Code & applications

Projects

Independent applications, tools, and resources from my GitHub repositories. Code accompanying papers is linked from Publications.

View my GitHub

5 projects

Research resources

Evolutionary multiobjective neural architecture search

Paper collection · Benchmarks · Taxonomy · Updated 2026-10-03

A curated research collection covering evolutionary methods for discovering neural architectures that balance predictive performance with latency, energy, memory, and model complexity. It connects research papers with code, benchmarks, and deployment considerations.

GitHub Documentation

Project details

The collection organizes the field around search spaces, search strategies, objective functions, evaluation strategies, and techniques for accelerating search. It also covers hardware-aware design, robustness, emerging applications, and reproducible evaluation.

Applied machine learning

Telecom bandwidth forecasting and anomaly detection

Python · LSTM · Jupyter · Updated 2026-08-27

Time-series models for forecasting data-centre traffic and identifying unusual bandwidth patterns. LSTM forecasters use recent traffic history to predict future samples, while an LSTM autoencoder flags windows with unusually high reconstruction error.

GitHub Documentation

Project details

Separate workflows cover univariate forecasting, multivariate forecasting, and anomaly detection. Shared preprocessing uses chronological train-test splits and training-only scaling. Forecasts are evaluated against a persistence baseline with MAE, RMSE, and R².

Applied machine learning

Telecom alarm classification

Python · BERT · TextCNN · Updated 2026-08-27

A language-model pipeline for organizing short Chinese telecom alarm reports into meaningful categories. It first discovers groups through sentence embeddings and clustering, then trains a BERT classifier to assign new alarms to those groups.

GitHub Documentation

Project details

The project supports multiple category counts, includes a character-level TextCNN baseline, and provides notebooks and command-line tools for clustering, training, and prediction. Evaluation includes per-class reports, confusion matrices, and a majority-class baseline.

Computer vision

5G-V2X roadside perception system

Python · PyTorch · ONNX · Updated 2026-08-27

A roadside perception pipeline that detects and tracks road users, maps detections into road coordinates, estimates collision risk, and shares scene information with nearby vehicles. It brings vision, tracking, and communication together in an edge system.

GitHub Documentation

Project details

The repository includes object detection, semantic segmentation, Kalman-filter tracking, camera calibration, optional camera-radar fusion, and versioned messages over UDP. A virtual onboard unit supports local testing; physical C-V2X communication requires a vendor-specific transport integration.

Probabilistic modeling

Visualizing Gaussian process regression

Python · scikit-learn · Jupyter · Updated 2026-08-23

A notebook exploration of how covariance kernels shape Gaussian process predictions on weather observations. Visualizations compare predicted temperatures with measurements and show the uncertainty associated with each prediction.

GitHub Documentation

Project details

The notebook defines Rational Quadratic, periodic, Matérn, RBF, and combined kernels. It uses scikit-learn’s GaussianProcessRegressor to estimate predictive means and standard deviations, alongside regression error metrics. Its random holdout evaluates interpolation rather than a chronological forecasting task.