Muhammad Jahanzaib Awan — Writing https://jahanzaibawan.com/blog.html Short technical posts on machine learning, evaluation and data science. en-gb Sun, 05 Jul 2026 09:00:00 +0000 Attention, Intuitively: Queries, Keys, and Values https://jahanzaibawan.com/blog/attention-queries-keys-values.html https://jahanzaibawan.com/blog/attention-queries-keys-values.html Sun, 05 Jul 2026 09:00:00 +0000 Attention is a soft lookup table: queries ask, keys answer, values deliver. Here is that idea made concrete with numbers. Deep Learning Embeddings 101: What Cosine Similarity Actually Measures https://jahanzaibawan.com/blog/cosine-similarity-embeddings.html https://jahanzaibawan.com/blog/cosine-similarity-embeddings.html Sat, 04 Jul 2026 09:00:00 +0000 Cosine similarity measures angle, not distance. Understanding that difference will save you from subtle bugs in retrieval and clustering systems. NLP Train, Validation, Test: Why Three Splits Not Two https://jahanzaibawan.com/blog/train-validation-test-splits.html https://jahanzaibawan.com/blog/train-validation-test-splits.html Sat, 04 Jul 2026 09:00:00 +0000 Two splits let your decisions leak into your test score. A validation set is what keeps your final number honest. Evaluation The Trap of Optimizing the Wrong Metric https://jahanzaibawan.com/blog/optimizing-the-wrong-metric.html https://jahanzaibawan.com/blog/optimizing-the-wrong-metric.html Sat, 04 Jul 2026 09:00:00 +0000 High accuracy can hide a useless model. Here is why the metric you optimize quietly decides what your system actually learns. Evaluation Feature Scaling: Which Models Care and Which Do Not https://jahanzaibawan.com/blog/feature-scaling-which-models-care.html https://jahanzaibawan.com/blog/feature-scaling-which-models-care.html Fri, 03 Jul 2026 09:00:00 +0000 A practical guide to why distance and gradient based models demand scaled features while tree based models shrug them off entirely. Machine Learning Class Weights vs Resampling for Imbalanced Data https://jahanzaibawan.com/blog/class-weights-vs-resampling.html https://jahanzaibawan.com/blog/class-weights-vs-resampling.html Fri, 03 Jul 2026 09:00:00 +0000 Class weights and resampling both target imbalance, but they change different things. Here is how to pick the right one and evaluate it honestly. Evaluation Early Stopping vs Regularization: Do You Need Both https://jahanzaibawan.com/blog/early-stopping-vs-regularization.html https://jahanzaibawan.com/blog/early-stopping-vs-regularization.html Thu, 02 Jul 2026 09:00:00 +0000 Early stopping and regularization both fight overfitting, but they work through different mechanisms. Understanding the difference tells you when you need one, the other, or both. Machine Learning A tuned GRU beat LoRA-fine-tuned GPT-2, here's why https://jahanzaibawan.com/blog/gru-vs-gpt2.html https://jahanzaibawan.com/blog/gru-vs-gpt2.html Sat, 20 Jun 2026 09:00:00 +0000 A 117M-parameter transformer lost to a small recurrent net on every metric. Not because transformers are bad, but because the baseline was done properly. Baselines How a naïve train/test split inflated my UAV F1 by 0.5 https://jahanzaibawan.com/blog/uav-data-leakage.html https://jahanzaibawan.com/blog/uav-data-leakage.html Fri, 12 Jun 2026 09:00:00 +0000 The same model, the same data, and a macro-F1 that fell from 0.78 to 0.26 the moment I split the data honestly. A walk through the most expensive bug in ML. Evaluation