In the movie "Interstellar," we get an up-close look at a supermassive black hole. Set against a backdrop of bright gas, the black hole's massive gravitational pull bends light into a ring. However, this isn't a real photograph, but a computer graphic rendering -- an artistic interpretation of what a black hole might look like. 在電影《星際穿越》中， 我們得以近距離觀察一個超級黑洞。 在明亮氣體構成的背景下， 黑洞的巨大引力 將光線彎曲成環形。 但是，（電影中的）這一幕 並不是一張真正的照片， 而是電腦合成的效果—— 它只是一個對於黑洞 可能樣子的藝術表現。
A hundred years ago, Albert Einstein first published his theory of general relativity. In the years since then, scientists have provided a lot of evidence in support of it. But one thing predicted from this theory, black holes, still have not been directly observed. Although we have some idea as to what a black hole might look like, we've never actually taken a picture of one before. However, you might be surprised to know that that may soon change. We may be seeing our first picture of a black hole in the next couple years. Getting this first picture will come down to an international team of scientists, an Earth-sized telescope and an algorithm that puts together the final picture. Although I won't be able to show you a real picture of a black hole today, I'd like to give you a brief glimpse into the effort involved in getting that first picture. 一百多年前， 阿爾伯特·愛因斯坦 第一次發表了廣義相對論學說。 在之後的數年裡， 科學家們又對此提供了許多佐證。 但相對論中所預測的一點，黑洞， 卻始終無法被直接觀察到。 儘管我們大致知道一個黑洞 看起來應該是什麼樣， 卻從未真正拍攝過它。 不過，這個現狀可能很快就會改變。 在接下來幾年內，我們或許就能 見到第一張黑洞的圖片。 這一重任會落在一個由 各國科學家組成的團隊上， 同時需要一個 地球大小的天文望遠鏡， 以及一個可以讓我們合成出 最終圖片的算法。 儘管今天我不能讓你們 見到真正的黑洞圖片， 我還是想讓你們大致瞭解一下 得到第一張（黑洞）圖片 所需要的努力。
My name is Katie Bouman, and I'm a PhD student at MIT. I do research in a computer science lab that works on making computers see through images and video. But although I'm not an astronomer, today I'd like to show you how I've been able to contribute to this exciting project. 我叫凱蒂·伯曼， 是麻省理工學院的一名博士生。 我在計算機科學實驗室中進行 讓電腦解析圖片和視頻信息的研究。 儘管我並不是個天文學家， 今天我還是想向大家展示 我是怎樣在這個項目中貢獻 自己的一份力量的。
If you go out past the bright city lights tonight, you may just be lucky enough to see a stunning view of the Milky Way Galaxy. And if you could zoom past millions of stars, 26,000 light-years toward the heart of the spiraling Milky Way, we'd eventually reach a cluster of stars right at the center. Peering past all the galactic dust with infrared telescopes, astronomers have watched these stars for over 16 years. But it's what they don't see that is the most spectacular. These stars seem to orbit an invisible object. By tracking the paths of these stars, astronomers have concluded that the only thing small and heavy enough to cause this motion is a supermassive black hole -- an object so dense that it sucks up anything that ventures too close -- even light. 如果你遠離城市的燈光， 你可能有幸看到銀河系 那令人震撼的美景。 而如果你可以穿過百萬星辰， 將鏡頭放大到 2.6萬光年以外的銀河系中心， 我們就能抵達（銀河系）中央的 一群恆星。 天文學家們已經穿過星塵，使用紅外望遠鏡 觀察了這些恆星整整十六年。 但是天文學家們所看不到的東西 才是最為壯觀的。 這些恆星似乎是在圍繞一個 隱形的物體旋轉。 通過觀測這些星星的移動路徑， 天文學家們得出結論， 體積足夠小，而質量又大到能導致 恆星們如此運動的唯一物體 就是超級黑洞—— 它的密度極大，高到它能吸進 周圍所有東西， 甚至光。
But what happens if we were to zoom in even further? Is it possible to see something that, by definition, is impossible to see? Well, it turns out that if we were to zoom in at radio wavelengths, we'd expect to see a ring of light caused by the gravitational lensing of hot plasma zipping around the black hole. In other words, the black hole casts a shadow on this backdrop of bright material, carving out a sphere of darkness. This bright ring reveals the black hole's event horizon, where the gravitational pull becomes so great that not even light can escape. Einstein's equations predict the size and shape of this ring, so taking a picture of it wouldn't only be really cool, it would also help to verify that these equations hold in the extreme conditions around the black hole. 那麼，如果我們繼續放大下去， 會發生什麼？ 是不是就可能看見一些， 理論上不可能看到的東西呢？ 事實上，如果我們以 無線電波長放大， 我們會看到一圈光線， 是由圍繞著黑洞的 等離子體引力透鏡產生的。 換句話說， 這個黑洞，在背後明亮物質的襯托下， 留下一個圓形的暗影。 而它周圍那明亮的光環 指示了黑洞邊境的位置。 在這裡，引力作用變得無比巨大， 大到就連光線都無法逃離。 愛因斯坦用公式推測了 這個環的大小和形狀， 所以，給光環拍照不僅很酷， 還能幫助我們檢驗這些公式在 黑洞周圍的極端環境下是否成立。
