AI & Machine Learning
9 clusters, spanning half-life layers 0–3: Math foundations, Core Python data stack, Classical ML, Deep learning frameworks, Architectures, Domains, MLOps, Managed platforms, Model serving.
Clusters
Section titled “Clusters”Math foundations
Layer 0 and permanent. The single highest-return investment in this domain, and the one most often skipped.
Core Python data stack
The array/broadcast model under NumPy transfers to every framework in this domain, including PyTorch’s tensors. DataFrame APIs transfer poorly…
Classical ML
The estimator/transformer/pipeline interface, cross-validation and leakage avoidance transfer across all of them and are most of what matters.…
Deep learning frameworks
Autodiff, the computational graph, the training loop and device placement transfer across all of them. Eager vs traced execution splits them, and…
Architectures
Layer 1 and the durable core: attention, convolution and the sequence-modelling problem outlive every framework that implements them.
Domains
Task framing, dataset construction and evaluation metrics transfer within a modality and rarely across one. Library APIs are Layer 3.
MLOps
Experiment tracking, model registries, reproducibility and drift monitoring are the same five jobs in every tool. Nothing else transfers.
Managed platforms
The managed-training and managed-endpoint shape transfers across all three, and maps to the cloud cluster’s primitive-set argument. SDKs and job…
Model serving
Batching, concurrency, warm-up and the latency/throughput trade transfer across all of them and are the content. ONNX is the interchange layer under…