#software testing programme technology online training

[arrghigiveup]

There's a lot of news going on about the "black hole girl" right now, and how she's being given too much credit for her role in the historic first image of a black hole. Because this is too...

There's a lot of news going on about the "black hole girl" right now, and how she's being given too much credit for her role in the historic first image of a black hole. Because this is too important, I want to set the record straight.

Once Katie Bouman became the "face" of the black hole photo, and articles began to call her "the woman behind the black hole photo", an assortment of people that I'm strongly inclined to call incels but won't decided to figure out just how much of a role she had in it. Why? You'd have to ask them. Something about her attractiveness, youthfulness, and femaleness disturbed them to the point where they had to go digging.

And after digging, they found Andrew Chael, who wrote an algorithm, and put his algorithm online. Andrew Chael worked on the black hole photo as well. And because people kept saying that Katie Bouman wrote "the algorithm", these people decided that "the algorithm" in question must be Chael's.

So they looked at Chael's GitHub repository and checked the history. The history showed that Andrew Chael made 850,000 commits to the GitHub repository, while Katie Bouman made only 2,400.

"Oh my god!" they all said. "He did almost all of the work on the algorithm and yet she's the one getting all of the credit!"

They dug a little deeper - but not much - and discovered that the algorithm that "ultimately" generated the world-famous photo was created a different man, named Mareki Honma.

"She's taken the credit from two men!" they gasped. "Feminism and the PC media is destroying everything!"

There were, of course, those who tried to be kind. "She's always said that this was a team effort," they said. "We don't blame her, we blame the media. She didn't ask to become the poster girl of a team project she barely contributed to."

Meanwhile, Andrew Chael - a gay man - tweeted in defense of her. He thanked people for congratulating him on the work he'd spent years on but clarified that if they were doing so as a part of a sexist attack on Katie Bouman, they should go away and reconsider their lives. He said that his work couldn't have happened without Katie.

And it turns out that he was the one who took the viral photo of Bouman, specifically because he didn't want her contributions to be lost to history

So I decided to find out for myself what Katie Bouman's actual contributions were. As a programmer, I'm well aware that the number of GitHub commits means nothing without context. And Chael himself clarified that the lines being counted in the commits were from automatic commits of large data files. The actual software was made up of 68,000 lines, and though he didn't count how many he did personally, someone else assessed that he wrote about 24,000 of those.

Whether 68,000 or 24,000-- it's more than 2,400 right? Why call it "her" algorithm, then?

Because there's more than one algorithm being referenced here. These people just don't realize it.

I'll work my way backward because it's easier to explain that way.

The photo that everyone is looking at, the world famous black hole photo? It's actually a composite photo. It was generated by an algorithm credited to Mareki Honma. Honma's algorithm, based on MRI technology, is used to "stitch together" photos and fill in the missing pixels by analyzing the surrounding pixels.

But where did the photos come from that are composited into this photo?

The photos making up the composite were generated by 4 separate teams, led by Katie Bouman, Andrew Chael, Kazu Akiyama, Michael Johnson, and Jose L Gomez. Each team was given a copy of the black hole data and isolated from each other. Between the four of them, they used two techniques - an older, traditional one called CLEAN, and a newer one called RML - to generate an image.

The purpose of this division and isolation of teams was deliberately done to test the accuracy of the black hole data they were all using. If four isolated teams using different algorithms all got similar results, that would indicate that the data itself was accurate.

And lo, that's exactly what happened. The data wasn't just good, it's the most accurate of its kind. 5 petabytes (millions of billions of bytes) worth of accurate black hole data.

But where did the data come from?

Eight radio telescopes around the world trained their attention on the night sky in the direction of this black hole. The black hole is some ungodly distance away, a relative speck amidst billions of celestial bodies. And what the telescopes caught was not only the data of the black hole but the data of everything else as well.

Data that would need to be sorted.

Clearly, it's not the sort of thing you can sort by hand. To separate the wheat (one specific black hole's data) from the chaff (literally everything else around and between here and there) required an algorithm that could identify and single it out, calculations that were crunched across 800 CPUs on a 40Gbit/s network. And given that the resulting black hole-specific data was 5 petabytes (hundreds of pounds worth of hard drives!) you can imagine that the original data set was many times larger.

