Links for July 2016

The results from the arXiv survey are in. Nature characterizes them as very conservative, but I as shocked to find that ~58% of responses thought “Allow readers to comment on papers” was very important, important, or somewhat important. From Andrej Karpathy:

I developed and maintain Arxiv Sanity Preserver (http://www.arxiv-sanity.com/), one of the Arxiv overlays the article mentions. I built it to try address some of the pains that the “raw” arXiv introduces, such as being flooded by paper submissions without any support or tools for sifting through them.

I’m torn on how Arxiv should proceed in becoming more complex. I support what seems to be the cited poll consensus (“The message was more or less ‘stay focused on the basic dissemination task, and don’t get distracted by getting overextended or going commercial’”) and I think the simplicity/rawness of arXiv was partly what made it succeed, but there is also a clear value proposition offered by more advanced search/filter/recommendation tools like Arxiv Sanity Preserver. It’s not clear to me to what extent arXiv should strive to develop these kinds of features internally.

Whether they go a simple or more complex route, I really hope that they keep their API open and allow 3rd party developers such as myself to explore new ways of making the arXiv repository useful to researchers. Somewhat disappointedly, the arXiv poll they ran did not include any mentions of their API, which in my opinion are a critical, overlooked and somehow undervalued. For example, today their rate limits are very aggressive and make it tricky to pull down publication metadata for Arxiv Sanity Preserver, even when this undoubtedly costs minimal bandwidth. In the future, I’m concerned they will discontinue the API altogether and prevent similar 3rd party overlays from being built.

The arXiv has issued a nice report on the results of their survey. From the summary:

The combination of multiple choice responses (see Appendix B) and the extensive and thoughtful open text comments pinpointed areas that need to be upgraded and enhanced. Improving the search function emerged as a top priority as the users expressed a great deal of frustration with the limited search capabilities currently available, especially in author searches. Providing better support for submitting and linking research data, code, slides and other materials associated with papers emerged as another important service to expand. Regardless of their subject area, users were in agreement about the importance of continuing to implement quality control measures, such as checking for text overlap, correct classification of submissions, rejection of papers without much scientific value, and asking authors to fix format-related problems. Several users commented on the need to randomize the order of new papers in announcements and mailings. There were several useful remarks about the need to improve the endorsement system and provide more information about the moderation process and policies.

In regard to arXiv’s role in scientific publishing, some users encouraged the arXiv team to think boldly and further advance open access (and new forms of publishing) by adding features such as peer review and encouraging overlay journals. On the other hand, many users strongly emphasized the importance of sticking to the main mission and not getting side-tracked into formal publishing. There was a similar divergence of opinion about encouraging an open review process by adding rating and annotation features. When it comes to adding new features to arXiv to facilitate open science, the prevailing opinion was that any such features need to be implemented very carefully and systematically, and without jeopardizing arXiv’s core values…

While many respondents took the time to suggest future enhancements or the finessing of current services, several users were strident in their opposition to any changes. Throughout all of the suggestions and regardless of the topic, commenters unanimously urged vigilance when approaching any changes and cautioned against turning arXiv into a “social media” style platform.

I wish more of the text write-in answers were released, if only for ideas.

But they were not able to get good estimates of pet loss, since it is hard to pin down which pets that disappear were killed by cougars….

Also, they could not account for the obvious emotional response to predators. Even if the estimate is correct that five times as many people would be saved by cougars as would be killed, death by deer and cougar are different.

“The idea of being killed in a car crash with a deer just doesn’t scare people the way the idea of a cougar leaping on your back in the woods does,” Dr. Prugh said.

I, for one, would much rather be killed by a cougar attack than a deer accident.

No longer just hints, they are now making concrete plans to recover a three-part first stage (the central core, plus two boosters) on the soon-to-launch Falcon Heavy.

A Hawaii overprint note is one of a series of banknotes (one silver certificate and three Federal Reserve Notes) issued during World War II as an emergency issue after the attack on Pearl Harbor. The intent of the overprints was to easily distinguish US currency captured by Japanese forces in the event of an invasion of Hawaii and render the bills useless.

Presumably this is the base-level emotion tapped into by “preppers” and a serious bias to correct for regarding arguments about existential risk.

What’s the alternative? The satisfaction that comes from transparent, mechanistic understanding. The emotional appeal of the latter is less powerful, but more enduring and tracks reality better.

The program allows for living donors to donate a kidney in advance of when a friend or family member might require a kidney transplant.

…“It’s the brainchild of a grandfather who wanted to donate a kidney to his grandson nearing dialysis dependency, but the grandfather felt he would be too old to donate in a few years when his grandson would likely need a transplant.”

Nine other transplant centers across the U.S. have agreed to offer the gift certificate program, under the umbrella of the National Kidney Registry’s advanced donation program. Veale anticipates that more living donors will come forward to donate kidneys, which could trigger chains of transplants. Then, when a patient redeems his or her gift certificate, the last donor in the chain could donate a kidney to that recipient.

