Engineering / advanced
Train and serve learning over graphs
Follow graph representation, neighborhood sampling, message passing, partitions, and update paths while testing leakage and stale-neighbor effects.
Your route through engineering
Follow the path or begin with a question. Every lesson is open.
Your place, at your pace. Progress is optional and saved only in this browser when available.
Read / explore / reason
Inside this module
Module 01 advanced
Train and serve learning over graphs
Follow graph representation, neighborhood sampling, message passing, partitions, and update paths while testing leakage and stale-neighbor effects.
- 01
Store graph neighborhoods with explicit CSR and CSC orientation
Connect pointer arrays, edge multiplicity, degree counts and feature access patterns.
24 min - 02
Budget sampled neighborhoods without losing estimator semantics
Track occurrence growth, unique nodes and probability weights separately.
24 min - 03
Specify message direction, normalization and nonlinear placement
Distinguish sums, means, GCN normalization and attention rather than treating them as interchangeable averages.
24 min - 04
Compact minibatches while preserving global node identity
Share feature rows without accidentally merging messages, layers or time contexts.
24 min - 05
Balance partitions using the communication the model actually performs
Relate edge cuts, boundary-node reuse and per-layer traffic under explicit assumptions.
24 min - 06
Own, cache and update distributed embedding rows explicitly
Separate trainable ID tables from feature-based embedding generation and account for optimizer state.
24 min - 07
Build graph state from information available at prediction time
Separate event time, arrival time, tie policy and state-update order.
24 min - 08
Define what a negative edge means before sampling it
Control false negatives, information boundaries and candidate distributions in link prediction.
24 min - 09
Serve graph features and representations with a coherent freshness policy
Trace dependencies, stale publications and model-compatible cache versions.
24 min - 10
Match graph evaluation splits to the claimed generalization
Separate transductive visibility, unseen-node encoding and temporal prediction protocols.
24 min