Infoman

Member since 2017-07-15T03:50:57Z. Last seen 2025-08-21T09:12:55Z.

2756 blog posts. 128 comments.

Followers: 0 Following: 2

Posts by Infoman

• 
  

  Infoman

  Differences Between Interactive and Linear Narrative

  At their core, interactive and linear narratives are quite different because of the difference in the role of the audience versus the player. Though an audience member of course brings her own background and interpretations to any media that she consumes, she is still unable to change the actual media itself, only her perce ption thereof. However, a player is constantly affecting the media in which she is taking part, and therefore a player has actual agency in the interactive narratives that she experiences. This means that authors of interactive narrative must be aware of some core differences in how they can craft their narratives.

  Plot Versus Free Will

  One of the most difficult things to give up when crafting interactive narratives is control over the plot. Both authors and readers/viewers are accustomed to plots with elements like fore-shadowing, fate, irony, and other ways in which the intended outcome of the plot actually influences earlier parts of the story. In a truly interactive experience, this would be impossible because of the free will of the player. Without knowing what choices the player will make, it is very difficult to intentionally foreshadow the results of those choices. There are several possibilities for dealing with this dichotomy, some of which are used often and others of which are used in situations like pen-and-paper RPGs but have not yet been implemented in many digital games

  Limited possibilities

  Limited possibilities are actually a part of nearly all interactive narrative experiences. In fact, most games, at their inscribed level, are not actually interactive narratives. all the most popular series of games over the past decade (Prince ofpersio. ca ofDuty, Halo, Uncharted, and so on) have exclusively linear stories at their core. No matter what you do in the game, your choices are to either continue with the narrative or quit the game. In fact, Spec Ops The Line by Yager Development explored this issue beautifully, placing the player and the main character of the story in the same position of having only two real choices : continue to perform increasingly horrific acts orjust stop playing the game.

  In Prince ofpersio The Sands ofTime, this is handled by having the narrator (the prince of the title and the protagonisti say"No. No, no that's not the way it happened, Shall l start again?"whenever the player dies and the game has to back up to the most recent check Point. In the Assossin Creed series, this is handled by stating that you have become "desynchronized"from your ancestor's story if (through lack of player skill) the ancestor is allowed to die.

  There are also several examples of games that limit choices to only a few possibilities and base those on the player's actions throughout the game. Both Fable, by Lionhead Studios, and St(7r Wars Kni夕hts ofthe O/dRepub/ic, by Bioware, claimed to be watching the player throughout the game to determine the final game outcome, but though each did track the player on a good versus evil scale throughout the game, in both cases (as in many other games), a single choice made at the end of the game could override an entire game of good or evil behavior.

  Other games like the Japanese RPGs Final Fantasy V// and Chrono Trigger have more subtle and varied possibilities. In Final Fan tosy W, there is a point where the main character, Cloud, goes on a date with someone at the Golden Saucer amusement part. The default is for Cloud to go out with Aeris ; however, if the player has ignored Aeris throughout the game and kept her out of their battle party, Cloud will instead go out with Tifa. The possibilities for the date are the characters Aeris, Tifa, Yuffie, and Barrett, although it takes resolute effort to have the date with Barrett. The game never explains that this math is happening in the background but it is always there, and the Final Fantasy team used a similar strategy in Fino/Fantosyxto determine who the protagonist, Tidus, would ride on a snowmobile with in a romantic scene. Chrono Trigger uses several metrics to determine which of the game's thirteen endings to choose (and some of those endings have multiple possibilities within them). Again, the calculations for this are largely invisible to the player.

  AIIow the player to choose from several linear side quests

  Many of Bethesda Softwork's open-world games use this strategy, including the recent games Fa/tout 3 and Skyrim. While the main quest is generally pretty linear for these games, it is only a small fraction of the game's total content. In Skyrim, for instance, the main quest takes about 12 to 16 hours to complete, but the game has over 400 hours of additional side quests. A player's reputation and history in the game lead to some side quests being unlocked and exclude her from playing others. This means that each individual who plays the game has the potential to have a different combination of linear experiences that add up to a different overall game experience from other players.

  Foreshadowing multiple things

  If you foreshadow several different things that might happen, some of them probably will happen. players will generally ignore the foreshadowing that is not paid off while noticing that which does. This happens often in serial television shows where several possibilities for future plots are put in place but only a few are ever actually executed (e. g., the Nebari plot to take over the universe that is revealed in the Far-scope episode"A clockwork Nebari"and the character of the Doctor's daughter from the Doctor who episode"The Doctor's Daughter"who never returns to the show).

  Develop minor nonplayer characters (Npcs) into major ones

  This is a tactic used often by game masters of pen-and-paper RpGs. An example of this would be if the players were attacked by a group of ten bandits, and the players defeated the bandits, but one got away. The game master (GM) could then choose to have that bandit return at some point with a vendetta against the players for killing his friends. This differs significantly from games like Final Fantasy vl (originally titled Final Fantasy m in the U. S.), where it is pretty obvious from early in the game that Kefka will be a recurring, annoying, and eventually wholly evil nemes is character. Though the characters in the player's party don't realize this, just the fact that the developers chose to give Kefka a special sound effect for his laugh makes it apparent to the player.

  Tip

  Pen-and-paper RPGs still offer players a unique interactive gaming experience, and I highly recommend them. In fact, when I taught at the University of Southern California, I required all of my students to run an RPG and play in a couple run by their peers. Roughly 40% of the students each semester listed it as their favorite assignment.

  Because pen-and-paper RPGs are run by a person, that game master (GM) can craft the narrative in real time for the players in a way that computers have yet to match. All of the strategies listed earlier are used by GMs to guide their players and make their experiences seem fated, foreshadowed, or ironic in ways that are usually reserved for linear narrative. The perennial RPG Dungeons & Dragons, by Wizards of the Coast, is a good place to get started, and there are a tremendous number of source books for it. However, I have found that D & D campaigns tend to be very combat focused. For an experience that allows you to most easily create and experience interactive stories, I recommend the FATE system by Evil Hat Productions.

  2017/10/15
• 
  

  Infoman

  Game Design - Inscribed Narrative - premise, setting, character, and plot

  In both linear and interactive narrative, the components of the dramatics are the same premise, setting, character, and plot.

  ■ Premise The premise is the narrative basis from which the story emerges:

  A long time ago in a galaxy far, far away, an intergalactic war is brought to the doorstep of a young farmer who doesn't yet realize the importance of his ancestry or himself. Gordon Freeman has no idea about the surprises that are in store for him on his first day of work at the top secret Black Mesa research facility. Edward Kenway must fight and pirate his way to fortune on the high seas of the Carib- bean while discovering the secret of the mysterious Observatory, sought by Templars and Assassins alike.