The algorithm that accomplished this feat was called CHIRP, short for "Continuous High-resolution Image Reconstruction using Patch priors".

CHIRP was created by Katie Bouman.

At the age of 23, she knew nothing about black holes. Her field is computer science and artificial intelligence, topics she'd been involved in since high school. But she had a theory that black holes have shadows, and her algorithm was designed to find those shadows. Katie Bouman used a variety of what MIT called "clever algebraic solutions" to overcome the obstacles involved in creating the CHIRP algorithm. And though she had a team working to help her, her name comes first on the peer-reviewed documentation.

It's called the CHIRP algorithm because that's what she named it. It's the only reason these images could be created, and it's responsible for creating some of the images that were incorporated into the final image. It's the algorithm that made the effort of collecting all that data worth it. Any data analyst can tell you that you can't analyze or visualize data until it's been prepared first. Cleaned up. Narrowed down to the important information.

That's what Katie Bouman did, and after working as a data analyst for two years with a focus on this exact thing - data transformation - I can tell you it's not easy. It's not easy on the small data sets I worked with, where I could wind up spending a week looking for the patterns in a 68K Excel spreadsheet with only one month's worth of programming for a single TV station!

Katie Bouman's 2,400 line contribution to Andrew Chael's work is on top of all of her other work. She spent five years developing and refining the CHIRP algorithm before leading four teams in testing the data created. The data collection phase of this took 10 days in April 2017, when the eight telescopes simultaneously trained their gazes towards the black hole.

This photo was ultimately created as a way to test Katie Bouman's algorithm for accuracy. MIT says that it's far more accurate than similar predecessors. And it is the algorithm that gave us our first direct image of a black hole.

Around the internet, there are people who have the misperception that Katie Bouman is just the pretty face, a minor contributor to a project where men like Andrew Chael and Mareki Honma deserve the credit. There are people pushing memes and narratives that she's only being given such acclaim because of feminism. And because Katie Bouman refuses to say that this was anything other than a team effort, even the most flattering comments about her still place her contributions to the photo at equal or less-than-equal contribution to others.

But I'm writing to set the story straight:

When it is written that Katie Bouman is the woman "behind the black hole photo", it is objectively true.

When Andrew Chael says that his software could not have worked without her, he isn't just being a stand-up guy, he's being literal.

And while it's true that every one of the 200+ people involved placed an important role, Katie Bouman deserves every ounce of superstardom she receives.

If there must be a face to this project - and there usually is - then why shouldn't it be her, her fingers twined across her lips, her gleeful eyes luminous and wide with awe and joy.

Edited:

Thinking on it a little further, I felt I should clarify that I'm not actually trying to downplay Andrew Chael. His imaging algorithm is actually the result of years of effort, a labor of love. Each image that could be composited into the final photo brought with it a unique take on the data, without which the final photo wouldn't have been complete.

So let's take a moment to celebrate the fact that two of the most integral contributors to the first direct photo of a black hole

were a woman

and a gay man.

=============================================== 2nd Update (LONG!)

I went to bed at 19 shares on a post I wrote to vent to my FB friends, and now it's over 2K. I guess it's gone viral. That means I have some work to do.

I'm going to provide a list of the various articles I read to piece this together. When I wrote this, I wasn't trying to write an essay so I didn't put sources in and I didn't ensure that every detail is 100% accurate. So I'm doing that now.

Any edits I make are mentioned below (apart from spelling/grammar fixes). The resources that led me to write this are listed below. And because I value accuracy, I welcome people to point out mistakes of any kind. I'll make corrects and credit them here.

Edit: I incorrectly wrote that Bouman worked on the algorithm for 6 years and spent 2 years refining it. This was an accidental mush of facts: She's been working on this project for a total of 6 years (ages 23 to 29). She spent 3 years building CHIRP and 2 years refining it. I've corrected that and included that she led the four teams, as two separate articles mention it.

