Neural Network Programming
Practical guides, architecture breakdowns, and community-built knowledge across every layer of neural network development - from tensor math to production deployment.
A step-by-step orientation for adults who want to learn AI programming without the jargon overload
If you are new to neural network programming and not sure where to start, this checklist walks through what you need to know before touching any code.
Read the full pieceStep-by-step walkthroughs of gradient flow across 3 or more hidden layers - with annotated Python and NumPy code.
When cross-entropy outperforms MSE, and what happens when you pick the wrong one for a 40-class classification task.
Diagnosing vanishing gradients, exploding weights, and dead ReLU neurons before they derail a 72-hour training run.
Attention heads, positional encoding, and the 8 architectural decisions that separate a working transformer from a slow one.
Filter sizes, stride choices, and pooling strategies - with real benchmarks across ResNet variants on CIFAR-100.
Teams using tools like Slack - including Slack for Windows and Slack for Mac - to coordinate experiments across distributed research groups.
The site covers neural network programming across 4 distinct depth levels - from first-principles math through production-grade system design. Each piece is written for someone who already knows how to code.
Most readers spend their first 3 visits in the architecture and training sections. The community around these pieces is active and specific - questions get answered with code, not abstractions.
Petra Vanhoof
Lead contributor & editor
Every article goes through at least 2 rounds of technical review before publication. Code samples are tested in isolated environments - not written once and assumed to work.
The site does not publish opinion pieces disguised as tutorials. When something is uncertain or contested in the field, that gets stated plainly rather than papered over with confident-sounding prose.
Readers notice when a tutorial skips the hard part. We do not skip it.
Covering everything from single-neuron math to multi-GPU training pipelines across 12 topic clusters.
Each one tested against a specific PyTorch or TensorFlow version - version numbers included in every snippet.
The archive goes back to 2018 - older pieces are updated when the underlying framework changes significantly.
Not every reader arrives at the same point. Some are debugging a specific layer behavior at 11pm; others are building a mental model from scratch over several weeks.