  ■ Setting The setting expands upon the skeleton of the premise to provide a detailed world in which the narrative can take place. The setting can be something as large as a galaxy far, far away or as small as a tiny room beneath the stairs, but it's important that it is believable within the bounds ofthe premise and that it is internally consistent ifyour characters will choose to fight with swords in a world full of guns, you need to have a good reason for it. In Stor Wars, when Obi Wan Kenobi gives the light saber to Luke, he answers all of these questions by stating that it is"not as clumsy or random as a blaster, an elegant weapon for a more civilized age

  ■ Character : Stories are about characters, and the best stories are about characters we care about. Narratively, characters are composed of a backstory and one or more objectives. These combine to give the character a role in the story protagonist, antagonist, companion, lackey, mentor, and so on.

  ■ Plot Plot is the sequence of events that take place in your narrative. Usually, this takes the form of the protagonist wanting something but having difficulty achieving it because of either an antagonist or an antagonistic situation getting in the way. The plot then becomes the story of how the protagonist attempts to overcome this difficulty or obstruction.

  2017/10/15
• 
  

  Infoman

  3D game piece printing

  One of the most exciting recent technological advancements for board game prototyping is 3D printing, and many board game designers are starting to print pieces for their game Prototypes. There are also several companies online now that will print your game board, cards, or pieces. There are also aspects of touch in digital games. The way that the controller feels in a Player's hands and the amount of fatigue that it causes are definitely aspects that the designer needs to consider. when the fantastic playstation 2 game okami was ported to the Nintendo Wii, the designers chose to change the attack command from a button press (the X on the Playstation controller) to a waggle of the wiimote (which mimicked the attacl gesture from The Legend ofzelda Twilight princess that had done very well on the wii).

  However, while attacks in the heat of battle in Twilight princess happen about once every couple of seconds, attacks in Okami happen several times per second, so the attack gesture that worked well in TwM 夕 ht Princess instead caused player fatigue in okami. with the rise of tablet and smartphone gaming, touch and gesture are elements that every game designer must consider carefully. Anothe, aspect of touch in digital games is'"mble-style player fgedback. As a designer. diffe'"nt book printing processes have di«erent smell* so too do diffgrent board and card game p, inting processes. Make su'"that you get a sample from you'"anufacturer before

  2017/10/15
• 
  

  Infoman

  Steve Wozniak announces tech education platform Woz U

  530 nh 1 day 106 https://techcrunch.com/2017/10/13/steve-wozniak-announces-tech-education-platform-woz-u/ news.ycombinator.com/item?id=15466599

  Steve Wozniak, the Apple cofounder who changed the world alongside Steve Jobs, has today announced the launch of Woz U,

  According to the release, Woz U will start out as an online learning platform focused on both students and companies that will eventually hire those students. Woz U is based out of Arizona, and hopes to launch physical locations for learning in more than 30 cities across the globe.

  At launch, the curriculum will center around computer support specialists and software developers, with future courses on data science, mobile applications and cybersecurity coming in the future.

  Alongside the education platform, Woz U will also offer platforms for tech companies to recruit, train and retain their workforce through on-site customized programs and subscription-based curricula. There will also be a platform for K-12 students, which will be distributed to school districts, that will offer STEAM programs to identify talent and nudge those individuals into a tech-based career.

  And if that weren’t enough, Woz U will eventually introduce an accelerator program ‘to identify and develop elite tech talent.’

  Here’s what Woz had to say in a prepared statement:

  Our goal is to educate and train people in employable digital skills without putting them into years of debt. People often are afraid to choose a technology-based career because they think they can’t do it. I know they can, and I want to show them how. My entire life I have worked to build, develop and create a better world through technology and I have always respected education. Now is the time for Woz U, and we are only getting started.

  Woz U will offer an app to help people understand which field of tech they’re best suited for, so they can set up their curriculum accordingly.

  As of yet, there’s no word on pricing.

  2017/10/15
• 
  

  Infoman

  Google commits $1B in grants to train U.S. workers for high-tech jobs

  327 thesanerguy 14 hrs 302

  https://techcrunch.com/2017/10/12/google-commits-1-billion-in-grants-to-train-u-s-workers-for-high-tech-jobs/

  http://news.ycombinator.com/item?id=15460354

  The nature of work is changing on a global level at a rapid pace. Sure, it’s not the first time work has been dramatically impacted by technology, but the growth of automation, robotics, AI and the like have the potential to displace jobs at an unprecedented rate. And Google will almost certainly be one of the driving forces behind that transformation.

  The search giant has regularly expressed a desire to help stem some of that negative impact, and now it’s putting its money where its mouth is to the tune of $1 billion. CEO Sundar Pichai announced Grow with Google at an event earlier today in Pittsburgh, PA. Over the next five years, the initiative will commit $1 billion to nonprofits aimed at training American workers and helping build business.

  The location of the event will not be lost on anyone who has followed Pittsburgh’s growth over the last few decades. The Steel City has long served as an ideal example of an economy that’s rebounded from the brink of disaster. In Pittsburgh’s case, technology was a primary driver, thanks to Carnegie Mellon, which has helped transform it from post-Rust Belt depression to one of the country’s leading tech hubs. These days, the walls of Pittsburgh’s former factories house cutting-edge innovations in fields like robotics and autonomous driving.

  Pichai noted in his address that the city also holds special meaning for him. “It was the first city I saw in America when I came here 24 years ago,” he explained. “I was here before the internet really took off, but the city was already changing. The number of high-tech jobs had doubled.”

  The company is committing $10 million to Goodwill as part of the initiative — the largest Google.org has committed to one organization. That money will be used to help launch the Goodwill Digital Career Accelerator, aimed at preparing the American workforce for high-tech jobs. Grow with Google also will take the form of a national tour hosted by libraries and community organizations aimed at bringing training and career advice directly to local towns and cities. That’s part of the company’s goal of committing one million hours to employee volunteering over the next five years.

  “At Google, our mission is to make sure that information serves everyone, not just a few,” Pichai explained in the address. “A child in a school here in Pittsburgh can access the same information on Google as a professor at Carnegie Mellon. In the end, the internet is a powerful equalizer, capable of propelling new ideas and people forward.”

  Additional online training can be found at the Grow with Google hub.

  Featured Image: LLUIS GENE/AFP/Getty Images

  2017/10/13
• 
  

  Infoman

  投資不應預設立場

  投資市場有自己的一套定律：

  投資市場沒有永遠的明星，盈利是關鍵。投資市場不應預設立場，如果只懂買升，而不是對沖操作，盈利不斷改善的股票就是好股票，即是賺錢的股票。

  作為一位絕對回報操作者，也不能完全解讀市場。只講亮麗的說話卻不說出風險，其實也是不負責任。我有朋友，重傷出場；幸存者一些可把資金曲線不斷上升，從複式增長當中製造更多財富；也有炒爆戶口的，50歲後重新上路。也有一位要好的明星朋友和我說道，外表光鮮亮麗的日子他嘗試過，結果由負25萬美元重新上路、踏實的工作（電視劇及拍戲），可預計能承受風險的投資，有適當的生活調劑，達到生活財富兼顧的境界。 3）世界市值最大的公司蘋果進入一個iPhone「10年新時代」，股票最高價由164.94美元回落至156.55美元。我依然深信蘋果、亞馬遜、Facebook是極佳的顛覆性公司。亞馬遜的生意模式經過了20年，由最初的實體書送運，到電子書及網上及實體貨倉的儲存及零售「一條龍」，可說已達至「無量境界」。至於Facebook，喜歡它與否，盡在不言中。 我的投資方法包含了對沖策略，不會是「齋long」做長倉。我期待11月2號蘋果的業績公佈，也認為一些對沖基金有機會或者提早年尾收爐，因為一整年ETF的世界已經令他們獲利甜美。Happy trading!