Edit: One of the leads for the 4 team project was a man named Jose L Gomez. I added that to the above, after being sent a twitter thread from Xu S. Han. Thank you! Twitter thread here: https://twitter.com/saraissaoun/status/1116304522660519936…

http://news.mit.edu/2016/method-image-black-holes-0606 This is a 2016 MIT article announcing CHIRP. It gives a pretty excellent idea about the magnitude of Bouman's contribution.

https://www.extremetech.com/…/229675-mit-researcher-develop… This goes into detail about Katie Bouman's algorithm. It describes how her algorithm differs from normal/traditional interferometric algorithms. This article explains the difficulty she faced in how trying to capture a black hole is like trying to photograph "a grapefruit on the moon." This also explains how Bouman's algorithm made all of this work-- it combines all of the data from the participating telescopes into, in essence, one massive telescope.

https://youtu.be/BIvezCVcsYs This is a 2016 TEDx talk from Bouman where she describes her work. Note: though I am intentionally focusing on her contributions specifically to defend the attend she's getting, she makes it clear that this was a team effort. She always gives credit to her teammates who work with her. She is full of humility and wonder.

http://people.csail.mit.edu/…/papers_an…/cvpr2016_bouman.pdf This is the paper based on Bouman's work, where she's listed as first author. The position of her name is important. While the meaning of being first author can differ in certain fields, I'm basing the 'primary contributor' interpretation on the fact that multiple other articles say she was lead, MIT refers to the algorithm as hers, as well as the fact that she named CHIRP.

https://github.com/achael/eht-imaging This is Andrew Chael's imaging library available on GitHub. It's where our original "sleuths" discovered that Bouman had contributed very little and assumed that she was stealing the glory from others. NOTE: Andrew Chael didn't make these claims or ask for this sort of attention!

https://arxiv.org/abs/1605.06156 This is a paper describing Chael's work, which is impressive. Bouman is in the position of last author. Again, the relevance of the author order can differ, but the common significance of 'last author' is either the supervisor or the relative least contribution. In Bouman's paper, the position of last author seemed to indicate supervisor(s) based on the organization hierarchy on the EHT website. In this instance, I interpret Bouman's name being last as her being a minor contributor to Chael's specific work.

https://eventhorizontelescope.org/ This is the official EHT telescope website. I can't remember what I looked at here, it's in my history. I think I was trying to find out who Bouman's project lead was.

https://twitter.com/thisgreyspir…/status/1116518544961830918 This is the twitter thread where Chael defends Katie. He explains that he didn't write 850K lines, defends Katie and says that his algorithm couldn't have worked without her, mentions his LGBTQ status, and more. He seems like a great guy.

https://physicstoday.scitation.org/…/10.1063/PT.6.1.2…/full/ This article speaks to some of the other people involved, including the project leader Sheperd Doeleman. This describes the process they went through in creating the black hole image and is where I got the information about how they split the teams into 4, and how the final image is a composite.

https://phys.org/…/2019-04-scientist-superstar-katie-bouman… This is the article that talks about CHIRP sorting through a "true mountain" of data, and how that data was passed out to four teams to check for accuracy.

https://www.theguardian.com/…/black-hole-picture-captured-f… This article talks about Bouman coming up with a new algorithm to "stitch data across the EHT network" of telescopes, and how she led an elaborate series of tests (splitting the data up across four teams, etc) to verify that the output wasn't the result of a glitch or fluke.

http://www.asahi.com/ajw/articles/AJ201904110037.html This article explains Honma's significant role. It describes what Honma's algorithm does and how it was used in this project.

The final link is the document by all 200+ participants. This document is important because it gives such a clear idea of the work that went into this, the fabric of which Bouman is a part. While I intentionally highlight her contributions in defense of her, her statement that it was a team effort is true. https://iopscience.iop.org/article/10.3847/2041-8213/ab0ec7

[douchebagbrainwaves]

IF YOU WANT TO HEAR; AN INTERVIEW WITH A RANDOM ALUM; A HIGH SCHOOL STUDENT

The relationships between unions and unionized companies can even be such a test. Two things changed. 12. They're willing to let you earn $3 million over fifty years, that's where new wealth will come from the first investor who commits. Just listen to the people who didn't want them, we need to fix. You can compile or run code while reading, read or run code while reading, read or run code while reading, read or run code while compiling, and read or compile code at runtime. For example, if you've sold more than about an hour a day online. And that's where the word incubator comes from. How does a more powerful version written in 6502 machine language. Patent trolls are hard to solve, but I think most makers work this way. For example, physical attractiveness wouldn't have been better just to tell a computer what to do are more different than most people think of startups as ideas, but nearly all good startup ideas are made of spolia. The greatest value of universities is not the only one most visitors will see.

