tinsir888s hjemmeside

Why visualizing The inner workings of ConvNets After training, each layer progressively extracts higher and higher-level features of the image, until the final layer essentially makes a decision on what the image shows. early layers: features like “corners” and “blobs” late layers: semantic like “wheel” and “door”. assemble features into complete interpretations Motivation why visualize CNN? Debugging with neural networks we have little insight about learning and internal operation through v ...

Orthogonal Problems aka 正交问题 input: set PPP of nnn points goal: process PPP in a DS​ query: axis-aligned rectangle rrr orthogonal range counting: number of points in rrr orthogonal range reporting: list of points in rrr 3-sided queries: 3 boundaries Rectangle stabbing aka 穿刺矩形 input: set PPP of nnn rectangles goal: process PPP in a DS query: a point qqq output: list of rectangles containing qqq 1D Orthogonal Range Searching input: set PPP of nnn points in 1D goal: process PPP in a DS query: a ...

last lecture about mechanism design mechanism without money? kidney exchange and stable matching Kidney exchange patients with a final stage kidney disease in the waiting list for kidney transplant by a deceased donor. 已故捐献者肾移植等候名单中患有末期肾病的患者。 98k in the US, 800 in Denmark demand>>supply! Transplantations from living donors 活体移植 e.g. transplantation from mother to child unfortunately, such transplantations are allowed between relatives and are not always possible Incompatibilities of bloo ...

Unsupervised learning regression vs. classification why unsupervised learning? few of data are labeled pre-train an unsupervised model anomaly detection - learn what data distribution typically looks like cheaper than supervised learning Generative models a generative model is a class of statistical model that contrasts with discriminative models. informally discriminative model - classifier data∼pdata(x)\text{data}\sim p_{data}(x)data∼pdata​(x) generative model - generate new data instance ...

Duality there is a 1-to-1 mapping between points and non-vertical lines map every point p to a line: p∗:y=pxx−pyp^*:y=p_xx-p_yp∗:y=px​x−py​, p∗∗:y=pxx+pyp^{**}:y=p_xx+p_yp∗∗:y=px​x+py​ every line l:y=ax+bl:y=ax+bl:y=ax+b to a point l∗=(a,−b)l^*=(a,-b)l∗=(a,−b). claim: if ppp is on(below/above) l⇔l∗l\Leftrightarrow l^*l⇔l∗ is on(below/above) p∗p^*p∗. Proof: p(px,py),l:y=ax+bp is on l⇒py=apx+bl∗(a,−b),p∗:y=pxx−pywe plug l∗ into p∗:pxa−(−b)=apx+b=pyQ.E.D.p(p_x,p_y),l:y=ax+b\\ \text{p is on l}\Rig ...

Inhalt Gesten sagen mehr als tausend Worte … Sprachen kinderleicht?! Smalltalk - Die Kunst der kleinen Worte Wenn zwei sich streiten 2-1 Sprich mit mir verbale Kommunikation und nonverbale Kommunikation 语言交际和非语言交际 非语言交际的方式 Gesichtsausdruck 面部表情 Mimik Körperhaltung und Körperbewegung 体态和动作 Berührung 触摸 räumliche Distanz 空间距离 stimmliche Merkmale 声音特征: Tonfall 语调, Sprechgeschwindigkeit 语速, Betonung 重音, Pause 停顿 usw. demütig 恭顺的，谦虚的 leidend 虚弱的 hochnäsig 自大的，目空一切的，傲慢的 das Piktogramm-e 形象符号，标志牌 ...

Budget constraints limit the amount of money that an agent can pay in practice: in a sponsored search auction, every bidder is asked for her bid-per-click and her daily budget. small but drastic change in modeling the behavior of participant iii with budget BiB_iBi​ and valuation viv_ivi​. utility for outcome ω\omegaω with payment pip_ipi​ {vi(ω)−pi, pi≤Bi−∞, pi>Bi\begin{cases} v_i(\omega)-p_i,\ p_i\le B_i\\ -\infty,\ p_i\gt B_i \end{cases} {vi​(ω)−pi​, pi​≤Bi​−∞, pi​>Bi​​ A single-item ...

Object localization Classification + Localization: ImageNet 1000 classes each image has 1 class and at least one bounding box 800 training images per class in the competition, algorithm should produce 5 (class, box) predictions correct if at least one prediction has the correct class and a bounding box with at least 0.5 intersection over union IoU Evaluation mean average precision mAP compute average precision separately for each class, then average over classes. a detection is a true positiv ...

Planar Point Location using Persistency 使用持久化数据结构进行平面点定位（在线查询） Given a planar subdivision with nnn segments, pre-process it for point location queries. 给定一个具有 n 段的平面细分，对其进行预处理以进行点位置查询。 nnn disjoint line segments that divide the plane into some regions query: given a point, ask which region it lies in. Goal: O(n)O(n)O(n) space O(log⁡n)O(\log n)O(logn) query time O(nlog⁡n)O(n\log n)O(nlogn) pre-processing Recall the solution for line segment intersection: sweep a vertical line lll over the plan ...

Geografi og vejr vejr 天气 地点前面使用 i 或者 på： 使用 i 的情况： by 城市 i Aarhus 在奥胡斯 halvø 半岛 i Jylland 在日德兰半岛 land 国家 i Danmark 在丹麦 使用 på 的情况 ø 岛 på Sjælland 在西兰岛 Vejr og temperatur 天气 solen skinner 出太阳 det er skyet 多云 det er overskyet 阴天 det regner 下雨 det blæser 吹大风 det sner 下雪 det lyner og tordner 电闪雷鸣 温度 Hvad er temperaturen? 温度如何？ Hvor mange grader er det? 多少度？ Det er 2°. (to grader) 两度。 Det er ÷ 2°. (minus to grader) 零下两度。 Det er omkring/cirka 5°. 大约五度。 Det er mellem 5 og 10°. 五到十度。 使用的摄氏温度！除了美国使用 ...