  錢志健 資深對沖基金經理 mdehedgecenter@gmail.com 本欄逢周五刊出

  2017/10/13
• 
  

  Infoman

  Rejecting a candidate for over-qualification results in age bias

  408 KentBeck 8 hrs 258

  https://www.facebook.com/notes/kent-beck/over-qualification-wut/1671791376187053/

  http://news.ycombinator.com/item?id=15461439

  My father was laid off from SRI at 55. He spent the next two years bouncing around the Valley fruitlessly looking for work, lowering his expectations all the while. Eventually he gave up, tucked his tail between his legs, and retired to southern Oregon.

  My father wasn’t rejected for his lack of skills. He could make a 6502 bark, roll over, and shake hands. His code worked. He knew when to clean up, so he was able to add features indefinitely. He hit his deadlines. He could code, he just couldn’t find a job.

  I still carry some of the pain of his situation with me. I’m afraid of “aging out” of technology. I’ve worked to keep myself employable. I’m already seeing responses most easily explained by age-related bias, though, so my time is coming.

  I tweeted this the other day: “You are over-qualified for the position and we don’t think you’d be satisfied” is de facto age discrimination The subsequent conversation was diverse and intense enough that I want to follow up on the many interesting comments. I will make three points:

  Mechanism, not equivalence Effect, not motivation Thoughts, not words One set of responses to the tweet related how non-old people can also hear this response. To be clear, I didn’t say that citing over-qualification is age discrimination. There are legitimate reasons to cite over-qualification. I’m saying that citing over-qualification is also used as the mechanism for enforcing age-related bias, just as “lack of culture fit” is used as a mechanism for enforcing gender and cultural bias.

  Another set of responses speculated on the motivation of those citing over-qualification: fear of managing someone better, fear of being called out on bullshit, fear of salary pressure. You know what? I just don’t care. While there are certainly explanations for this kind of behavior there are no excuses. Look at the effects of your behavior.

  Compared to when I was a wee little engineer, the list of forbidden questions and phrases is much longer. I’m not interested in adding “over-qualified and likely dissatisfied” to that list. I am interested in addressing the thoughts that lead to that phrase.

  Have you used the phrase? Reflect on your thoughts just before you used it. Reflect on the beliefs that must underlie such thoughts. Hint: this is going to be uncomfortable work, at least if you’re honest.

  My father spent from 57 to 83 on his five acres on a hilltop. He puttered in his ham shack. He ran for county planning commissioner and lost. He cared for my aging mother. He never wrote another line of code.

  Part of that is on him (about which more later), but part is on the system you and I are still part of. The beliefs that prevent experienced programmers from contributing at their full capacity are still out there. It’s up to us to call them out.

  2017/10/13
• 
  

  Infoman

  Where's the Proof That Mindfulness Meditation Works?

  18 helloworld 2 hrs 4

  https://www.scientificamerican.com/article/wheres-the-proof-that-mindfulness-meditation-works1/

  http://news.ycombinator.com/item?id=15463056

  The concept of mindfulness involves focusing on your present situation and state of mind. This can mean awareness of your surroundings, emotions and breathing—or,

  more simply, enjoying each bite of a really good sandwich.

  Research in recent decades has linked mindfulness practices to a staggering collection of possible health benefits.

  Tuning into the world around you may provide a sense of well-being, an array of studies claim. Multiple reports link mindfulness with improved cognitive functioning. One study even suggests it may preserve the tips of our chromosomes, which whither away as we age.

  Yet many psychologists, neuroscientists and meditation experts are afraid that hype is outpacing the science. In an article released this week in Perspectives on Psychological Science, 15 prominent psychologists and cognitive scientists caution that despite its popularity and supposed benefits, scientific data on mindfulness is woefully lacking. Many of the studies on mindfulness and meditation, the authors wrote, are poorly designed—compromised by inconsistent definitions of what mindfulness actually is, and often void of a control group to rule out the placebo effect.

  The new paper cites a 2015 review published in American Psychologist reporting that only around 9 percent of research into mindfulness-based interventions has been tested in clinical trials that included a control group. The authors also point to multiple large placebo-controlled meta-analyses concluding that mindfulness practices have often produced unimpressive results. A 2014 review of 47 meditation trials, collectively including over 3,500 participants, found essentially no evidence for benefits related to enhancing attention, curtailing substance abuse, aiding sleep or controlling weight.

  Lead author of the report Nicholas Van Dam, a clinical psychologist and research fellow in psychological sciences at the University of Melbourne, contends potential benefits of mindfulness are being overshadowed by hyperbole and oversold for financial gain. Mindfulness meditation and training is now a $1.1-billion industry in the U.S. alone. “Our report does not mean that mindfulness meditation is not helpful for some things,” Van Dam says. “But the scientific rigor just isn’t there yet to be making these big claims.” He and his co-authors are also concerned that as of 2015, less than 25 percent of meditation trials included monitoring for potential negative effects of the intervention, a number he would like to see grow as the field moves forward.

  Van Dam acknowledges that some good evidence does support mindfulness. The 2014 analysis found meditation and mindfulness may provide modest benefits in anxiety, depression and pain. He also cites a 2013 review published in Clinical Psychology Review for mindfulness-based therapy that found similar results. “The intention and scope of this review is welcome—it is looking to introduce rigor and balance into this emerging new field,” says Willem Kuyken, a professor of psychiatry at the University of Oxford in England, who was not involved in research for the new report. “There are many areas where mindfulness-based programs seem to be acceptable and promising, but larger scale randomized, rigorous trials are needed.”

  Two trials published earlier this month in Science Advances also support mindfulness practices. The first found mindfulness-like attention training reduces self-perceived stress, but not levels of the hormone cortisol, a commonly used biological gauge of stress levels. The other trial links mindfulness-like attention training to increases in thickness of the prefrontal cortex, a brain region associated with complex behavior, decision-making and shaping personality. The authors called for further research into what these findings could mean clinically.

  Van Dam characterizes the research methods used in both of these studies as sound. Yet he points out both also represent the field’s larger problem—a lack of standardization. Varying mindfulness-like approaches have been investigated over the years, making comparisons of different studies difficult.

  Mindfulness is rooted in Buddhist thought and theory. In the West it was popularized in the 1970s by University of Massachusetts professor Jon Kabat-Zinn, a cognitive scientist who founded the university’s Stress Reduction Clinic and the Center for Mindfulness in Medicine. Kabat-Zinn developed what he called “mindfulness-based stress reduction,” an alternative therapy for a variety of often difficult-to-treat conditions. By the early 2000s, the concept of mindfulness had ballooned in popularity. It soon came to have many differing meanings and varying approaches to treatment. “We specifically commented in our article on the fact that many continue to develop novel interventions without fully evaluating those that are already being implemented,” Van Dam says. “I think these studies, while well-designed, may fit within the category of being just different enough from what we already have to prevent us from really knowing whether we could use these results as evidence for [the effectiveness of] other mindfulness-based practices.”