However, this black hole is so far away from us, that from Earth, this ring appears incredibly small -- the same size to us as an orange on the surface of the moon. That makes taking a picture of it extremely difficult. Why is that? Well, it all comes down to a simple equation. Due to a phenomenon called diffraction, there are fundamental limits to the smallest objects that we can possibly see. This governing equation says that in order to see smaller and smaller, we need to make our telescope bigger and bigger. But even with the most powerful optical telescopes here on Earth, we can't even get close to the resolution necessary to image on the surface of the moon. In fact, here I show one of the highest resolution images ever taken of the moon from Earth. It contains roughly 13,000 pixels, and yet each pixel would contain over 1.5 million oranges. 不過，這個黑洞離我們太過遙遠， 從地球上看，它非常，非常小—— 大概就和月球上的一個橘子一樣大。 這導致給它拍照變得無比艱難。 為什麼呢？ 一切都源於一個簡單的等式。 由於衍射現象， 我們所能看到的 最小物體是有限制的。 這個等式指出，當想要看到的 東西越來越小時， 望遠鏡需要變得更大。 但即使是地球上功能最強大的 光學望遠鏡， 其分辨率甚至不足以 讓我們得到月球表面的圖片。 事實上，這裡是一張有史以來 從地球上拍攝的最高清的 月球圖片。 它包含約1.3萬個像素， 而每一個像素裡包含超過150萬個橘子。
So how big of a telescope do we need in order to see an orange on the surface of the moon and, by extension, our black hole? Well, it turns out that by crunching the numbers, you can easily calculate that we would need a telescope the size of the entire Earth. If we could build this Earth-sized telescope, we could just start to make out that distinctive ring of light indicative of the black hole's event horizon. Although this picture wouldn't contain all the detail we see in computer graphic renderings, it would allow us to safely get our first glimpse of the immediate environment around a black hole. 所以，我們需要多大的望遠鏡 才能看到月球表面的橘子， 以及，那個黑洞呢？ 事實上，通過計算， 我們可以輕易得出所需的 望遠鏡的大小， 就和整個地球一樣大。 而如果我們能夠建造出這個 地球大小的望遠鏡， 就能夠分辨出那指示著視界線的 獨特的光環。 儘管在這張照片上，我們無法看到 電腦合成圖上的那些細節， 它仍可以讓我們對於 黑洞周圍的環境有個大致的瞭解。
However, as you can imagine, building a single-dish telescope the size of the Earth is impossible. But in the famous words of Mick Jagger, "You can't always get what you want, but if you try sometimes, you just might find you get what you need." And by connecting telescopes from around the world, an international collaboration called the Event Horizon Telescope is creating a computational telescope the size of the Earth, capable of resolving structure on the scale of a black hole's event horizon. This network of telescopes is scheduled to take its very first picture of a black hole next year. Each telescope in the worldwide network works together. Linked through the precise timing of atomic clocks, teams of researchers at each of the sites freeze light by collecting thousands of terabytes of data. This data is then processed in a lab right here in Massachusetts. 但是，正如你預料， 想建造一個地球大小的射電望遠鏡 是不可能的。 不過，米克·賈格爾有一句名言： “你不可能永遠心想事成， 但如果你嘗試了，說不定就 正好能找到 你所需要的東西。” 通過將遍佈全世界的望遠鏡 連接起來， “視界線望遠鏡”， 一個國際合作項目，誕生了。 這個項目通過電腦製作一個 地球大小的望遠鏡， 能夠幫助我們找到 黑洞視界線的結構。 這個由無數小望遠鏡構成的網絡 將會在明年拍下它的 第一張黑洞圖片。 在這個網絡中，每一個望遠鏡 都與其他所有望遠鏡一同工作。 通過原子鐘的準確時間相連， 各地的研究團隊們通過收集 上萬千兆字節的數據來定位光線。 接下來，這份數據會在 麻省的實驗室進行處理。