H/t Alex Tabarrok. Also from him, a discussion of results-blind review at the journal Comparative Political Studies. See Gwern for some of the main times and places this idea has appeared.

https://www.youtube.com/watch?v=IwIZ-wntgdo&t=31m48s

I’ve always thought the nice way to formalize this would be an amendment requiring a supermajority to pass a laws without sunset provisions.

In 2015, Comparative Political Studies embarked on a landmark pilot study in research transparency in the social sciences. The editors issued an open call for submissions of manuscripts that contained no mention of their actual results, incentivizing reviewers to evaluate manuscripts based on their theoretical contributions, research designs, and analysis plans. The three papers in this special issue are the result of this process that began with 19 submissions. In this article, we describe the rationale for this pilot, expressly articulating the practices of preregistration and results-free review. We document the process of carrying out the special issue with a discussion of the three accepted papers, and critically evaluate the role of both preregistration and results-free review. Our main conclusions are that results-free review encourages much greater attention to theory and research design, but that it raises thorny problems about how to anticipate and interpret null findings. We also observe that as currently practiced, results-free review has a particular affinity with experimental and cross-case methodologies. Our lack of submissions from scholars using qualitative or interpretivist research suggests limitations to the widespread use of results-free review.

(H/t Gwern.)[PDF]

I recently had a discussion with a distinguished physicists who basically took Feynman’s position. I think it’s easy to not be careful enough in distinguishing between the following questions: (a) What are the predictions made by the quantum formalism? (b) Do the predictions made by the quantum formalism agree with reality? (c) Which aspects of our quantum formalism are necessary for making these predictions, and which are optional, or could be improved?

Now, the psychological/sociological criticism is fine as far as it goes — “you only are overly skeptical of quantum mechanics because you foolishly refuse to let go of your classical intuitions!”. And I certainly think that, if a departure from quantum mechanics is found, it will be even weirder than the departures of quantum mechanics from classical mechanics. But a bet against the importance of Bell’s theorem is a bet against us discovering a departure from quantum mechanics by direct analysis rather than, say, by investigating an apparently orthogonal frontier like high-energy which unexpectedly yields surprising results. And I think that bet is looking less attractive than it was in the ’70s, now that our experimental data is petering to a stop.

Barratt thinks solving the puzzle of VIIP [Visual impairment intracranial pressure syndrome] is going to require testing intracranial pressure in space, even if that means an invasive procedure. One option is an intracranial probe that would be surgically implanted months before flight and allow pressure to be measured at different points during spaceflight. It could also answer questions about other potential factors, including heightened carbon dioxide levels and the effect of in-flight exercise.

“This is one of those times I think aggressive science is extremely warranted,” Barratt said.

The National Highway Traffic Safety Administration is soon expected to issue rules that will mandate transponders for all new cars and most trucks. This will permit vehicles to broadcast their speed, heading and braking status to anyone or anything within 300 meters, which is well beyond the range of current onboard sensors. These devices, called “V2V” (vehicle-to-vehicle) communicators, can see around corners and convey a driver’s intent (such as, say, an impending left turn), along with other relevant information. The NHTSA estimates that, at intersections alone, this simple step could prevent 500,000 crashes a year, saving 930 lives….

A 2014 analysis in the MIT Technology Review estimated that the aggregate financial benefit of converting the bulk of the U.S. fleet to automated vehicles would be more than $3 trillion a year—an incredible 15% of current GDP.

I don’t necessarily agree with that article’s technical suggestions, such as separating human-driven from autonomous traffic. To me, the easiest infrastructure to add is embedded magnets. Several projects (e.g., Volvo’s) have had good results. Importantly, magnets are very cheap compared to the cost of resurfacing.

The magnet-based system relies on magnets embedded along the center of the lane 1.2 meters apart to define the roadway. The car tracks the magnets with less than 7.5 centimeters of error. Although this requires doing something to the road, it is very cost-effective to spend about $10,000 per lane mile to buy and install the magnets compared to spending from $1 million to $100 million per mile to build new lanes. And the life of the magnets is 30 to 50 years — longer than the best pavements.

I don’t understand why we haven’t been putting these in for years in the same way that local governments have laid down dark fiber when putting in pipes. (It’s cheap and broadly useful for the future, even if they don’t have an immediate use.) Adding magnets to all 160,000 miles of the US national highway system would be a one-time cost of a few tens of billions of dollars, which is tiny in comparison to the hundred of billions or trillions we lose every year that autonomous cars are delayed.

It wouldn’t be too hard to pattern the magnets to communicate other info to the car too, such as an upcoming sharp turn. More abstractly, magnets could mark the car’s position extremely precisely, and the car could then acquire regularly updated info about upcoming road conditions from the internet, or a short-range local radio broadcast.