  As Van Dam and his co-authors wrote, “[there is] neither one universally accepted technical definition of ‘mindfulness’ nor any broad agreement about detailed aspects of the underlying concept to which it refers.”

  “Overall, I suspect that a large number of the health promises will not be fulfilled, mostly because therapies, phone apps and other interventions are being rushed to market without sufficiently rigorous testing and appropriate implementation,” he says. “But given what we’ve seen to date, I suspect evidence may accumulate supporting mindfulness practices for anxiety, depression and stress-related conditions.”

  Behavioral and social sciences professor and director of Brown University’s Mindfulness Center Eric Loucks, who was not involved in researching the new paper, agrees there are multiple definitions of mindfulness. But it is the trickiness in bringing a rich spiritual concept into a standardized framework for testing and advising patients that he feels might be tough to tackle.

  “One element in defining mindfulness, if considering its roots in Buddhism, is…the Buddha's recommendation that descriptions of concepts like ‘mindfulness’ are like a finger pointing at the moon,” he explains. “It is important not to confuse the finger for the moon. There will always be variations in people's understanding of mindfulness. It is a personal experience.”

  2017/10/13
• 
  

  Infoman

  Spotify’s Discover Weekly: How machine learning finds new music

  336 kiyanwang 8 hrs 159

  https://hackernoon.com/spotifys-discover-weekly-how-machine-learning-finds-your-new-music-19a41ab76efe

  http://news.ycombinator.com/item?id=15447706

  This Monday — just like every Monday— over 100 million Spotify users found a fresh new playlist waiting for them. It’s a custom mixtape of 30 songs they’ve never listened to before but will probably love. It’s called Discover Weekly, and it’s pretty much magic.

  I’m a huge fan of Spotify, and particularly Discover Weekly. Why? It makes me feel seen. It knows my musical tastes better than any person in my life ever has, and I am consistently delighted by how it satisfies me just right every week, with tracks I myself would never have found or known I would like.

  For those of you who live under a musically soundproof rock, let me introduce you to my virtual best friend:

  A Spotify Discover Weekly playlist — specifically, mine. As it turns out, I’m not alone in my obsession with Discover Weekly—the user base went crazy for it, which has driven Spotify to completely rethink its focus, investing more resources into algorithm-based playlists.

  Ever since Discover Weekly debuted in 2015, I’ve been dying to know how it worked (plus I’m a fangirl of the company, so sometimes I like to pretend I work there and research their products.) After three weeks of mad googling and a great conversation on Spotify’s roof deck with data engineer Nikhil Tibrewal, I feel grateful to have finally gotten a glimpse behind the curtain.

  So how does Spotify do such an amazing job of choosing those 30 songs for each person each week? Let’s zoom out for a second to look at how other music services have done music recommendations, and how Spotify’s doing it better.

  Back in the 2000s, Songza kicked off the online music curation scene using manual curation to create playlists for users. “Manual curation” meant that some team of “music experts” or other curators would put together playlists by hand that they thought sounded good, and then listeners would just listen to their playlists. (Later, Beats Music would employ this same strategy.) Manual curation worked okay, but it was manual and simple, and therefore it couldn’t take into account the nuance of each listener’s individual music taste.

  Like Songza, Pandora was also one of the original players in the music curation scene. It employed a slightly more advanced approach, instead manually tagging attributes of songs. This meant a group of people listened to music, chose a bunch of descriptive words for each track, and tagged the tracks with those words. Then, Pandora’s code could simply filter for certain tags to make playlists of similar-sounding music.

  Around that same time, a music intelligence agency from the MIT Media Lab called The Echo Nest was born, which took a radically more advanced approach to personalized music. The Echo Nest used algorithms to analyze the audio and textual content of music, allowing it to perform music identification, personalized recommendation, playlist creation, and analysis.

  Finally, taking yet another different approach is Last.fm, which still exists today and uses a process called collaborative filtering to identify music its users might like. More on that in a moment.

  So if that’s how other music curation services have done recommendations, how does Spotify come up with their magic engine, which seem to nail individual users’ tastes so much more accurately than any of the other services?

  Spotify actually doesn’t use a single revolutionary recommendation model — instead, they mix together some of the best strategies used by other services to create its own uniquely powerful Discovery engine.

  In 2014, Spotify actually bought The Echo Nest to gain access to their data and algorithms surrounding audio and text analysis, and they also use collaborative filtering algorithms similar to those used at Last.fm.

  Therefore, to create Discover Weekly, there are three main types of recommendation models that Spotify employs:

  Collaborative Filtering models (i.e. the ones that Last.fm originally used), which work by analyzing your behavior and others’ behavior. Natural Language Processing (NLP) models, which work by analyzing text. Audio models, which work by analyzing the raw audio tracks themselves.

  Image credit: Chris Johnson, Spotify Let’s take a dive into how each of these recommendation models work!

  First, some background: When many people hear the words “collaborative filtering”, they think of Netflix, as they were one of the first companies to use collaborative filtering to power a recommendation model, using users’ star-based movie ratings to inform their understanding of what movies to recommend to other “similar” users.

  After Netflix used it successfully, its use spread quickly, and now it’s often considered the starting point for anyone trying to make a recommendation model.

  Unlike Netflix, though, Spotify doesn’t have those stars with which users rate their music. Instead, Spotify’s data is implicit feedback — specifically, the stream counts of the tracks we listen to, as well as additional streaming data, including whether a user saved the track to his/her own playlist, or visited the Artist page after listening.

  But what is collaborative filtering, and how does it work? Here’s a high-level rundown, as encapsulated in a quick conversation:

  Image by Erik Bernhardsson What’s going on here? Each of these two guys has some track preferences — the guy on the left likes tracks P, Q, R, and S; the guy on the right likes tracks Q, R, S, and T.

  Collaborative filtering then uses that data to say,

  “Hmmm. You both like three of the same tracks — Q, R, and S — so you are probably similar users. Therefore, you’re each likely to enjoy other tracks that the other person has listened to, that you haven’t heard yet.”

  It therefore suggests that the guy on the right check out track P, and the guy on the left check out track T. Simple, right?

  But how does Spotify actually use that concept in practice to calculate millions of users’ suggested tracks based on millions of other users’ preferences?

  …matrix math, done with Python libraries!

  In actuality, this matrix you see here is gigantic. Each row represents one of Spotify’s 140 million users (if you use Spotify, you yourself are a row in this matrix) and each column represents one of the 30 million songs in Spotify’s database.

  At the matrix’s intersections, where each user meets each song, there is a 1 if the user has listened to that song, and a 0 if the user hasn’t. So, if I listened to the song “Thriller”, the place where my row meets the column representing “Thriller” is going to be a 1. (Note: Spotify has experimented with using the actual number of streams, vs. a simple 1 vs. 0.)