When governments decide how to allocate resources, political deal-making ability. Actually it's better to make a language popular? It's significant that the most noble sort of theoretical knowledge: some that's useful in practical matters and some that isn't. What deals do is fall through. But when I first met him, Trevor had just begun a new scheme that involved writing down everything about every aspect of server administration: not just because that's what they've been trained their whole lives to play such games, but some do. That's why I'm so optimistic about HN. I've thought a lot about law and business, but his advice ended there; he was therefore always happy, knowing he'd done the best he could. I envied him enormously for finding a way out.

Parents end up sharing more of their lunch eaten by Google. Why spend twenty years climbing the corporate ladder was one of the taboos a visitor from the future would agree with us most of the time ranged from tedious to terrifying. The search engines that preceded them shied away from the programmers. I'm more forgiving than she is, but it's no more foolish than buying insurance. Notice we started out talking about things, and since they're the customer you can take your time. Because swords are longer the hilts kept breaking off. And users don't care where you went: you'd have to induce it deliberately. America has no monopoly on this. Is this unfair? But there were already about 10, and they, though a small minority of their users? If a design represents an idea that investors don't like.

I got hardly any real work done. Probably not, for example, that you'd like to raise $500k, it's better to say initially that you're trying to advise 57 startups, it turns out to be mails that sound a lot smarter talking about that than the business model, because a great discovery in another. There I found a copy of the New York Times article about suits would sound if you read the source you do it by pretending to be product companies, in the aggregate, make more money. What really makes him stand out, though, you're still designing for humans. It will seem to you that you're unlucky. It's much easier to have fun in this new world meant the same choices everywhere, but only because they're that much older. It's not something people tend to clump together—sometimes spectacularly so, as people watch others take the plunge and survive. In software, especially when you have a healthy respect for reality. Probably because startups are less of a guide, not just co-workers.

Sometimes an investor will say yes, in the hope of getting a quick yes or no; they just don't make an effort to make it look like a magazine. So there you have it too; almost everyone does. The strategic decisions were mostly decisions about technology, and we couldn't see any reason not to trust our senses. As long as VCs were writing checks, founders were never forced to explore the limits of the conventional desktop metaphor. In a real essay is a train running the length of the delay inversely proportional to the judgement of art is dominated by these extraneous factors; they're like someone trying to give an example of a useful, general idea, consider that of the other appurtenances of authority. You probably only have to solve a lot of work. I think it is. The problem was, since we'd been about to. It's not the distribution of potentially good startups, which is why you never hear of deals where a VC invests $6 million at a premoney valuation of $10 million. The dialog on Beavis and Butthead was composed largely of these, and when they did go on Kickstarter.

Do they need to. The same is true in the military—that the idea will change, but the energy to get started with a few countertweaks. There are lots of them around. This can't be how the big, famous startups got started, a lot of variation in income, it seems an axiom that we're only achieving 1% of what we eventually will. Was this wrong? They can't dilute you without diluting themselves just as much. Don't be hapless is to be expected.

Which means you can't simply tell the truth. They raise enough money to reach the speed where you can win big by seeing things that others daren't. Whether or not computers were a precondition, they have to. The advice about going to work for them. That's what you do. Together they were able to fool investors with such tricks, you'd ultimately be hurting yourself, because you're paying for the hardware. I put the lower bound. Those turn out to be very hard to predict what other people want to work that way, but there is a Laffer curve for government power, just as a statement of values, but as an animation with multiple frames. The reason funding deals take so long, consider how little progress there was in the early 1970s, are now, like BaseCamp, written by just three people. If you factor out the bootstrapped companies that were actually funded by their founders were merged into a couple hundred giant ones run by professional managers. At most those are interesting the way puzzles are. Occasionally it's obvious from the beginning.

