- 帶有太多行話、jargon的自黑,限於同行之間傳播就好。如果要公開傳播,切記笑點不要設置得太高。
- 在隨手轉推之前,先想清楚自己有沒get到point。
- 如果你真要寫钓鱼文,最起碼应该在结尾注明真相。Anyway,我个人还是不同意釣魚文这种與人性相悖的手段。
Saturday, December 8, 2012
關鍵詞:反讽、自黑、釣魚、stereotype
Wednesday, November 14, 2012
My workflow of copying and pasting content between Mac and iOS device
Thursday, September 20, 2012
转载:中年夫妻「相看兩相厭」?
轉載自親子天下雜誌
原文:
有些中年夫妻結婚許久,卻各自活在自己的世界,沒去理解對方應對問題的方式。這時,一個小小的溝通不良,就足以讓彼此覺得「真是夠了」...
============================
我想,很多夫妻情侶應該都有同感吧~
原文:
有些太太在工作上受挫,回家跟先生說今天工作不順,先生經常會答:「你就是怎樣怎樣不對……應該怎樣怎樣做才對……」通常太太希望聽到的是肯定與支持,所以當先生開始唸唸唸,太太覺得好厭煩:一是眼前這個男人何時變得這麼愛說教;二是他的觀點也許根本不正確。
=====================================
男人想法和女人真的不同,男人直覺想法就是想解決問題,所以才會碎碎念,說一堆分析和解決方式...
如果女人覺得不喜歡,也請直接或間接提醒一下,兩個人生活在一起就是要互相包容調整吧....
原文:
這種厭煩是一定會有的,幾乎沒有一個伴侶可以百分之百讓人滿意。面對先生的碎唸,這時請太太先問自己:「為什麼那麼厭煩?」是不是有些事情你必須倚賴先生,當他不能滿足時會讓人受不了;那麼這個東西,你有沒有辦法健全自己而不用去要求他。
說穿了,婚姻裡有兩種基本的安全感。第一是我跟你在一起,你喜不喜歡我,證明的方法是:我煮的湯很難喝你也要喝下去、我的屁很臭你也要聞、我媽要來你就得開門、我要你媽走你就要把她踢出去、我要做一件事你得認同我,不然就是看不起我。第二是我跟你在一起會不會被你害到,比方說你抽菸,我要你戒菸,否則十年後你生病了我要照顧你。這兩種安全感衍生出所有人跟人之間的問題,這是夫妻進入兩人生活最在意的事情。
太太跟先生抱怨工作不順時,若沒有得到他的支持,其實也不會怎樣;但若是得不到對方的支持,就會認為對方不夠關心、不夠肯定。這就是第一種安全感。
可是回頭去想,為什麼先生講不出好聽的話?我發現先生們常覺得太太如果受到挫折,自己也會感到挫折,夫妻之間的感覺是會流動分享的。所以先生會想趕快排除這個挫折感,他沒有辦法扛著你的挫折,跟你一起難過、罵人;他想趕快找個方法不要停留在挫折感裡面,結果他講出來的話就是急著排除感覺的方法。
=============================
男人真的只會想著如何解除這些煩人的問題,不想一直聽著不停的抱怨...
原文:
太太要克服厭煩的方式是,先告訴先生他的方法很好,然後說:「除了方法之外我想問你,在你心目中我是不是值得的人?我是不是特別笨?你可不可以給我一點鼓勵?」我也會鼓勵太太把感覺講清楚,比如說:「你剛剛的反應讓我發現,我今天工作的不愉快有感染到你,因為你看來也很不愉快,我剛剛講話是不是很不耐煩?」先生會說:「對,你的臉真的好臭。」太太可以說:「原來你很愛我,所以我感到挫折時,你馬上感到好像是自己的挫折,才有這麼強烈的反應。這樣我很安心,可是以後你不要這麼挫折,因為我講講就沒事了,你只要講一些無關痛癢的安慰話,反而我自己就會好了。」讓先生意識到「原來我的情緒會被太太感染」很重要。
============================
一直抱怨的壞心情,只會讓另一半覺得更心煩進而影響到心情,只是,很多人無法像作者這麼清楚的指出重點,當然就不可能解決這種問題...
