HN user
Yes, it's quite clever. An equivalent proof is dividing 0.999... by 9 using long division, which comes out to 0.111... which is equal to 1/9. Now use fraction notation and it simplifies to 9/9 = 1. Not quite as robust as the limit-based proofs but it's a quick answer and gets to the heart of the issue of repeating notation not capturing the whole picture.
How can the US have that little protection of their workers..?
The US has spent the last 65 years chipping away at unionization rates and other pro-employee protections. The overall rate of unionization in the US peaked in 1954 [1].
[1] http://www.dollarsandsense.org/archives/2011/0311reuss2.html
Yes - this is a major function of private equity in general. They even have investment banks and firms that specialize in "buy side" versus "sell side" of the transaction.
I once had a consulting gig where the customer desperately wanted to build a Spark/Scala ML pipeline, for a dataset that was 10 MB. We spent 3 months hammering it together for a flat Python process that would've taken us 2 weeks.
Neil Postman was my single most favorite find in a thrift store.
Long-haul trucking would be a great use-case. Even if you need humans for local pilot guidance, you can queue them up to a "cell phone lot" and courier the last-mile drivers in between shifts of going back & forth. It would also mean drivers would be based more locally to points of loading, rather than having to be constantly on the road away from home.
Cash or a bank demand account is not without their own inherent risks. A fire, robbery, or forced currency exchange could destroy the value of the physical commodity of cash, and the FDIC only insures individual account bank deposits up to a certain limit so a bank institution failure could cause losses to individual accounts.
Drinking multiple _liters_ of water in a relatively short time can cause water intoxication [1] due to the drop in relative electrolyte levels. I got it once at scout camp where, unfortunately, the medic misdiagnosed me with dehydration and had me drink a gallon of water. I ended up urinating about every ten minutes for the next 24 hours, and is one of the most unpleasant experiences I've gone through.
Most MLM's tout a flexible working schedule which women tend to prefer due to being saddled with a higher average proportion of the child-rearing and elder-care responsibilities.
Over time the attrition rate for attention increases. As a result it becomes harder to justify further investments in usability features or content quality because you can no longer leverage the increased attention of more users in the future. This can lead towards a death spiral of lower frequency updates leading to less viewers leading to less updates.
For a venture-backed site even hitting a 20% growth rate per month would be a warning that you're not growing fast enough to justify continued investment, as was the case for a recent postmortem on Gumroad [1].
Given that this post was from 10 years ago that's definitely not the case for Hacker News, but in this climate it's gotten even more extreme if you're relying on venture backing.
I think we would have to understand better how instinctive behavior ends up affecting this kind of cognitive processes, but I think it would be possible. Examples that immediately come to mind would be some cats' aversion to cucumbers due to resemblance to snakes, or superprecociality [1] in newborn animals who can walk within minutes of being born or fly on the same day they hatch.
DOB and date of last checkup are highly schematizable. This specific example is very well-suited to a SQL solution dumping into an autocaller queue.
Maybe this was a terrible example, and the author didn't grasp a good example of legitimately non-schematic data points?
Funny - half of my experiences talking with nonprofits were telling people with an Access database to migrate to a proper SQL database.
Mid-tier Midwestern cities are quickly gentrifying with bikeshares, kombucha shops, craft breweries, and bookstore/brewpub combos. Madison, Indianapolis, & Cincinnati often get mentioned in these conversations, but I know Detroit is having a bike renaissance right now and land is _dirt cheap_ in all of these cities.
I know Indianapolis in particular has had an influx of tech companies opening large campuses, including Salesforce, Genesys, and Infosys. This has gone hand-in-hand with rises in indie music festivals and a blossoming craft brewing industry.
I would add that humans also integrate gyroscopic & acceleration information from the inner ear to understand relative balance. Multiple sources of sensor data is a net benefit, not a drawback.
Fascinating. I've used my own hacked-together unsupervised learning to facilitate topic labeling, and this seems to address the bottleneck problem for supervised learning. Will have to dig in deeper though to pick out specifics.
Also interesting that one of the fundamental problems the authors note is "The limits and bias of learning about the world through text", which is essentially a Godelian incompleteness problem. One could say the reverse also applies to embodied/visual data, and a good argument for studying established literature in the abstract.
Logistic regression is also referred to as a "supervised classification" problem, which this book only addresses in the specialized space of document clustering or image classification. They do also address Support Vector Machines, which is a generalized algorithm for classification. However, there are a wide variety of specific implementations of logistic regressions that require quite a bit more conversation (dummy variables, log-odds ratios, ordinal variables) that are more directly applicable to a general stats background to machine learning itself.
Considering that the authors are all CS professors or researchers and not statisticians, that makes sense to me why they don't view logistic regression as foundational.
Well - there _is_ a bit of Student's Paradox involved in requirements-gathering for machine learning or analytics, and it's not always apparent that you're using the right algorithm for the problem or what specific metrics you actually need to be optimizing. Most of the time when I've gone through User Story collection processes, the actual end-state users describe is vague, not based on hypothesis testing, and not supportable by the data available onhand. A big part of this process is discovering what the data _can_ tell you, and if necessary to rewrite the technical requirements entirely to align with the actual business need.
