Explore by topic
Mathematics, probability, statistics, formal logic, and algorithmic principles.
Canada's coastline length is a scale-dependent measurement: coarse national figures (~150,000–200,000 km) and high-resolution geographic surveys (~244,000 km) coexist because of the coastline paradox, where finer units capture more fractal-like indentations.
A human riffle shuffle can look mixed yet retain structured order because its release pattern is not independent, individual cards do not have equal selection chances, and rigid spacing preserves positional relationships.
Distortion improves readability when the map preserves connection order and direction while replacing messy real-world geometry with a small set of constrained angles, even spacing, and clear color-coded route layers.
Entropy is not a vague measure of disorder; it is precisely the logarithm of how many microscopic arrangements are consistent with what we observe at the macroscale, and the macrostate we see is the one with the largest count of microstates.
Entropy is a count of microstates consistent with a macroscopic state, and that single count governs data compression, the minimum energy cost of computation, and the irreversible cooling of the cosmos.
The real-world sampling frames pollsters use to recruit about 1,000 respondents, and why weighting makes that sample represent millions.
A truly random shuffle gives every exact ordering of a deck the same chance—and 'looking unpredictable' is not the test.
How a random sample of about 1,000 people can estimate the opinions of millions, and why the same method fails when sampling is not random.
Public online security comes from private keys and one-way math, not from keeping the encryption method secret.
Investigate how many riffle shuffles it takes for a 52-card deck to lose all trace of its original order—and why 'truly random' is a measurable milestone, not just a feeling.
Why a deliberately distorted map can be more useful for navigating a train network than a true-to-life one.
探索高峰流量下绕行收益被变道和合流‘吃掉’的机制，并找到让绕行真正省时的临界条件。
你将会明白为什么遮挡、侧脸和暗光会击中人脸检测器的不同弱点，并能在新照片中预判检测结果。
Why a random poll of about 1,000 people can reliably reflect a country of millions—and when it can't.
奶茶店扎堆是因为每家都在追逐同一群顾客，最小化彼此间的竞争距离，从而抢占市场份额的下限。
独立随机误差在求平均时会按 √n 缩小，多人差异相互抵消，于是平均分天然比个人分数更"聚拢"。
揭示导航系统绕行决策背后的代价函数结构，并用它解释一次'被坑'的典型情形。
A node is locked the moment it is extracted because every alternative path to it would have to detour through an unvisited node and therefore be longer, making its tentative distance permanent.
The proof that Dijkstra's one-shot extraction is correct: the invariant maintained by the priority queue, the role of non-negative edge weights, and the geometric meaning of the settled frontier.
A capacity-controlled threshold separates approximation from interpolation: once a network has enough parameters to fit every training point exactly, the optimization problem shifts from finding shared patterns to finding any solution, and the learned function changes character accordingly.
地铁 App 用从起点按乘车时间逐层向外扩展的贪心搜索，每个站点只保留一次最优到达方式，因此几秒内就能锁定全程最短的换乘路线。
A node is locked the first time it is extracted as the minimum from the priority queue, because every still-unlocked node is at least as far away as the just-locked one, so no future relaxation can produce a shorter path to it.
Traffic waves are a useful approximation, but only on uniform stretches of road. Real networks break the assumption the moment intersections, signals, lane drops, or merges interrupt the free flow that the wave equation requires.