  Of course, this makes for a very sparse matrix— there are way more songs a given user hasn’t listened to than the ones he/she has, so the majority of the entries in the matrix are just ‘0’. But the placement of those few ‘1’s holds critical information.

  Then, the Python library runs this long, complicated matrix factorization formula:

  Some complicated math… When it finishes, we end up with two types of vectors, represented here by X and Y. X is a user vector, representing one single user’s taste, and Y is a song vector, representing one single song’s profile.

  The User/Song matrix produces two types of vectors: User vectors and Song vectors. Now we‘ve got 140 million user vectors — one for each user — and 30 million song vectors. The actual content of these vectors is just a bunch of numbers that are essentially meaningless on their own, but they are hugely useful for comparison.

  To find which users have taste most similar to mine, collaborative filtering compares my vector with all of the other users’ vectors using a mathematical dot product. Whichever produces the lowest product is the most similar user to me. The same goes for the Y vector, songs — you can compare a song’s vector with all the other song vectors, and find which songs are most similar to the one you’re looking at.

  Collaborative filtering does a pretty good job, but Spotify knew they could do even better by adding another engine. Enter NLP.

  The second type of recommendation model that Spotify employs are Natural Language Processing (NLP) models. These models’ source data, as the name suggests, are regular ol’ words — track metadata, news articles, blogs, and other text around the internet.

  Natural Language Processing — the ability of a computer to understand human speech as it is spoken — is a whole vast field unto itself, often harnessed through sentiment analysis APIs.

  The exact mechanisms behind NLP are beyond the scope of this article, but here’s what happens on a very high level: Spotify crawls the web constantly looking for blog posts and other written texts about music, and figures out what people are saying about specific artists and songs — what adjectives and language is frequently used about those songs, and which other artists and songs are also discussed alongside them.

  The most-used terms bucket up into what Spotify calls “cultural vectors” or “top terms.” Each artist and song has thousands of daily-changing top terms. Each term has a weight associated, which reveals how important the description is (roughly, the probability that someone will describe music as that term.)

  “Cultural vectors”, or “top terms”. Table from Brian Whitman Then, much like in collaborative filtering, the NLP model uses these terms and weights to create a vector representation of the song that can be used to determine if two pieces of music are similar. Cool, right?

  First, a question. You might be thinking:

  But, Sophia, we already have so much data from the first two models! Why do we need to analyze the audio itself, too? Well, first of all, including a third model further improves the accuracy of this amazing recommendation service. But actually, this model serves a secondary purpose, too: Unlike the first two model types, raw audio models take into account new songs.

  Take, for example, the song your singer-songwriter friend put up on Spotify. Maybe it only has 50 listens, so there are few other listeners to collaboratively filter it against. It also isn’t mentioned anywhere on the internet yet, so NLP models won’t pick up on it. Luckily, raw audio models don’t discriminate between new tracks and popular tracks, so with their help, your friend’s song can end up in a Discover Weekly playlist alongside popular songs!

  Ok, so now for the “how” — How can we analyze raw audio data, which seems so abstract?

  …with convolutional neural networks!

  Convolutional neural networks are the same technology behind facial recognition. In Spotify’s case, they’ve been modified for use on audio data instead of pixels. Here’s an example of a neural network architecture:

  Image credit: Sander Dieleman This particular neural network has four convolutional layers, seen as the thick bars on the left, and three dense layers, seen as the more narrow bars on the right. The input are time-frequency representations of audio frames, which are then concatenated to form the spectrogram.

  The audio frames go through these convolutional layers, and after the last convolutional layer, you can see a “global temporal pooling” layer, which pools across the entire time axis, effectively computing statistics of the learned features across the time of the song.

  All of this information then arrives at the output layer, which predicts an understanding of the song’s personality: Does it have a high tempo? Is it acoustic? Does it have high danceability? All of these characteristics can be found with pretty high accuracy just from letting these neural networks loose on the audio file.

  That covers the basics of the three major types of recommendation models feeding the Recommendations pipeline, and ultimately powering the Discover Weekly playlist!

  Of course, these recommendation models are all connected to Spotify’s much larger ecosystem, which includes giant amounts of data storage and uses lots of Hadoop clusters to scale recommendations and make these engines work on giant matrices, endless internet music articles, and huge numbers of audio files.

  I hope this was informative and tickled your curiosity like it did mine. For now, I’ll be working my way through my own Discover Weekly, finding my new favorite music, knowing and appreciating all the machine learning that’s going on behind the scenes. 🎶

  — —  If you enjoyed this piece, I’d love it if you hit the clap button 👏 so others might stumble upon it. You can find my own code on GitHub, and more of my writing and projects at http://www.sophiaciocca.com.

  Also, if you work at Spotify or know someone who does, I’d love to connect! I’m putting my dream to work at Spotify out into the world 😊

  Thanks also to ladycollective for reading this article over and suggesting edits.

  2017/10/11
• 
  

  Infoman

  Google Data Studio

  128 mkalygin 2 hrs 70

  https://www.google.com/analytics/data-studio

  http://news.ycombinator.com/item?id=15446049

  Google Data Studio (beta) turns your data into informative dashboards and reports that are easy to read, easy to share, and fully customizable. Dashboarding allows you to tell great data stories to support better business decisions.

  Create unlimited Data Studio custom reports with full editing and sharing.

  Sign up for free

  scroll icon charts icon Easily access all the data sources you need to understand your business and make better decisions.

  integrated icon Transform your raw data into the metrics and dimensions needed to create easy-to-follow reports and dashboards — no code or queries required.

  gears icon Data Studio gives you the ability to create meaningful, shareable charts and graphs that bring your data to life.

  headphones icon Harness the collective wisdom of your team and work together quickly, from anywhere.

  Learn more

  Explore our gallery of report and dashboard examples. Then if you see something you like, make a copy and start connecting your data.

  Visit the gallery

  Our pre-built data connectors make it easy to bring together all the data you need, how you need it, to create interactive and dynamic reports.

  Add Google Analytics data to your dashboard to track and analyze the performance of your websites and mobile apps.

  Add Google AdWords data to your dashboard to track and analyze the performance of your digital marketing.

  Add YouTube data to your dashboard to track and analyze your channel’s key video metrics.

  See all data connectors

  The Google Analytics 360 Suite combines enterprise analytics, tagging, site optimization, data visualization, market research, attribution, and audience management into a powerful measurement solution for your business. Designed to deliver a seamless user experience, cross-product data integration helps your enterprise gain useful insights and get better results every day.

  If you need assistance, dedicated support and services ensure that help is there when you need it.

