Work
Time-series forecasting, deep learning, and the data work that has to happen before either is useful.
- Forecasting S&P 500 volatility with deep learning MSc dissertation. LSTM variants benchmarked against ARIMA and GARCH, plus a hybrid of the two. The hybrid lost to the standalone LSTM — and why it lost turned out to be the finding.
- Wildfire detection with CNNs — literature review Conventional convolutional networks compared against CNNs using Learning without Forgetting, for continual wildfire detection that does not need full retraining on every new region.