AI Foundations Route
This route is for readers who want to fill in AI fundamentals systematically: data, models, training, evaluation, neural networks, deep learning math, and safety boundaries. Each stage favors a small runnable experiment over a pile of terminology.
| Module | Core question | Hands-on evidence | Common mistake |
|---|---|---|---|
| Data and features | How input columns, labels, normalization, and missing values affect the model. | Record fields, scaling method, and train/test split. | Looking only at accuracy without checking leakage or class balance. |
| Classical ML | How K-means, linear baselines, and metrics create a reference point. | Save parameters, SSE or confusion matrix, and rerun different seeds. | Treating one random run as a stable conclusion. |
| Neural networks | How matrix multiplication, activations, loss, and gradients connect. | Check one small batch through forward pass, loss, and gradients. | Skipping shape checks and immediately scaling the model. |
| Safety boundaries | Where model input, training data, and external tools can go wrong. | Write a threat model, misuse scenario, and mitigation note. | Moving demo scripts directly into production. |
AI learning path
A practical route from AI foundations to AI security
Track local progress from AI concepts to machine learning workflow, evaluation, neural networks, implementation projects, traffic-defense notes, and AI security engineering.
-
1
Concept map
AI Basics Learning Roadmap
Separate AI, machine learning, and deep learning before going into implementation details.
-
2
Workflow
Machine Learning Workflow
Follow the practical path from data and features to training, prediction, and evaluation.
-
3
Evaluation
Model Training and Evaluation
Understand loss, overfitting, train/test splits, accuracy, recall, and F1.
-
4
Neural networks
Neural Network Basics
Move from perceptrons to activation, forward propagation, backpropagation, and training loops.
-
5
Matrix calculus
Matrix Calculus for Neural Networks
Derive dL/dW for y = Wx + b and verify it with finite differences.
-
6
Backpropagation
Backpropagation as a Computation Graph
Trace local gradients through ReLU and softmax cross-entropy in a two-layer MLP.
-
7
Optimizers
Gradient Descent and Optimizer Geometry
Compare gradient descent, momentum, and Adam on a visible quadratic loss surface.
-
8
Convolution math
Convolution and Receptive Field Math
Compute convolution output size, receptive fields, channel mixing, and im2col layout.
-
9
Attention math
Transformer Attention Math
Hand-calculate Q/K/V scores, softmax weights, masks, multi-head structure, and KV cache.
-
10
LLM internals
LLM Visualizer
Inspect tokenization, embeddings, attention, next-token sampling, and KV cache with a local browser simulation.
-
11
Practice
Python AI Mini Practice
Run a small scikit-learn classification task and read the experiment output.
-
12
Security map
AI Security Threat Modeling
Use NIST, MITRE ATLAS, and OWASP to build a reviewable AI defense map.
-
13
Robustness
Adversarial Examples and Robust Evaluation
Run an FGSM-style digits experiment and compare clean and perturbed accuracy.
-
14
Data integrity
Data Poisoning and Backdoor Defense
Measure poison rate, trigger behavior, and attack success rate.
-
15
Privacy
Model Privacy and Extraction Defense
Measure membership signal and surrogate fidelity for a local toy model.
-
16
LLM security
LLM, RAG, and Agent Security
Separate instructions from data and enforce tool permission boundaries.
Review method
After each topic group, use the same dataset for three checks: run the smallest useful baseline, change one parameter and observe the output, then document one failed input. That moves the work from reading a tutorial to judging model behavior.