  Talk to sales Learn more

  APMEX APMEX, Inc. increases email conversion rates by 62% with Analytics 360 and Google BigQuery.

  pdf icon View PDF Skyscanner Integrating Analytics 360 and BigQuery improves efficiency and insights for Skyscanner.

  pdf icon View PDF U.S. Cellular U.S. Cellular turns to Analytics 360 and its integration with BigQuery to analyze terabytes of data and reveal the true impact of digital media on sales.

  pdf icon View PDF See all Success Stories

  2017/10/11
• 
  

  Infoman

  Sourced from A Day Magazine

  發揮D.I.Y.的最高精神：IKEA推出「料理說明書」，組裝出一道美味料理只需要5分鐘 WANJUN 16, 2017

  Facebook IKEA最著名的就是「D.I.Y.」的精神，所有物品都可以藉由照著簡單易懂的說明書，只要照著步驟，就能拼裝出完整的家具用品。近期IKEA的創意點子真的相當多元，除了組裝家用品之外，你有想過D.I.Y出一道料理嗎？

  “Cook This Page”

  IKEA的加拿大分公司推出一系列的「料理說明書 Cook This Page 」，每份說明書上都已經清楚的列出所有需要材料，你只要按照格子大小，放上適量的材料，接著將印有可食用墨水的烤盤紙，捲起、放入烤箱內，一切就大功告成。除此之外，因為說明書上的原料大小，會依照IKEA生鮮販賣部的食材大小設計，如果想要更方便、省時，也可以考慮在IKEA購買料理食材。比起看著食譜上的文字敘述，Step1.2.3..的繁複步驟，這份「料理說明書」可以說是懶人與手拙女孩的救星。

  1

  瑞典肉丸義大利肉醬水餃

  放上與說明書上同樣大小的一小搓鹽巴、蒜末，擺好羅勒葉、塗上醬料，捲起來，完成！

  2

  香草佐檸檬烤鮭魚

  拿出IKEA販售的新鮮鮭魚塊，將檸檬切成說明書中的大小，適量的鹽巴、洋蔥粉，把鮭魚塊一一整齊排列，放入烤箱，完成！

  3

  橄欖油漬番茄蝦

  準備好同樣數量的蝦仁數，將食材一一放入說明書中的格子內，放入烤箱中，完成！

  Photos：Ikea

  目前這一系列「料理說明書」尚未正式上市，但是想必屆時推出會造成一股轟動。喜歡料理，又總覺得太過發費時間的女孩們，一起引頸期待吧！

  2017/10/10
• 
  

  Infoman

  2017 California school test scores: Why are they flatlining?

  38 masonic 2 hrs 63

  http://news.ycombinator.com/item?id=15438694

  Breaking with its steadily upward trend, California’s annual test scores have stagnated, with fewer than half of students proficient in math and English, and a wide ethnic achievement gap persisting.

  State scores released Wednesday show just 49 percent of students proficient in English and 37 percent proficient in math. The numbers are half a percentage-point different from 2016 — down in English and up in math. Tests were administered last spring to students in third through eighth grades and 11th grade.

  

  The state’s education leaders played down officials’ attitude.

  Reading this on your phone? Stay up to date with our free mobile app. Get it from the Apple app store or the Google Play store. “Test scores are not the only way to look at how students are doing,” state Board of Education President Michael Kirst, a professor emeritus at Stanford, wrote in an email. He cited graduation rates, suspension, enrollment in college-prep and advanced-placement classes as other measures of school performance.

  “Our watchwords are patience, persistence, humility and continuous improvement,” he wrote.

  Likewise, Tom Torlakson, state superintendent of schools, issued a written statement celebrating 2016 gains (of 3 percent), the number of students tested (3.2 million) and the rigor of the tests.

  Critics, however, lambasted those responses.

  “The state has given itself permission to fail,” said Ryan Smith, executive director of the Education Trust-West, an Oakland-based research and advocacy organization.

  As evidence of that, he said, “couple scores with an accountability plan that doesn’t hold schools responsible and a report card that parents need a degree in analytics to read.” He was referring to California’s education plan that critics see as letting failing schools off the hook, and the state’s colorful but confusing online “dashboard” that’s intended to depict school progress.

  California isn’t alone in its stagnating scores and in fact compares favorably with the 12 other states, plus the U.S. Virgin Islands, that administered the same tests last spring from the Smarter Balanced Assessment Consortium. Scores decreased in every state, and California’s decline was about the smallest, retired educational testing specialist Douglas McRae said.

  Those languishing scores have raised questions about the quality of the test itself.

  “It stands out like a sore thumb,” said McRae, of Monterey. “You have to ask the question: Is it because of the performance of students in schools, or is it something with the test?”

  McRae thinks the test is faulty.

  Statewide, scores flatlined, from high-scoring suburbs to inner cities and rural areas.

  SJM-L-CALATESTS-0927-90In both the Fremont and Pleasanton unified school districts, English and math scores were identical to previous years. In Palo Alto, where high scores fuel a roaring housing market, both math and English scores fell 3 points. In the San Mateo-Foster City School District, scores in various grade levels and sub-areas were flat or declined up to 2 percentage points — “not a significant change,” spokeswoman Amber Farinha wrote in an email.

  The plateau in scores is not surprising after second-year familiarity with the tests bumped up scores, Gilroy Unified Superintendent Debbie Flores said in an email. Her district dropped 1 point in English, to 48, and stayed at 40 in math.

  Charter schools posted varied scores but mostly reflected the marginal changes that other public schools did. Rocketship Education’s 12 Bay Area elementary schools (its newest campus did not have reportable scores) gained 3 points, reaching 44 percent in English and 54 percent in math. The charter chain serves predominantly students from poor and immigrant families.

  Last spring was the third administration of the California Assessment of Student Performance and Progress, aswhich includes Smarter Balanced and other examsis known in this state and which replaced tests in the previous STAR system. The CAASPP It is a computer adaptive test — meaning as the test progresses, questions become harder or easier depending on the student’s answers. The test also includes “performance tasks” where students apply knowledge to analyze a real-world problem.

  While proficiency levels took a dive when the state switched from multiple-choice tests to the more sophisticated and challenging CAASPP, scores were expected to rise as the Common Core curriculum and the tests became more familiar.

  McRae cites national data indicating a test problem and decries the opaque testing body that he said has refused to divulge critical information that could help analyze the integrity of the test.

  Flat or declining test scores alone do not signal a school problem, especially in districts such as Piedmont Unified, where scores fell but remain in the mid-80s.

  But persistently low achievement rates of African-American, Latino, poor, English-learner and disabled students, who together make up a majority of the state’s public-school students, fuel criticism of the educational status quo.

  In Santa Clara and San Mateo counties, home to some of the state’s highest-performing schools, Latino English proficiency is identical to the state’s: 37 percent. Latino math proficiency in the two counties is 27 percent, two points higher than the state figure — but more than 50 points lower than Asian math proficiency in the counties.

  Several Bay Area school districts serving large numbers of Latinos can’t even match the state’s proficiency levels. San Jose Unified’s Latino math proficiency is 22 percent, a 1-point increase from last year, but 5 points below the state. The figures are 21 percent in Mount Diablo Unified in Concord, 16 percent in West Contra Costa Unified in Richmond, and 15 percent in Oakland Unified. Scores for African-American students are equally dismal.

  “Any parent looking at these numbers is justifiably alarmed,” Matt Hammer, CEO of the advocacy group Innovate Public Schools, wrote in an email.