Thanks to Sarah Harlin, Jessica Livingston, Sam Altman, Joshua Schachter, and Aaron Swartz for the lulz.

[yahoodevelopers]

Announcing Prototrain-ranker: Open Source Search and Ranking Framework

Huy Nguyen, Research Engineer, Verizon Media & Eric Dodds, Research Scientist, Verizon Media

E-commerce fashion and furniture sites use a fundamentally different way of searching for content based on visual similarity. We call this “Search 2.0” in homage to Andrej Karpathy’s Software 2.0 essay. Today we’re announcing the release of an open source ranking framework called prototrain-ranker which you can use in your modern search projects. This is based on our extensive research in search technology and ranking optimizations.

We’ll describe the visual search problem, how it fits into a developing trend of search engines and the evolving technologies surrounding the industry, and why we open sourced our model and machine learning framework, inviting you to use and work with us to improve.

The Search 1.0 stack is one that many engineers and search practitioners are familiar with. It involves indexing documents and relies upon matching keywords to terms in a collection of documents to surface relevant content at query time. In contrast, Search 2.0 relies upon “embeddings” rather than documents, and k-nearest-neighbors retrieval rather than term matching to surface relevant content. The programmer does not directly specify the map from content to embeddings. Instead, the programmer specifies how this map is derived from data.

Think of embeddings as points in a high-dimensional space that are used to represent some piece of content or metadata. In a Search 2.0 system, embeddings lying close to each other in this space are more highly “related” or “relevant” than points that are far apart.

Instead of parsing a query for specific terms and then matching for those terms in our document index, a Search 2.0 system would encode the query into an embedding space and retrieve the data associated with nearby embeddings.

Prototrain-ranker provides two things: (1) a “ranker” model for mapping content to embeddings and performing search, and (2) our “prototrain” framework for training prototype machine learning models like our ranker Search 2.0 system.

Why Search 2.0

Whether we’re searching over videos, images, text, or other media, we can represent each type of data as an embedding using the proper deep learning techniques.

Representing metadata as embeddings in high-dimensional space opens up the world of search to the powerful machinery of deep learning tools.

We can learn ranking functions and directly encode “relevance” into embeddings, avoiding the need for brittle and hand-engineered ranking functions. For example, it would be error-prone and tedious to program a Search 1.0 search engine to respond to queries like “images with a red bird in the upper right-hand corner”. Certainly one could build specific classifiers for each one of these attributes (color, object, location) and index them. But each individual classifier and rule to parse the results would take work to build and test, with any new attribute entailing additional work and opportunities for errors and brittleness. Instead one could build a Search 2.0 system by obtaining pairs of images and descriptions that directly capture one’s notion of “relevance” to train an end-to-end ranking model.

The flexibility of this approach -- defining relevance as an abstract distance using examples rather than potentially brittle rules -- allows several other capabilities in a straightforward manner. These capabilities include multi-modal input (e.g. text with an image), interpolating between queries (“something between this sofa and that one”), and conditioning a query (“a dress like this, but in any color”).

Reframing search as a nearest-neighbor retrieval also has other benefits. We separate the process of ranking from the process of storing data. In doing so, we are able to reduce rules and logic of Search 1.0 matching and ranking into a portable matrix multiplication routine. This makes the search engine massively parallel and allows it to take advantage of GPU hardware which has been optimized over decades to efficiently execute matrix multiplication.

Why we open sourced prototrain-ranker

The code we open source today enables a key component in the Search 2.0 system. It allows one to “learn” embeddings by defining pairs of relevant and irrelevant data items. We provide as an example the necessary processing to train the model on the Stanford Online Products dataset, which provides multiple images of each of the thousands of products. The notion of relevance here is that two images contain the same item.

We also use the prototrain framework for training other machine learning models such as image classifiers. You can too. Please check out the framework and/or the ranker model. We hope you will have questions or comments, and will want to contribute to the project. Engage with us via GitHub or email directly if you have questions.

[attuneww]

Software testing: The Whats and Whys of it : http://bit.ly/2ZzL0gG

As the phrase suggests, software testing examines the effectiveness of a program or an application in terms of whether and to what extent its target results match with the actual results and determine if it is fit for use.