This is the most easy trek, the most touristy one. We didn't have much equipment, and we just dropped our bags and kept going through the storm,"（译注：这条路线向来很简单，很游客化，我们并没有携带太多装备，风暴之下，我们扔了背包挣扎前行）（信息来源: http://theguardian.com）
To walk through life in a comfortable way is still not my goal. This is why I want to try to climb Annapurna a third time. I would like to implement my dreams and visions into reality. Annapurna is one of them.（译注：安稳度过此生并不是我的目标，这也是我为什么第三次尝试安纳普尔纳；我想将我所有的梦想都变成现实，安纳普尔纳攀登正是其中之一。）
Choiniere J N, Xu X, Clark J M et al. 2010. A basal alvarezsauroid theropod from the early Late Jurassic of Xinjiang, China. Science, 327(5965): 571-574
Gauthier J 1986. Saurischian monophyly and the origin of birds. Memoirs of the California Academy of Sciences, 8: 1-55
Heilmann G 1927. The origin of birds. New York: Bover Publications, INC.
Hu D, Hou L, Zhang L et al. 2009. A pre-Archaeopteryx troodontid theropod from China with long feathers on the metatarsus. Nature, 461(7264): 640-643
Huxley T H 1868. On the animals which are most nearly intermediate between birds and reptiles. Annals and Magazine of Natural History. 4th, 2: 66–75
Ji Q, Ji S A 1996. On the discovery of the earliest fossil bird in China (Sinosauropteryx gen. nov.) and the origin of birds. . Chinese Geology, 233: 30-33
Ostrom J H 1973. The ancestry of birds. Nature, 242: 136–136
Xu X, Clark J M, Forster C A et al. 2006. A basal tyrannosauroid dinosaur from the Late Jurassic of China. Nature, 439(7077): 715-718
Xu X, Zhou Z, Dudley R et al. 2014. An integrative approach to understanding bird origins.Science, 346(6215): 1253293
Xu X, Zhou Z, Wang X et al. 2003. Four-winged dinosaurs from China. Nature, 421(6921): 335-340
Zhou Z 2004. The origin and early evolution of birds: discoveries, disputes, and perspectives from fossil evidence. Naturwissenschaften, 91(10): 455-471
References / 参考资料： [1] APA.org. Breakups aren't all bad: Coping strategies to promote positive outcomes. [2] Battaglia, D. M., Richard, F. D., Datteri, D. L., & Lord, C. G. (1998). Breaking up is (relatively) easy to do: A script for the dissolution of close relationships. Journal of Social and Personal Relationships, 15(6), 829-845. [3] Kross, E., Berman, M. G., Mischel, W., Smith, E. E., & Wager, T. D. (2011). Social rejection shares somatosensory representations with physical pain. Proceedings of the National Academy of Sciences, 108(15), 6270-6275. [4] Fisher, H. E., Brown, L. L., Aron, A., Strong, G., & Mashek, D. (2010). Reward, addiction, and emotion regulation systems associated with rejection in love. Journal of neurophysiology, 104(1), 51-60. [5] Wiki How: How to benefit from a breakup? [6] Lauren Howe on TheAtlantic.com. Why Some People Take Breakups Harder Than Others? (2016) [7] Alexandra Nicole Torres on ThoughtCatalog.com. The Psychology Of Letting Go: 11 Tips To Help You Through Your Breakup (2017). [8] Denise Baptiste on BoldSky.com. 12 Awesome Benefits of a Breakup (2014).
记得以前 Google 使用的在线笔试题，大多是偏重算法的题目，可以看做是简化版的 ACM。这的确是一种设计开卷笔试题的思路，因为很多经典算法问题，都可以被包装到一个很具体的场景中。学生考试时，即便用搜索引擎去查找这个具体场景，只要以前没考过类似的题目，就很难搜到正确答案。最起码，学生需要有一定的算法知识，能够将遇到的问题抽象成特定的算法模式，然后再去搜索对应的解决方案。
但是，回到 DeeCamp，如果按照简化版 ACM 的思路，就不能满足要求了。因为 DeeCamp 希望招聘的学生标准与 Google 要招聘的学生标准有很大不同。Google 是一家公司，想招进来的是最聪明的、编程能力和算法能力最强的学生。DeeCamp 是一个训练营，本来就是以培训为目的，当然不能把入门考试的要求设得太高。
也就是说，我们创新工场 AI 工程院这个命题小组，要设计的是一套 AI 入门级的开卷笔试题，无论是数学、编程、算法还是 AI，都只适合考察最基本的知识技能，而不适合强调题目的专深。同时，这套题还需要在题目的趣味性、灵活度等方面下功夫，最好有一些平常很少见到的新场景，以考察同学们在解决新问题时的实战技能。更重要的，这些题目需要在学生开卷答题的情况下，还能有足够的区分度。
其实，TensorFlow 白皮书里，"5.1 Common Subexpression Elimination"就是关于计算图优化的一个举例说明。有关符号执行对于计算图优化的作用，MXnet 文档里有一篇特别浅显的讲解：Programming Models for Deep Learning。其中对流行的框架做了一个归类："Examples of imperative style deep learning libraries include Torch, Chainer and Minerva. While the examples of symbolic style deep learning libraries include Theano, CGT and Tensorflow."
在正常配置的计算机里，这个题目里的代码是很难跑出结果的。这也是为什么题目描述里会强调"假设可以不考虑计算机运行资源（如内存）的限制"的原因。不相信这一点的同学当然会去尝试运行这段代码，但结果么，多半是内存异常，或者持续运行无响应。最糟糕的，电脑甚至会直接崩溃。实际考试时，同学报告了好几个崩溃案例；我猜，主要是在 Windows 系统中吧（笑脸）。
所以，遇到这种题是不能硬来的。做题的学生还是要回过头来，读代码，理解代码的含义。
这个代码的函数名sieve其实暗示了代码的功用，Sieve of Eratosthenes，这是用筛法计算质数个数的一个经典方法。当然，不知道筛法这个词也不重要，重要的是要从代码逻辑里，读懂最关键的几行其实是在把所有非质数都筛掉。就算读起来费劲，那么，至少能想到，把最后一行改成比较小的输入数字，比如 10，比如 100，比如 1000，运行一下试试吧。
![(a*v)[n]=\sum_{m=-\infty}^{\infty}{a[m]v[n-m]}](https://lh3.googleusercontent.com/blogger_img_proxy/AEn0k_tnw2NJrhE8sD4F6YuRyvWNzQl1XbC11OPYeiU3SYj4cxvEbuAN8gW2AOBjY_q6GrjMqWpR3xvPlXgxPI3WNBz5lmXKKB_JdXx1MZrLwYNvHvYi0n3OV43cGBYK6I2hfPLpQfoyFpd6Vmb4hdYeif0lhYKuQYF_tisnH6xGfJtkknB1D6Xcx0a-YyhQw9m4nWqUiuf1Nvz0=s0-d)
