Large-scale spatial clustering
Algorithms designed to cluster very large spatial datasets efficiently (near-linear behavior in practice), supporting robust behavior across different point distributions.
Research
My work focuses on building efficient intelligent systems by combining algorithms, hardware, and real-world data. This page gives a curated overview; detailed project reports and code are linked in the Archive.
Sparse Intelligent Systems
Many real problems become infeasible when complexity grows. A recurring theme in my work is to represent and solve large problems using sparsity, hierarchy, and validation-driven reduction.
Validation-driven reduction
Step-by-step reduction of complex systems with continuous validation, to preserve correctness while simplifying representations.
Sparse connectivity in neural memory
Alternatives to fully-connected associative memory using optimized partial connectivity to scale to larger sizes.
Efficiency as a design principle
Designs focused on minimizing compute, memory, and power while maintaining scientific usefulness and repeatable performance.
Neural Network Architectures
My neural network work spans foundational models, learning rules, and tool-building—aimed at systems that are interpretable, scalable, and hardware-friendly.
Stochastic neuron model
Probabilistic neuron behavior inspired by biological firing, along with training rules and comparisons to classical activation functions.
Partially connected feedback networks
Sparse associative memory networks as a practical alternative to fully connected Hopfield-style models.
Neural toolboxes
Tooling for building and experimenting with neural networks and pattern recognition systems—built for research productivity and reproducibility.
Hierarchical / modular neural systems
Architectures where pre-trained networks operate as computational units within a larger adaptive system (a “net-of-nets” concept).
Pruning and compression
Methods to reduce network size while preserving function—supporting efficient deployment and analysis.
Hardware-aware learning
Neural designs and learning rules that consider practical implementation constraints from the beginning.
Scientific Instrumentation
Building instruments and measurement systems that extract useful information from signals—combining electronics, signal processing, and AI.
3D imaging microscopy
Reflective light microscope workflow for 3D reconstruction using focus stacking and intelligent focus/defocus detection.
Radiation signal enhancement
Improving counting efficiency in noisy conditions using noise-signature discrimination and anti-coincidence strategies.
Space payload electronics
Embedded modules and FPGA-oriented designs for scientific payload applications with strict constraints.
Sensor systems
Low-power sensing, data logging, and analysis pipelines to turn raw signals into actionable knowledge.
Bioinformatics & Data Mining
Hybrid knowledge + neural approaches for extracting patterns from biological sequences and large biological databases.
Protein secondary structure prediction
Combining knowledge-base properties and neural networks for improved prediction and analysis.
Proteomic data mining
Tools and workflows to analyze and organize large biological datasets for research tasks.
Database utilities
Practical utilities for managing, extracting, and interpreting biological datasets and sequence properties.
Planetary Exploration Tools
Planetary science software and analysis tools for terrain visualization, crater-related workflows, and astro-material analysis.
Crater simulation & detection
Modeling and analysis tools supporting crater identification and classification tasks.
3D terrain browsers
Visualization systems for exploring lunar and Martian surfaces, including stereo/anaglyph style browsing.
Astro-material analysis
Classification and analysis workflows for planetary material and rock-type datasets.
Web-based exploration tools
Tools intended to make planetary surface browsing more accessible for research and education.
Embedded & Distributed AI
Designing intelligent systems close to sensors—where power, memory, latency, and reliability matter as much as accuracy.
Low-power analytics
RMS measurement, filtering, and compression methods suitable for microcontrollers and long-running deployments.
Sensor networks
Distributed measurement and data fusion ideas that convert local observations into global models.
Distributed compute concepts
Modular compute nodes and architecture thinking for scalable, affordable AI experimentation.