Systems Engineering (Data Engineering) student @ Wrocław University of Science and Technology
Exploring the intersection of computational neuroscience and bio-inspired ML
I'm a 3rd-year Systems Engineering student with a strong interest in Machine Learning, Computational Neuroscience, and Biomedical Analysis. Most of my work lives at the boundary between biology and deep learning, spanning astrocyte-modulated neural networks and clinical image analysis.
- Bio-inspired & Computational Neuroscience — astrocyte-neuron network dynamics, spiking neural networks, biologically plausible alternatives to attention
- Medical Image AI — volumetric tumor segmentation, histopathology classification
- Mechanistic Interpretability — studying learned representations in transformer embeddings and the grokking phenomenon
| Project | Description |
|---|---|
| Grokking | Studying the grokking phenomenon on algorithmic tasks (modular arithmetic, complex multiplication) across DOT, MHDOT, and a novel astrocyte-inspired linear-attention transformer (NAT) |
| biologically-plausible-transformer | Linear RNN inspired by astrocyte-neuron interactions, as a biologically grounded alternative to attention |
| Neuron-astrocyte-Network-Dynamics | Vectorized PyTorch implementation of tripartite synapse (neuron-astrocyte) network dynamics for RL |
| brats24-glioma-research | 3D multi-modal brain tumor segmentation (BraTS 2024) with PyTorch & MONAI — high-precision 3D U-Net |
| her2-classification | HER2 breast cancer status classification from histopathology — 96.6% accuracy, Grad-CAM interpretability |
| model-compression | Neural network quantization implemented from scratch in C++ (INT8 post-training quantization), verified against a Res8 keyword-spotting model trained in PyTorch |
Notes and derivations from aleczo.github.io:
| Article | Topic |
|---|---|
| Optimization Methods on an Energy Landscape | Comparing Gradient Descent, SGD, Momentum, RMSProp, and Adam on three synthetic non-convex landscapes, with interactive 3D and 2D simulations |
| Reprezentacje grup cyklicznych w sieciach neuronowych | How a two-layer perceptron learns to represent modular addition via cyclic group character theory |
| Transformer Architecture - Self-Attention | Analysis of self-attention mechanisms |
| Maximum Likelihood Estimation for Multivariate Gaussian | Step-by-step derivation of the MLE for mean vector and covariance matrix |
| Numerical Analysis of Dynamical Systems: The Euler Method | Formal derivation and error analysis of the Euler method |
Open to discussion on theoretical neuroinformatics, system dynamics, and machine learning. Reach out via email or LinkedIn.