resolution and face-image pseudo-coloring using deep-learning techniques, including a GAN-based approach.
Academic paper ↗Algorithms in C++ and C#
Alongside my commercial work I publish smaller projects: gealib, a fast parallel genetic algorithm library in C++; Annotator, an image-annotation tool in C#; a GAN-based image colorizer; and an A*-based puzzle solver.
GitHub profile ↗Academic background
Hellenic Open University
Computer Science.
Aristotle University of Thessaloniki
School of Geology.