Bohdan Hirianskyi

In August 2026 I defended a PhD in Computer Science at Igor Sikorsky Kyiv Polytechnic Institute. The dissertation studies how evolutionary algorithms optimize neural networks: four methods that prune and compress networks, tune hyperparameters, and merge models. Author of 5 research publications, including a paper in a Scopus-indexed journal. From September 2025 to June 2026 I worked as a junior software developer. I taught a university course on algorithm design. I am looking for applied, product-focused ML work.

PhD dissertation (122 Computer Science): “Optimization of Parameters and Structure of Neural Networks Using Evolutionary Algorithms”. Defended on 28 August 2026 before one-time specialized academic council DF 26.002.419 at Igor Sikorsky KPI; unanimous vote.

Code and experiment results

1 Scopus-indexed paper, 3 papers in Ukrainian peer-reviewed journals, 1 conference abstract.

1. Effectiveness of evolutionary algorithms in neural-network optimization (2026)

   Compared evolutionary (CMA-ES), gradient-based and Bayesian methods on unconstrained minimization, structured pruning and hyperparameter optimization. CMA-ES + SynFlow with energy compensation gave the best quality under extreme pruning (98%, F1 = 0.81). Reported the limits honestly: Adam trains weights better, and magnitude pruning wins at moderate sparsity.

   Hirianskyi, B., & Bulakh, B. (2026). Effectiveness of evolutionary algorithms in neural network optimization tasks: Unconditional minimization, pruning, and hyperparameter optimization. Control, Navigation and Communication Systems, 1(83), 45–50. https://doi.org/10.26906/sunz.2026.1.045

2. Neuroevolutionary synthesis of neural networks (2025, in Ukrainian)

   Implemented two neuroevolution algorithms from scratch, memetic and hybrid, that automatically select network architecture and weights. On non-linear binary-classification tasks (XOR, Moons, Breast Cancer) they outperformed a fixed-architecture perceptron. Python, NumPy, scikit-learn.

   Hirianskyi, B., & Bulakh, B. (2025). [Creating neural networks through neuroevolution and automated learning using evolutionary algorithms]. Science and Technology Today, 8(49), 1214–1227. https://doi.org/10.52058/2786-6025-2025-8(49)-1214-1227

3. Evolutionary optimization of neural networks (2024)

   Applied genetic algorithms (DEAP, sklearn-genetic) to optimize the architecture, hyperparameters and weights of a multilayer perceptron on the Wine Quality dataset. Reported honestly that the evolutionary approach did not, on average, beat the baseline in quality and required far more computation.

   Hirianskyi, B., & Bulakh, B. (2024). Using evolutionary algorithms for training and optimizing neural networks. Science and Technology Today, 8(36), 808–823. https://doi.org/10.52058/2786-6025-2024-8(36)-808-823

4. A review of evolutionary algorithms for neural-network synthesis and training (2023, Scopus-indexed)

   Systematized how genetic algorithms are applied to neural networks: feature selection, weight optimization, changes to connection structure, and optimizer tuning.

   Hirianskyi, B., & Bulakh, B. (2023). A review of practice of using evolutionary algorithms for neural network synthesis and training. Technology Audit and Production Reserves, 4(2(72)), 22–26. https://doi.org/10.15587/2706-5448.2023.286278

5. International conference abstract (Athens, Greece, 24–26 December 2025, in Ukrainian)

   Hirianskyi, B. (2025). [Using evolutionary algorithms for neural network synthesis and training: State of the art and prospects]. Science and Information Technologies in the Modern World, 165–168. https://doi.org/10.70286/isu-24.12.2025.006

The applications were built together with the Claude Code agent.

Igor Sikorsky Kyiv Polytechnic Institute · Institute for Applied System Analysis (IASA) · Department of System Design

• Ukrainian — native
• English — B2