So how does this even work? Remember if we want to see the black hole in the center of our galaxy, we need to build this impossibly large Earth-sized telescope? For just a second, let's pretend we could build a telescope the size of the Earth. This would be a little bit like turning the Earth into a giant spinning disco ball. Each individual mirror would collect light that we could then combine together to make a picture. However, now let's say we remove most of those mirrors so only a few remained. We could still try to combine this information together, but now there are a lot of holes. These remaining mirrors represent the locations where we have telescopes. This is an incredibly small number of measurements to make a picture from. But although we only collect light at a few telescope locations, as the Earth rotates, we get to see other new measurements. In other words, as the disco ball spins, those mirrors change locations and we get to observe different parts of the image. The imaging algorithms we develop fill in the missing gaps of the disco ball in order to reconstruct the underlying black hole image. If we had telescopes located everywhere on the globe -- in other words, the entire disco ball -- this would be trivial. However, we only see a few samples, and for that reason, there are an infinite number of possible images that are perfectly consistent with our telescope measurements. However, not all images are created equal. Some of those images look more like what we think of as images than others. And so, my role in helping to take the first image of a black hole is to design algorithms that find the most reasonable image that also fits the telescope measurements. 那麼，這一項目到底是 怎麼運作的呢？ 大家是否記得，如果要看到 銀河系中心的那個黑洞， 我們需要一個地球大小的望遠鏡？ 現在，先假設我們可以 將這個望遠鏡建造出來。 這可能有點像是把地球變成 一個巨大的球形迪斯科燈。 每一面鏡子都會收集光線， 然後，我們就可以將這些光線 組合成圖片。 但是，現在，假設我們將 大多數鏡子移走， 只有幾片留了下來。 我們仍可以嘗試將信息合成圖片， 但現在，圖片中有很多洞。 這幾片留下來的鏡子就代表了 地球上的幾處天文望遠鏡。 這對於製成一張圖片來說， 還遠遠不夠。 不過，儘管我們只在寥寥幾處 地方收集光線， 每當地球旋轉時，我們便可以 得到新的信息。 換言之，當迪斯科球旋轉時， 鏡子會改變位置， 而我們就可以看到圖片的各個部分。 我們開發的生成圖片的算法 可以將迪斯科球上的空缺部分填滿， 從而建造出隱藏的黑洞圖片。 如果我們能在地球上每一處 都裝上望遠鏡， 或者說能有整個迪斯科球， 那麼這個算法並不算重要。 但現在我們只有少量的樣本， 所以，可能有無數張圖像 符合望遠鏡所測量到的信息。 但並不是每一張圖片都一樣。 有些圖片，比其他一些 看起來更像我們想象中的圖片。 所以我在拍攝黑洞 這一項目中的任務是， 開發一種既可以找到最合理圖像， 又能使圖像符合望遠鏡 所測量到的信息的算法。
Just as a forensic sketch artist uses limited descriptions to piece together a picture using their knowledge of face structure, the imaging algorithms I develop use our limited telescope data to guide us to a picture that also looks like stuff in our universe. Using these algorithms, we're able to piece together pictures from this sparse, noisy data. So here I show a sample reconstruction done using simulated data, when we pretend to point our telescopes to the black hole in the center of our galaxy. Although this is just a simulation, reconstruction such as this give us hope that we'll soon be able to reliably take the first image of a black hole and from it, determine the size of its ring. Although I'd love to go on about all the details of this algorithm, luckily for you, I don't have the time. 就像法醫素描師通過有限的信息， 結合自己對於人臉結構的認知 畫出一張畫像一樣， 我正在開發的圖片算法， 是使用望遠鏡提供的有限數據 來生成一張看起來像是 宇宙裡的東西的圖片。 通過這些算法，我們能從散亂 而充滿干擾的數據中 合成一張圖片。 這裡是一個用模擬數據 進行重現的例子： 我們假設將望遠鏡指向 銀河系中心的黑洞。 儘管這只是一個模擬，像這樣的 重建工作給了我們 真正給黑洞拍攝可行照片的希望， 之後便可以決定其光環的大小。 雖然我很想繼續描繪 這個算法的細節， 但你們很幸運，我沒有這個時間。
But I'd still like to give you a brief idea of how we define what our universe looks like, and how we use this to reconstruct and verify our results. Since there are an infinite number of possible images that perfectly explain our telescope measurements, we have to choose between them in some way. We do this by ranking the images based upon how likely they are to be the black hole image, and then choosing the one that's most likely. 可我仍然想大概讓你們瞭解一下 我們是怎樣定義宇宙的樣子， 以及是怎樣以此來重建 和校驗我們的結果的。 由於有無數種可以完美解釋 望遠鏡測量結果的圖片， 我們需要找到一個方式進行挑選。 我們會按照這些圖片是 真正黑洞圖片的可能性進行排序， 然後選出可能性最高的那一張。