Even when a customer's given me a "clean" dataset, I've had to write 400+ lines of code to do the feature engineering on a relatively straightforward logistic regression. Then there's all the other times when a customer asks me to deploy one type of algorithm, and their business problem is actually solved by an entirely different class of algorithm entirely.
Zayd over at Stanford has a nice blog post [1] describing why machine learning is several more dimensions of complexity compared to traditional software development. There _is_ a specific set of Data-first skills that is complemented by dev and CS experience, but a fundamental reason why ML projects fail is due to lack of appreciation for the many different skillsets needed to succeed.
[1] http://ai.stanford.edu/~zayd/why-is-machine-learning-hard.ht...
I found it simply fascinating when I learned that Blue Wildebeests are born knowing how to stand within minutes of their birth, and within a day can outrun adult hyenas:
"Extremely precocial species are called "superprecocial". Examples are the Megapode birds, which have full flight feathers and which, in some species, can fly on the same day they hatch from their eggs. Another example is the Blue Wildebeest, whose calves can stand within an average of six minutes from birth and walk within thirty minutes; they can outrun a hyena within a day."
What would you propose as being a "market-decides" approach to health care, beyond abolishing any-party insurers and revert to a 19th-century model?
We have lots of examples of governments doing this with ranges from poor to fair to good results. To my knowledge, the US is the only widescale attempt at market-based health care, with results of 2x the OECD average of health care cost per capita [1].
[1] https://read.oecd-ilibrary.org/social-issues-migration-healt...
I had a very bitter argument about whether Kansas was Midwestern or not. An important distinction is that what's now considered the "Midwest" used to be just called the "West", and that the Great Plains formed a natural barrier of relatively lower rainfall that settlers headed out to Oregon or California would be crossing over before they hit the Mountain West. As a result, settlers in Kansas had lots of natural connections to Missouri, which is fairly clearly in the Midwest.
Neat observation. The "3-longest-words" approach probably works well because grammatical words tend to be elided down to as short of an implementation as possible, while longer words tend to be more demonstrative of the actual topic at hand, rather than grammatical structure.
Eh, that really comes down to applied theory vs. pure theory. There's no one Grand Unifying Theory of Natural Language Processing, and not likely to be a strong candidate for a while yet. Until then, there can still be a lot of good problem-solving that can be used with either traditional NLP or with neural networks, or even a hacked-together hybrid approach, and both application and research will feed into each other to refine the processes.
The specific instance I was remembering was from interviews Hinton's given about these papers, but this is the section of the arXiv paper that's relevant:
Now that convolutional neural networks have become the dominant approach to object recognition, it makes sense to ask whether there are any exponential inefficiencies that may lead to their demise. A good candidate is the difficulty that convolutional nets have in generalizing to novel viewpoints. The ability to deal with translation is built in, but for the other dimensions of an affine transformation we have to chose between replicating feature detectors on a grid that grows exponentially with the number of dimensions, or increasing the size of the labelled training set in a similarly exponential way. Capsules (Hinton et al. [2011]) avoid these exponential inefficiencies by converting pixel intensities into vectors of instantiation parameters of recognized fragments and then applying transformation matrices to the fragments to predict the instantiation parameters of larger fragments. Transformation matrices that learn to encode the intrinsic spatial relationship between a part and a whole constitute viewpoint invariant knowledge that automatically generalizes to novel viewpoints. Hinton et al. [2011] proposed transforming autoencoders to generate the instantiation parameters of the PrimaryCapsule layer and their system required transformation matrices to be supplied externally. We propose a complete system that also answers "how larger and more complex visual entities can be recognized by using agreements of the poses predicted by active, lower-level capsules".
More broadly speaking, the benefit of being able to recognize slightly transformed viewing angles leads to dramatically fewer needed training observations that are still clearly identifiable as the same object.
Was very pleased to find this out when I first started studying word embeddings (the abstract principles of word2vec). Essentially it comes down to words having similar verbs and objects that come up most frequently together, so they end up being semantically close.
Anaphora resolution is one of the really hard problems not only in NLP but in linguistics in general.
This was one of the most frustrating parts of studying Latin rhetoric. The speakers would keep referring to "That thing I was talking about," and it's a noun from a subordinate clause 2 and a half paragraphs ago.
Well, there's also observations from neuroscience. CNN's were directly inspired by observations of closely-grouped neurons firing when detecting certain images, and capsule networks are directly inspired by cortical columns.
The major advantage proposed for capsule networks is the ability to train off far fewer number of observations, not necessarily the full accuracy. At this point CNN's are consistently approaching or even exceeding human levels of accuracy, and thus be benefiting from a slower but more accurate methodology that relies on far more training data.
Before the printing press, there was dramatic variation in regional dialects of European nations. After the printing press, standardization and state centralization have been grinding these regional dialects into the ground. Even traditional regional languages like Irish or Scots Gaelic are going extinct.