Blinding of reviews is another fertile area of study. In 1998, my colleagues and I conducted a five-journal trial6 of double-blind peer review (neither author nor reviewer knows the identity of the other). We found no difference in the quality of reviews. What’s more, attempts to mask authors’ identities were often ineffective and imposed a considerable bureaucratic burden. We concluded that the only potential benefit to a (largely unsuccessful) policy of masking is the appearance, not the reality, of fairness. Since then, online technologies for blinding have increased, as have numbers of scientists (and thus the difficulty of guessing who authors may be). It will be interesting to see how similar studies work out now, and whether double-blind reviewing affects acceptance rates for women and under-represented minorities.

Good introductory video to the mission by one endearingly nerdy scientist:

The solar arrays only collect 500 W, so they “run the whole spacecraft on a third or a fourth of what it takes to run a hairdryer”. Here’s the first image. That’s from 2.7 million miles, to be compared with the 3,000 miles above the cloud tops that JunoCam will reach in the coming weeks:

Jupiter itself will only appear to be 75 pixels across from JunoCam when Juno reaches the furthest point of its orbit around the planet. At its closest approaches JunoCam could achieve 15 km/pixel resolution from 4300 km, while Hubble has taken images of up to 119 km/pixel from 600 million km….The camera uses a Kodak image sensor, the KODAK KAI-2020, capable of color imaging at 1600 x 1200 pixels. It has a field of view of 18 x 3.4 degrees with three filters to provide color imaging.

… the…highest vacuum generated by pumps in the laboratory amounts to 10^-17 bar, which corresponds to just a few hundred particles in one cm³ and which was also obtained at CERN. One should compare this with a number of ~ 3 × 10¹⁹ particles per cm³ at atmospheric pressure and room temperature. Therefore, between the highest and lowest pressure there are 47 orders of magnitude in between. But even the best vacuum obtained on earth is a high-pressure area compared with the almost total emptiness between stars in space. Accordingly, besides pressure vacuum is characterised by the density of particles. The interstellar particle density in the Milky Way, for instance, consisting of gas, plasma and dust, is only ~ 5 × 10⁴ particles per m³. Between Galaxies one has only one particle or at most a few of them per m³. If one would distribute homogeneously the total matter of the universe in space, one would still have an extremely low particle density of only 3 particles per m³.

The head of the National Highway Transportation Safety Administration, which is investigating the autopilot’s performance in the collision, has called for such systems to demonstrate that they’re at least twice as safe as human drivers.

That to me is a very reasonable (and marketable) number. If the public can be sold on “two”, my optimism for self-driving cars clearing regulatory hurdles is much higher.

This would allow us not only to avoid some of the duplication across the teams but also to take more of a leadership position in Effective Altruism; focussing on supporting greater effectiveness and dedication among the many brilliant people involved in the movement. At the same time, we want to ensure that we do not damage the growth of Giving What We Can and it’s ability to bring in effective donors and we are carefully reviewing the strategy.

SER-109 is expected to become the first oral microbiome therapy approved by the FDA, though Seres declined to predict exactly when it will arrive. Results from the latest trials are due by midyear, and Phase 3 trials are scheduled to follow later in the year. Seres hopes to follow up quickly with SER-287, a drug to treat ulcerative colitis, which could be the first microbiome drug to treat a chronic disease, and SER-262, to treat primary C. diff before it turns into the recurrent kind.

Also

Another top contender is Rebiotix. Its RBX2660 is also designed to treat recurrent C. diff but, unlike SER-109, is administered with an enema; an oral version is in development…Seres does have at least one distinct market advantage. “Patients have different preferences,” Hecht observes, but “in general, people don’t particularly like enemas.”

Note that it seems to go back and forth from a walk-like stroke to a trot-like stroke, but doesn’t use a canter- or gallop-like strokes.

“Run” = canter, and “sprint” = flying gait, which is not the gallop.

Wifi was never meant to go insanely long distances, and now spamming APs inside of even a smaller home is a cheap and effective way of dealing with signal problems.

60ghz (Added by 802.11ad) can’t penetrate sheetrock effectively, let alone bother your neighbors, a much smaller distance than 5ghz (which has half-solved the dense urban environment issue on it’s own, but not completely…).

I support abandoning 2.4ghz for Wifi and anything else that is high bandwidth. The good news is, the 802.11 Working Group and the major Wifi impl manufs agree with me, which is why 802.11ac is 5ghz only, and 802.11ad is 60ghz first, both of them removing 2.4ghz for anything but 802.11 b/g/n legacy connectivity.

Also, as a side note, I want the FCC to ban the LTE-over-5ghz plan. 5ghz is ours; they can have 2.4ghz, we’re done with it.

[The] motion planning process is both one of the most important skills a robot can have (since it’s necessary for robots to “do stuff”), and also one of the most time and processor intensive….[R]esearchers from the Duke Robotics group… describe how they can speed up motion planning by three orders of magnitude while using 20 times less power….Rather than using general purpose CPUs and GPUs, they instead developed a custom processor that can run collision checking across an entire 3D grid all at once.