原文:
有些夫妻會發展出一些暗號:「對不起,我沒有問你意見喔。」提醒另一半不用那麼辛苦的給意見。因為有些男人覺得給意見才是愛你跟認真,要是他像連續劇裡演的那樣摸摸老婆的頭、親親她,完全不費力,他就是個爛男人。
===========================
我也是屬於"覺得給意見才是愛妳跟認真的",只是,反而會讓女人覺得碎碎念(已經在改進了)...
有些女性的確也不需要男人的意見,純粹只是發洩情緒,
如果沒有發展出像作者說的一些暗號提醒另一半,
<真是夠了>這種心情就會常常出現,
其實,這就是需要雙方的包容和相互的調整,
沒有人是完美無瑕的....
只是,一般男人很難理解女人的情緒變化,如果這個男人能這麼了解女人的話,
我想,應該是情聖才辦的到吧,而情聖通常不會只有一個女人...
Saturday, June 30, 2012
Saturday, May 12, 2012
Compressed Sensing
Compressed sensing is an interesting and hot topic recently. I wrote an report on it for my optimization course. I want to summarize it in a more succinct way in a blog post.
Overview
Consider an underdetermined linear system Ax = b, where A is a nxN matrix and n << N, x in R^N. If A is not degenerate (or has rank n), then there will be infinitely many solutions to this system. However, if we know the solution x is sparse, i.e., the number of non-zeros is only a few, then the solution is unique and under a certain condition we may find it exactly.
Application Background
Many application can be formulated in the above way. For example, in signal processing, to recover the signal x, one must have enough samples to do so. According to Nyquist-Shannon sampling theorem, the sampling rate must be no less than half of the frequency. The matrix A is usually called measurement matrix, which corresponds to the measurement, or sensing in our context. However, in compressed sensing we can still recover the signal even when the samples are not enough. This may seem contradicting, but it is not. The sampling theorem does not make any assumption to the signal, but we does – the signal is sparse.
Another example is image 'sensing'. When we are taking images using a digital camera, we must capture all of the pixels. If the resolution is high, the resulting image file will be very large, and people may want to compress it. The modern approach in image compression is wavelet transform, which is used in JPEG2000 standard. An image can be viewed as a 2D signal, and can be represented using a set of wavelets and coefficients. This is similar to Fourier transform, where the signal can be represented by a set of sinusoids and coefficients. The wavelet or sinusoids are called 'basis'. Interestingly, we may always find a basis such that only a few components in it are significant, while others are not. The compressions relies on such an idea, we may safely discard the unimportant components but keep those important components. This will not hurt the image quality too much, and human-eyes cannot see the difference. Intuitively, we can plot the coefficient of the components, and we will see some 'sparks' in the plot, while the others are very small magnitude that is close to zero. We threshold those small magnitude and keep only those sparks. This is roughly how the image compression works. Nevertheless, if we are to abandon lots of information anyway, why not just only capture the important ones from the very beginning?
Key Discoveries of CS
The following summarizes the keys:
- If the solution x is sparse enough and matrix A satisfies 'Restricted Isometry Property' (RIP), then x is unique.
- RIP itself is hard to check, but if we obtain A using some random process, it is almost always guaranteed (or, with high probability, with high confident).
- We can formulate an optimization to find the solution: minimize ||x||_0 s.t. Ax=b, where ||x||_0 is the l0-norm of x. Note that the norm is not a true 'norm', but just counting the number of non-zeros in x. However, l0-norm is clearly not continuous, not differentiable and not convex, the minimization is combinatorially hard, or, NP-hard.
- We can convexify the problem. Instead of minimizing l0-norm, we minimize l1-norm, which can be recast as linear program and solved efficiently.
- When A satisfies some property, the l1-minimization has the exact solution as l0-minimization.
Disclaimer
The above description is based on the paper I read and my interpretation. Some description may not be very accurate or rigorous in math.