So what do I mean by this exactly? Let's say we were trying to make a model that told us how likely an image were to appear on Facebook. We'd probably want the model to say it's pretty unlikely that someone would post this noise image on the left, and pretty likely that someone would post a selfie like this one on the right. The image in the middle is blurry, so even though it's more likely we'd see it on Facebook compared to the noise image, it's probably less likely we'd see it compared to the selfie. 我這話到底是什麼意思呢？ 假設我們正在建立一個能夠 指出一張圖出現在臉書上的 可能性的模型。 我們希望這個模型能指出 不太可能有人會上傳最左邊的圖像， 而像右邊那樣的自拍照 畫出一張圖片一樣， 中間那張圖有點模糊， 所以它被髮表的可能性 比左邊的噪點圖像大， 但比右邊自拍發表的可能性要小。
But when it comes to images from the black hole, we're posed with a real conundrum: we've never seen a black hole before. In that case, what is a likely black hole image, and what should we assume about the structure of black holes? We could try to use images from simulations we've done, like the image of the black hole from "Interstellar," but if we did this, it could cause some serious problems. What would happen if Einstein's theories didn't hold? We'd still want to reconstruct an accurate picture of what was going on. If we bake Einstein's equations too much into our algorithms, we'll just end up seeing what we expect to see. In other words, we want to leave the option open for there being a giant elephant at the center of our galaxy. 但是當模型的主角變成 黑洞的照片時， 一個難題出現了：我們從未 見過真正的黑洞。 在這樣的情況下， 什麼樣的圖才更像黑洞， 而我們又該怎樣假設黑洞的結構呢？ 我們或許能夠使用模擬試驗 得出的圖片， 比如《星際穿越》裡的那張黑洞圖。 但這樣做可能會引起 一些嚴重的問題。 如果愛因斯坦的理論是錯的怎麼辦？ 我們仍然想要得到一張 準確而真實的圖片。 而如果我們在算法中摻入太多 愛因斯坦的公式， 最終只會看到我們所希望看到的。 換句話說，我們想保留在銀河系中心 看到一頭大象這樣的可能性。
Different types of images have very distinct features. We can easily tell the difference between black hole simulation images and images we take every day here on Earth. We need a way to tell our algorithms what images look like without imposing one type of image's features too much. One way we can try to get around this is by imposing the features of different kinds of images and seeing how the type of image we assume affects our reconstructions. If all images' types produce a very similar-looking image, then we can start to become more confident that the image assumptions we're making are not biasing this picture that much. 不同類型的照片擁有 完全不同的特徵。 我們可以輕鬆分辨出 一張黑洞模擬圖 和我們日常拍的照片的差別。 我們需要在不過度提供某類圖片 特徵的情況下， 告訴我們的算法，一張正常的圖片 應該是什麼樣。 做到這一點的一種方法是， 向算法展示擁有不同特徵的圖片， 然後看看這些圖片會怎樣 影響重建的結果。 如果不同類型的圖片都產生出了 差不多的圖像， 那麼我們便可以更有信心了， 我們對圖片的假設並沒有 導致結果出現太大偏差。
This is a little bit like giving the same description to three different sketch artists from all around the world. If they all produce a very similar-looking face, then we can start to become confident that they're not imposing their own cultural biases on the drawings. One way we can try to impose different image features is by using pieces of existing images. So we take a large collection of images, and we break them down into their little image patches. We then can treat each image patch a little bit like pieces of a puzzle. And we use commonly seen puzzle pieces to piece together an image that also fits our telescope measurements. 這就有點像讓來自不同國家的 三個法醫素描師 根據同樣的文字描述來作畫。 如果他們畫出的臉都差不多， 那麼我們就能比較確信， 他們各自的文化背景 並沒有影響到他們的畫。 將不同圖片的特徵賦予 （算法）的一個方法 就是使用現有的圖片的碎片特徵。 所以，我們將大量的圖像 分解成無數小圖片， 然後像拼圖一樣處理這些小圖片。 我們用其中常見的拼圖碎片 來組合成一張 符合望遠鏡所測量數據的完整圖片。
