Course and AIMA-aligned sequence
Use this as the default full-course order. It begins with state-space search, moves through logical and probabilistic reasoning, then develops learning, representation, vision, and reinforcement learning.
Curriculum Map
The suite supports two different orders. The course-aligned sequence preserves conceptual dependencies and follows the broad AIMA progression from search to logic, probability, machine learning, and reinforcement learning. The quick-entry sampler is optimized for a first visit, not for teaching the full course in order.
Quick-entry sampler
Use this for demonstrations, open houses, or a four-lesson introduction. It deliberately jumps across the field to create four immediate and contrasting experiences.
Quick-entry four-app sampler
Visible comparison of BFS and A*.
A memorable base-rate reversal.
Training success versus generalization.
Trial and error producing a policy.
Why this is not the course order: the sampler maximizes variety and first-visit impact. It skips prerequisite development and should not replace the full sequence below.
Foundations / course track基础 / 课程路径
This sequence matches the structure used to build the course and the broad ordering of foundational topics in Artificial Intelligence: A Modern Approach. It is an instructional alignment, not a chapter-by-chapter reproduction.
Pathfinding VisualizerHill Climbing and Simulated AnnealingWumpus WorldCNF and SAT BuilderBayes Rule PlaygroundBayesian NetworkK-Nearest NeighborsOverfitting ExplorerTiny Neural NetworkK-Means ClusteringConvolution PlaygroundQ-Learning GridworldTransformer Language Modeling
| # | Applet | Concept area | Why it appears here |
|---|---|---|---|
| 1 | Pathfinding Visualizer | Search and problem solving | Introduce controlled exploration, optimality, and heuristic focus. |
| 2 | Hill Climbing and Simulated Annealing | Search and optimization | Show why local improvement can stall and why escape strategies matter. |
| 3 | Wumpus World | Logic and knowledge | Bridge perception, inference, uncertainty, and safe action. |
| 4 | CNF and SAT Builder | Logic and knowledge | Formalize propositional reasoning and search over assignments. |
| 5 | Bayes Rule Playground | Probability | Surface base-rate neglect before moving to graphical models. |
| 6 | Bayesian Network | Probabilistic reasoning | Extend conditional probability into dependence and explaining away. |
| 7 | K-Nearest Neighbors | Machine learning | Introduce supervised classification through visible neighborhood votes. |
| 8 | Overfitting Explorer | Machine learning and evaluation | Separate training performance from generalization on new data. |
| 9 | Tiny Neural Network | Neural networks | Build intuition for hidden representations and nonlinearity. |
| 10 | K-Means Clustering | Unsupervised learning | Introduce iterative unsupervised clustering and initialization sensitivity. |
| 11 | Convolution Playground | Computer vision | Connect local filters to feature maps and learned visual representations. |
| 12 | Q-Learning Gridworld | Reinforcement learning | Conclude with trial-and-error learning, delayed reward, and policy emergence. |
| 13 | Game Trees: Minimax and Alpha-Beta Pruning | Adversarial search | Extend search to an opponent: back terminal utilities through alternating MIN/MAX nodes, then prune branches that cannot change the exact minimax result. |
Modern AI extensions现代 AI 扩展
These labs extend the course-facing foundations into contemporary AI systems. They are optional extensions, not prerequisites for the classical introductory sequence.这些实验把面向课程的基础内容延伸到当代 AI 系统。它们属于可选扩展，而不是经典入门序列的先修要求。
Transformer Language Modeling
A bridge between advanced introductory NLP and modern generative language models: inspect causal self-attention and exact next-token probabilities.连接高级入门 NLP 与现代生成式语言模型：检查因果自注意力与精确的下一词元概率。
Agent Tool Use and Context Protocols
Go beyond the traditional introductory backbone to inspect tool schemas, authorization, execution, observations, provenance-aware context, and stopping.超出传统入门课程主干，检查工具模式、授权、执行、观察、带来源信息的上下文与停止决策。
Beginning-of-course Quick Assigns学期初快速任务
These Level-1 activities turn the existing in-applet response packets into stable, directly assignable 10-15 minute tasks. The link opens the applet in Use in class mode so the response packet is visible immediately.
