learning through writinghttps://shaijm.com/https://shaijm.com/posts/autocorrelation-through-simulation/https://shaijm.com/posts/autocorrelation-through-simulation/Using simulated processes, lag plots and the autocorrelation function to distinguish dependence from spurious patterns.Fri, 23 Oct 2026 00:00:00 GMThttps://shaijm.com/posts/time-series-decomposition-stl/https://shaijm.com/posts/time-series-decomposition-stl/Trend, seasonality and remainder components, and why robust STL is often more flexible than classical decomposition.Thu, 22 Oct 2026 00:00:00 GMThttps://shaijm.com/posts/why-adam-works/https://shaijm.com/posts/why-adam-works/Deriving momentum, RMSprop and Adam, with an interactive comparison on an ill-conditioned objective.Tue, 29 Sep 2026 00:00:00 GMThttps://shaijm.com/posts/why-scale-attention/https://shaijm.com/posts/why-scale-attention/A variance argument for scaled dot-product attention, with an interactive softmax experiment.Mon, 12 Oct 2026 00:00:00 GMThttps://shaijm.com/posts/word2vec-negative-sampling/https://shaijm.com/posts/word2vec-negative-sampling/How predicting contexts produces word embeddings, and why negative sampling changes the learning problem.Tue, 15 Sep 2026 00:00:00 GMT