Different types of images have very distinctive sets of puzzle pieces. So what happens when we take the same data but we use different sets of puzzle pieces to reconstruct the image? Let's first start with black hole image simulation puzzle pieces. OK, this looks reasonable. This looks like what we expect a black hole to look like. But did we just get it because we just fed it little pieces of black hole simulation images? Let's try another set of puzzle pieces from astronomical, non-black hole objects. OK, we get a similar-looking image. And then how about pieces from everyday images, like the images you take with your own personal camera? Great, we see the same image. When we get the same image from all different sets of puzzle pieces, then we can start to become more confident that the image assumptions we're making aren't biasing the final image we get too much. 不同類型的圖片擁有 完全不同的拼圖碎片。 所以，當我們使用相同的數據和 截然不同的拼圖類型來 重現圖像時，會發生什麼呢？ 我們先從黑洞模擬類的拼圖開始。 這張圖看起來還比較合理。 它比較符合我們預料中黑洞的樣子。 但我們得到這個結果 是否僅僅是因為我們拿的是 黑洞模擬拼圖呢？ 我們再來試試另一組拼圖， 這組拼圖由宇宙中不是黑洞的 各種天體構成。 很好，我們得到了一幅相似的圖片。 那如果我們拿日常照片的拼圖 會怎麼樣呢， 就像你每天拿自己的相機 拍的那種照片？ 太好了，我們看到了和之前 一樣的圖像。 當我們通過不同類型的拼圖 得出一樣的圖片時， 我們就有充足的自信說 我們對圖片進行的推測， 並沒有引起最終結果的太大偏差。
Another thing we can do is take the same set of puzzle pieces, such as the ones derived from everyday images, and use them to reconstruct many different kinds of source images. So in our simulations, we pretend a black hole looks like astronomical non-black hole objects, as well as everyday images like the elephant in the center of our galaxy. When the results of our algorithms on the bottom look very similar to the simulation's truth image on top, then we can start to become more confident in our algorithms. And I really want to emphasize here that all of these pictures were created by piecing together little pieces of everyday photographs, like you'd take with your own personal camera. So an image of a black hole we've never seen before may eventually be created by piecing together pictures we see all the time of people, buildings, trees, cats and dogs. Imaging ideas like this will make it possible for us to take our very first pictures of a black hole, and hopefully, verify those famous theories on which scientists rely on a daily basis. 我們能做的另一件事是， 用同一組拼圖， 比如源自日常圖片的那一種， 來得到不同類型的源圖片。 所以，在我們的模擬試驗中， 我們假設黑洞看起來像一個 非黑洞天體， 以及在銀河系中心的一頭大象。 當下面一排算法算出的圖片 看起來和上面一排實際圖片 十分相似時， 我們就能對我們的算法 有更多信心了。 在這裡我想強調， 此處所有的圖片都是由 拼接日常照片而得出的， 就像你自己用相機拍的照片一樣。 所以，一張我們從未見過的 黑洞的照片， 最終卻可能由我們日常 熟悉的圖片構成： 人，樓房，樹，小貓，小狗…… 想象這樣的想法使拍攝第一張 黑洞的圖片成為可能， 同時使我們有望校驗 科學家們每天所依靠的著名理論。
But of course, getting imaging ideas like this working would never have been possible without the amazing team of researchers that I have the privilege to work with. It still amazes me that although I began this project with no background in astrophysics, what we have achieved through this unique collaboration could result in the very first images of a black hole. But big projects like the Event Horizon Telescope are successful due to all the interdisciplinary expertise different people bring to the table. We're a melting pot of astronomers, physicists, mathematicians and engineers. This is what will make it soon possible to achieve something once thought impossible. I'd like to encourage all of you to go out and help push the boundaries of science, even if it may at first seem as mysterious to you as a black hole. 但是，要想讓如此充滿想象力的 點子實際工作， 離不開這些我有幸一同工作的 出色的研究者團隊。 我仍然對此感到振奮： 雖然在項目開始時我沒有任何 天文學背景知識， 我們通過這一獨特合作 所達成的成就， 可能導致世界上第一幅 黑洞照片的誕生。 像視界線望遠鏡這樣大項目的成功 是由來自不同學科的人們 用他們各自的專業知識， 一起創造的結果。 我們是一個由天文學家，物理學家， 數學家和工程學家構成的大熔爐。 這就是我們能夠很快達成 一個看起來不可能達成的 成就的原因。 在此我想鼓勵你們所有人，走出去， 推動科學的邊際， 儘管剛開始它看起來可能 和一個黑洞一樣神秘。
Thank you. 謝謝大家。
閱讀更多 TED演講學習 的文章