这些 Level 1 活动把 applet 内已有的作答包正式化为稳定、可直接布置的 10-15 分钟任务。链接会直接以“课堂使用”模式打开 applet，使作答包立即可见。
| ID | Activity活动 | Focus重点 | Time时间 |
|---|---|---|---|
| QA-SEARCH-01 | A* vs BFS: same goal, different work A* 与 BFS：同一目标，不同搜索工作量 | Compare frontier ordering, explored work, and shortest-path behavior for BFS and A* on the same maze. 比较同一迷宫中 A* 与 BFS 的路径质量、搜索工作量和前沿顺序。 | 10-15 min / 10-15 分钟 |
| QA-LOCAL-01 | Why local search gets stuck or escapes 为什么局部搜索会卡住或逃离 | Compare one local-search trajectory or matched seeded restarts, separating reliability from final and best solution cost. 比较单次局部搜索轨迹或匹配的种子重启，并区分成功可靠性、最终成本与最佳成本。 | 10-15 min / 10-15 分钟 |
| QA-WUMPUS-01 | Safe, risky, or unknown? 安全、危险还是未知？ | Use percept evidence to distinguish a square that is proven safe, possibly hazardous, or unresolved. 根据感知证据区分已证明安全、可能有危险和仍未确定的状态。 | 10-15 min / 10-15 分钟 |
| QA-SAT-01 | SAT, UNSAT, or entailed? SAT、UNSAT，还是蕴含查询？ | Convert a knowledge base to CNF, inspect DPLL or CDCL evidence, and explain any learned clause and backjump without overclaiming solver fidelity. 把知识库转换为 CNF，检查 DPLL 或 CDCL 证据，并解释学习子句与非按时间顺序回跳。 | 10-15 min / 10-15 分钟 |
| QA-BAYES-01 | Base rates and false alarms 基础率与误报 | Connect prior prevalence, true positives, false positives, and posterior probability. 连接先验流行率、真阳性、假阳性与后验概率。 | 10-15 min / 10-15 分钟 |
| QA-BN-01 | Explaining away 解释消除 | Predict and explain how evidence for one cause can lower belief in another after conditioning on a common effect. 预测在共同结果已知后，一项原因证据如何降低对另一原因的信念。 | 10-15 min / 10-15 分钟 |
| QA-KNN-01 | Which neighbors vote—or get averaged? 哪些邻居投票或被平均？ | Compare categorical voting with continuous-target averaging while holding neighbor selection and relevant controls constant. 在保持邻居选择和相关控件不变时，比较类别投票与连续目标平均。 | 10-15 min / 10-15 分钟 |
| QA-OVERFIT-01 | Fit the training set, fail the future 拟合训练集，却败给新数据 | Separate training error from validation/test behavior as model capacity changes. 随模型容量变化区分训练误差与验证/测试表现。 | 10-15 min / 10-15 分钟 |
| QA-NN-01 | Why nonlinearity changes capacity 为什么非线性改变表示能力 | Compare an affine-only network with a nonlinear representation and explain the decision-boundary change. 比较纯仿射网络与非线性隐藏表示并解释决策边界的变化。 | 10-15 min / 10-15 分钟 |
| QA-KMEANS-01 | Assign, move, repeat 分配、移动、重复 | Predict one assignment and centroid update, then explain the alternating k-means cycle. 预测一次分配和质心更新，再解释 k-means 的交替循环。 | 10-15 min / 10-15 分钟 |
| QA-CNN-01 | One convolution cell 一个卷积输出单元 | Predict one output-cell multiply-and-sum before revealing the feature-map value. 在揭示特征图值之前预测一个输出单元的乘加计算。 | 10-15 min / 10-15 分钟 |
| QA-QL-01 | One Q-update 一次 Q 更新 | Predict an action, TD target, and update direction before stepping the learner. 在学习器执行一步之前预测动作、TD 目标与更新方向。 | 10-15 min / 10-15 分钟 |
| QA-MINIMAX-01 | Same answer, less search 同一答案，更少搜索 | Explain why a safe alpha-beta cutoff can reduce evaluated work without changing the minimax result. 解释安全的 Alpha-Beta 截断为何减少评估工作却不改变 minimax 结果。 | 10-15 min / 10-15 分钟 |
| QA-TRANSFORMER-01 | Attend, then predict 先注意，再预测 | Connect a controlled representation/attention change to the next-token probability distribution. 把受控的表示/注意力变化连接到下一 token 概率分布。 | 10-15 min / 10-15 分钟 |
| QA-AGENT-01 | A proposed call is not an executed action 提出工具调用不等于执行动作 | Trace a proposed tool call through validation, authorization, execution, observation, and context update. 追踪一个工具调用通过验证、授权、执行、观察与上下文更新。 | 10-15 min / 10-15 分钟 |
Applet map
Each card uses the same accent color as its applet, so the concept remains visually recognizable across the landing page, Teacher Pack, and curriculum materials.