  Those numbers dismay critics who see no political will to act. “California’s current direction is not moving us forward,” said Smith, who blames state leaders, from Gov. Jerry Brown on down. “There’s a lack of a sense of urgency on this issue.”

  Finding your school’s test scores For 2017 school, district, county and state standardized test results, go to caaspp.cde.ca.gov, then click on “English Language Arts/Literacy and Mathematics.”

  On the next page, click on the blue “Search Smarter Balanced Results.” On the following page, select either “View Statewide Results” or select the county and, if desired, school district, then school. (Most charter schools are listed among the school districts.) Click on “View Test Results.”

  The resulting page will show test scores for all students of your selected school. Click on “Select Student Group/Subgroup” to see results by gender, ethnicity, poverty and other categories. Note: Results of the California Alternate Assessments, given to some special-education students, and of state science exams will not be posted until later this year.

  2017/10/10
• 
  

  Infoman

  Hong Kong's wildest dining experience

  Text by Maggie Hiufu Wong, video by Amanda Sealy and Daniel Campisi, CNN • Updated 9th October 2017

  Editor's Note — CNN Travel's series often carries sponsorship originating from the countries and regions we profile. However, CNN retains full editorial control over all of its reports. Read the policy . Hong Kong (CNN) — It's 8 p.m. in a rather dull multi-story government building and wet market center in North Point, Hong Kong . Most of the lights are off. The market stalls are all closed. Suddenly you hear the faint sounds of Michael Jackson classics, the vibrations getting stronger as you move towards the music. You're on the right path to Tung Po, the wildest restaurant in town. Related content See hilly Hong Kong on world's longest escalator system The dancing boss: Robby Cheung Tung Po co-owner Robby Cheung attracts almost as many diners as the food.
  Tung Po's different from most restaurants located inside cooked food centers and wet markets -- popular among locals in search of no-frills cheap but delicious eats -- thanks to one person: the restaurant's energetic owner Robby Cheung. Wearing an apron and a pair of white wellington boots with tops scissored into zig-zag shapes, Cheung is often seen doing the splits, drinking, dancing and taking selfies with customers. "The people just enjoy the food, enjoy the music and we dance together," the 55-year-old tells CNN. "I just like to dance. I'm 55 years old right now and I still dance. It [would be] so sad if I can't dance -- that would mean that I'm very old." Related content Hong Kong's best dim sum 'Favorite dish is black ink squid, because it's funny' Wind Sand Chicken is marinated and deep-fried till its skin is crispy but the meat remains moist. It's served buried in fried garlic and other spices. On any given night, it's almost impossible to find an empty table in Tung Po. Prior to taking over the spot 25 years ago with his friends, Cheung worked in various industries -- from acting to managing a disco to working in hotel restaurants. Far from being the cheapest restaurant in a cooked food center, Tung Po's menu ranges from classic Cantonese dishes to fusion cuisines such as prawn in orange sauce or tom yum-style fried rice cake. "My favorite dish is black ink squid," says Cheung. "It's a funny dish, you know? When you eat the black ink squid, you get a black mouth and black teeth." Other Tung Po signature dishes include Wind Sand Chicken, fried rice with duck sauce wrapped in lotus leaf and deep fried giant grouper with black pepper and toast. Related content 17 beautiful places to see in Hong Kong Tourist attraction Travelers around the world come to Tung Po to experience its famously rowdy atmosphere. Tung Po's rowdy atmosphere has made the restaurant a unique attraction, luring locals as well as travelers from around the world. "Every night we are all sold out. Over 60% is girls, office ladies," says Cheung. Like everything else in Tung Po, drinking beer is a dramatic affair. Instead of a glass, beer is served in an iced "battle beer bowl" -- a blue and white bowl with the Chinese word "victory" printed on the bottom. In Cantonese, yum sing (literally translated as drink victory) means bottoms up. Related content Top 10 things for visitors to do in Hong Kong The ultimate party trick

  With "Billie Jean" booming in the background, Cheung moonwalks to a table to deliver beers to his customers. Then he shows off one of Tung Po diners' favorite tricks -- opening a beer bottle with a chopstick. "A long, long time ago -- before we opened this restaurant, I saw it once outside a restaurant when they opened it with a chopstick," he recalls. "It was quite funny. Sometimes when a gweilo [foreigner in local slang] or a Chinese person wants to learn, I'll teach them." Check out the above video to see how he pulls off the trick. Reservations are recommended. Tung Po only accepts calls for bookings between 2:30-5:30 p.m. each day. Tung Po, 2F, Java Road Municipal Services Building, 99 Java Road, North Point ; +852 2880 5224 Read more WonderWhere the next adventure beginsSee all WonderThe Wonder List with Bill WeirDestinationsFood & DrinkPlayStayWonderGo to top of pageInstagramFacebookTwitter © 2017 Cable News Network. Turner Broadcasting System, Inc. All Rights Reserved. CNN Sans ™ & © 2016 Cable News Network. Terms of UsePrivacy PolicyAdChoicesAdvertise with usAbout usNewslettersWork for usHelpTranscriptsLicense FootageCNN Newsource closeI agreeBy continuing to browse our site you agree to our use of cookies, revised Privacy Policy and Terms of Service. More information about cookies.

  2017/10/10
• 
  

  Infoman

  香港生活忙碌，口頭禪是「為兩餐」，但其實飲食在我們的生活名單上排名甚低，現代都市人忙工作、忙煲劇、忙上網、忙自拍、忙WhatsApp等等，飲食的時間變得愈來愈奢侈了。

  因此，城市人的健康質素每況愈下，商家為了大家的身體健康，也為了自己的荷包着想，當然會推出不同的飲食商品，幫助大家食飯。

  被眾網友推為最「神」的飲食商品之一，便是宜家的「食譜組裝紙」，一向講創意的IKEA大概理解到新世代不擅廚藝，早前推出一套「食譜組裝紙」。

  看圖填食材

  其實每一張「食譜組裝紙」都是一份食譜，用家只要依照紙上所列的份量準備材料就是了，大家可能會奇怪，坊間食譜極為普遍，何以被稱為「神」？

  原來這張食譜紙就像一張填色畫紙，上面列明了食材及調味的圖案及位置，用家只要把各樣材料都放在圖上，然後依照指示把「食譜組裝紙」摺起，然後放入焗爐便可以了。

  這張「紙」當然不是一般的紙張，是專供煮食用的包裝紙，盛載液狀調味料在短時間內不會滲漏，把食物拿出時可以打開就吃，吃完包起廚餘便放到垃圾桶裏，連洗碟的過程也省去。

  有了這套「食譜組裝紙」，即使看起來很難炮製的菜餚，也因為「看圖識字」而變得超容易，「食譜組裝紙」更有多款菜式如「意大利餃伴瑞典肉丸」、「檸香三文魚」、「香蒜烤蝦」等，易煮好味又有營養，被封「神器」也非常合理。

  大家還記不記得小時候媽媽教導我們不要做懶人的寓言故事《大餅人》？話說大餅人四肢健全，卻懶得連食飯也要家人餵，某次家人出門，把大餅掛在他頸上，讓他自己吃，怎知他懶到只吃了頸前的大餅部分，餘下的因要用手拿才吃得到，他都懶得做，於是便餓死了。