Search搜索
Pathfinding Visualizer寻路可视化
Compare breadth-first search (BFS), depth-first search (DFS), Dijkstra’s algorithm, and A* on the same grid.在同一网格上比较广度优先搜索（BFS）、深度优先搜索（DFS）、Dijkstra 算法和 A*。
20 min · start
Local search局部搜索
Hill Climbing and Simulated Annealing爬山法与模拟退火
Compare hill climbing with simulated annealing on the same optimization problem and starting state.在同一优化问题和起始状态上比较爬山法与模拟退火。
25 min · core
Logical agents逻辑智能体
Wumpus WorldWumpus 世界
Follow a logical agent as percepts become safe-cell conclusions, unresolved risks, and forced choices.观察逻辑智能体如何把感知转化为安全结论、未解决风险和被迫选择。
25 min · core
Logic and SAT逻辑与 SAT
CNF and SAT BuilderCNF 与 SAT 构建器
Rewrite propositions into conjunctive normal form (CNF), then watch a satisfiability solver prune assignments.把命题改写为合取范式（CNF），再观察可满足性求解器如何剪枝赋值。
25 min · core
Probability概率
Bayes Rule Playground贝叶斯定理演练场
Change the base rate and test quality to see why high accuracy can still produce a surprising posterior.改变基率和检测质量，观察高准确率为什么仍会产生意外的后验概率。
20 min · start
Probabilistic inference概率推理
Bayesian Network贝叶斯网络
Edit evidence, compare exact inference with sampling, and see explaining away in a causal graph.编辑证据，比较精确推理与采样，并在因果图中观察解释消除。
30 min · core
Classification分类
K-Nearest NeighborsK 最近邻
Move a query point, change k and vote weighting, and inspect every neighbor that determines the class.移动查询点，改变 k 和投票权重，并检查决定类别的每个邻居。
20 min · start
Model evaluation模型评估
Overfitting Explorer过拟合探索器
Raise model complexity and compare training error with test error on fresh samples.提高模型复杂度，并比较训练误差与新样本上的测试误差。
25 min · start
Neural networks神经网络
Tiny Neural Network微型神经网络
Train a compact network and compare linear composition with a nonlinear hidden representation.训练一个小型网络，并比较线性组合与非线性隐藏表示。
30 min · advanced
Clustering聚类
K-Means ClusteringK 均值聚类
Change k and initialization, then watch assignment and center updates reshape the clusters.改变 k 和初始化方式，观察分配与中心更新如何重塑聚类。
25 min · core
Computer vision计算机视觉
Convolution Playground卷积演练场
Slide kernels across an image, inspect each multiply-and-sum, and compare hand-built with learned filters.让卷积核滑过图像，检查每次乘加，并比较手工滤波器与学习卷积核。
30 min · advanced
Reinforcement learning强化学习
Q-Learning GridworldQ 学习网格世界
Train by trial and error, inspect value backups, and compare exploration schedules on the same world.通过试错训练，检查价值回传，并在同一世界上比较探索策略。
30 min · advanced
Language models语言模型
Transformer Language ModelingTransformer 语言建模
Change a token, mask, or temperature and watch the next-token probabilities move.改变一个词元、掩码或温度，观察下一词元概率怎样变化。
30 min · advanced
Agent systems智能体系统
Agent Tool Use and Context Protocols智能体工具调用与上下文协议
Follow an agent as it checks a tool call, permission, result, and stopping point.跟随智能体检查工具调用、权限、返回结果和停止时机。
30 min · advanced
Foundations基础
Game Trees: Minimax and Alpha-Beta Pruning博弈树：Minimax 与 Alpha-Beta 剪枝
Back terminal utilities through alternating MIN/MAX turns, then inspect exact alpha-beta cutoffs and move-order work.通过交替的 MIN/MAX 回传终局效用，再检查精确的 Alpha-Beta 截断以及行动顺序对工作量的影响。
30 min · Intermediate
Assessment evidence
Predict and observe
Capture initial intuition, then cite a visible state, value, trace, or decision after running the experiment.
Explain and transfer
Name the mechanism that produced the result, then apply the idea to another AI system or decision.