  相信新世代都非常羨慕大餅人，只要張大口就有餅吃，省下的一雙手可以繼續做其他事情，想不到今天大家都「夢想成真」了。

  在歐美國家，政府為免司機一邊駕駛一邊講電話或拿食物而造成交通意外，於是規定司機必須雙手駕駛，但外國可不同香港，車程隨時幾小時，司機講電話時可以用免提裝置，但要吃東西的話就必須停車了，若行程趕急或者在高速公路上又碰巧肚餓怎麼辦？

  「殘廢餐」幫手

  於是一件名為「食包輔助器」的飲食神器便橫空出世了，原理就如大家熟悉的《大餅人》，把大包或大餅放在這個「輔助器」上再掛上頸，大包正好對着嘴，只要張開口便可以吃，不必勞動手與眼，當然，除了極為肚餓的司機，其他大忙人一樣可以使用。

  與「食包輔助器」有異曲同工之妙的飲食神器還有「免持水杯頸帶」，原理亦一樣，把水杯掛在頸上，不必用手拿杯，但設計只適宜靜位的用家，太大動作的話，隨時一身都是。

  相片：網上圖片

  撰文 : 瑪麗

  2017/10/10
• 
  

  Infoman

  Facebook Quietly Enters StarCraft War for AI Bots

  179 sus_007 13 hrs 83

  https://www.wired.com/story/facebook-quietly-enters-starcraft-war-for-ai-bots-and-loses/

  http://news.ycombinator.com/item?id=15433004

  In the distant Koprulu Sector of the Milky Way, Facebook’s Zerglings lingered in a restless swarm outside the enemy’s base. After the commander ill-advisedly opened the gate, the social network’s alien horde stormed in and slaughtered forces stationed inside, in a battle fought on the frontiers of artificial-intelligence research.

  The bloody incident was part of an annual competition of the videogame StarCraft for AI software bots that wrapped up Sunday. Facebook quietly entered a bot called CherryPi designed by eight people employed by or affiliated with its AI research lab.

  The social network’s stealthy space war suggests Facebook is serious about competing with Google and others to set showy new milestones in AI smarts. Google’s London-based DeepMind AI research unit made headlines last year when its AlphaGo software defeated a champion at the board game Go. In August, DeepMind declared StarCraft II, the latest version of the game, as its next target.

  The contest Facebook entered, like most AI research in the area, used an older version of StarCraft, which is considered equally difficult for software to master. Facebook’s AI research group, which lists 80 researchers on its website and is led by NYU professor Yann LeCun, has produced many research papers but not notched up an achievement as striking as Google’s with Go. Facebook has released three research papers on StarCraft, but not announced a special effort to conquer the game.

  Final results released Sunday indicate Facebook still has a way to go: CherryPi finished sixth in a field of 28; the top three bots were all made by lone, hobbyist coders.

  Gabriel Synnaeve, a research scientist at Facebook, described CherryPi to WIRED as an "baseline" on which to build future research on StarCraft. "We wanted to see how it compares to existing bots, and in particular test if it has flaws that need correcting," he said. CherryPi competed in a long-running contest that is part of AIIDE, an academic conference on applying AI in entertainment. Facebook also sponsored this year’s contest, paying for hardware used to run the thousands of bot-on-bot games.

  Related Stories

  Games such as tic-tac-toe, checkers, chess, and Go have been testbeds for new ideas in artificial intelligence since the field’s beginnings in the 1950s. These days, there’s also a serious business purpose, as companies increasingly use AI to hone their product and service offerings. Facebook, Google, and other tech companies use AI to improve ad-targeting and personalization systems, and enable new products, such as virtual assistants and augmented reality.

  StarCraft is alluring to AI researchers for more than just the fun of commanding weapons like the building-leveling Yamato plasma cannon. Although the videogame may appear more approachable than Go or chess, it is many times more complex, because players’ pieces and actions aren’t limited to a tightly regimented board and always in full view of their opponent. The number of valid positions on a Go board is a 1 followed by 170 zeros. Researchers estimate that you’d need to add at least 100 more zeros to get into the realm of StarCraft’s complexity.

  The winning bot in this year’s competition, ZZZKBot, was made by Chris Coxe, a software developer in Perth, Australia, who previously worked for NASDAQ. He built his bot alone, and lately took a break from work in part to dedicate more time to it. A day before the final results were announced, Coxe spoke self-deprecatingly of his handiwork. “It was supposed to be a proof of concept,” he said. “The source code isn’t all that great.”

  Like all StarCraft bots so far, ZZZKBot wouldn’t last long against even a moderately skilled human StarCraft player. The feats of planning and memory required to predict and react to the maneuvers of an alien army are beyond today’s software.

  The days of amateurs building the best StarCraft bots appear to be numbered now that two giant companies that compete in both online ads and AI prowess have taken an interest. David Churchill, a professor at Memorial University of Newfoundland who organized the AIIDE contest, predicts the StarCraft bot scene is set for a big shake up over the next few years.

  Facebook and Google say they are approaching StarCraft differently than most individual programmers have. Leading bots are based mostly on rules and strategies specified by their creators. Coxe says one of his bot’s best features is a simple learning feature, in which it tries out pre-programmed strategies against each bot it plays and notes which one works so as to be prepared in their next matchup. The tech giants are planning to lean more heavily on machine learning, planning to have bots develop their own strategies from scratch by examining large caches of data from past games, or repeated experimentation. Facebook didn't build ideas it has published along those lines into CherryPi. Machine learning was central to making Google’s AlphaGo unbeatable.

  Facebook’s bot may not have won the StarCraft competition, but Dan Gant, whose bot PurpleWave placed second, saw hints of the future in its play. Most bots choose to either attack frontally, or retreat, based on the relative numbers in opposing armies. In videos released from the contest prior to the final results, CherryPi appeared to know when it could move fast enough to sneak around an enemy to attack its base, says Gant.

  Still, don’t expect lone bot builders to disappear overnight—or StarCraft to be conquered soon. “The problem is still so difficult,” says Churchill. “For a couple of years I predict the hobbyist, mostly rule-based bots, will still do well.” He guesses it may be five years before any bot can beat expert humans—but acknowledges it may be sooner.

  Gant, a software developer in New York, took a break this year and spent months working full-time on PurpleWave. He says the entrance of tech giants adds to the appeal of a pursuit that presents a unique learning opportunity. “You can be Facebook or DeepMind or a kid just learning programming and you’re competing on a level playing field,” he says. “You’re limited by your own effort and what you can teach yourself.”

  Making a superhuman StarCraft player could deliver tech companies more than just satisfaction. Google says machine learning from DeepMind has helped cut cooling bills in its datacenters. A Microsoft research paper on machine learning this year said that improving predictions of when a user will click on an ad by just 0.1 percent would yield hundreds of millions of dollars in new revenue. A bot capable of leading armies of alien zergs to crush any human might quickly earn its keep.

  2017/10/10