> It is not for lack of trying that all the top papers in visual question answering end up doing this as a classification task. Results are really poor when it is used as RNN generation
I'd be curious to know if you have a reference for this. Given that the answers are one word, a word-level RNN language model output should basically be the same thing as a straight 1000-way softmax.
Model Q+I [1] Q+I+C [1] ATT 1000 ATT Full
ACC. 0.2678 0.2939 0.4838 0.4651
Where ATT Full represents using all the words in the vocabulary, as you
can see it performs worse than "Most frequent 1000 answers".
Source:
Chen, K., Wang, J., Chen, L. C., Gao, H., Xu, W., & Nevatia, R. (2015). ABC- CNN: An Attention Based Convolutional Neural Network for Visual Question Answering. arXiv preprint arXiv:1511.05960.
2.
(a) Several early papers about VQA directly adapt the image captioning
models to solve the VQA problem [10][11] by generating the answer using
a recurrent LSTM network conditioned on the CNN output. But these
models’ performance is still limited [10][11]
(b) our own implementation of this model is less accurate on [2] than other
baseline models
Above two quotes are from -
Xu, Huijuan, and Kate Saenko. "Ask, Attend and Answer: Exploring Question-Guided Spatial Attention for Visual Question Answering." arXiv preprint arXiv:1511.05234(2015).
However, I think my words were sloppy, as I could not find more concrete
proof in the literature, but I will revisit them with detail to recollect
where I read about RNN generating answers not overachieveing softmax
classification over Top K distribution of answers.
Also, I would like to note that, I am not using only "one word answers" as
the possible set of answers. It contains few two words, and very few three
and four word answers.
Here is the distribution
Key == Length of words | Value == Count of answers with those many
words
According to the paper, the parameters fit on one GPU (or at least that one GPU was able to train the model). It was just too slow, so they trained on 8 GPUs in parallel. But those GPUs were still on the same machine (one node, multiple GPUs).
Cool algorithm! One problem is that its probability density calculation assumes a random ship placement. If the opponent were aware of how the algorithm works (let's say they deduced it from its behavior, which seems plausible) then they could devise a counter-strategy of placing ships on the squares which have the fewest other possible ship placements (e.g. in the beginning of the game, this is along the edges of the board).
According to their "About" page, a one-way ride costs $22. Even if you buy 50 rides at once, the price is still $17.6 per one-way ride. That's comparable to what a typical UberX ride around Palo Alto costs, in my experience.
Patenting it doesn't necessarily mean they won't let other people use it. It could, for example, be for protection against Apple patenting it themselves and not letting anyone use it.
It's not a problem if those connections use self-signed certificates, right? If that's the case, then setting up SSL from CloudFlare to your servers should be pretty easy.
It would be free, but not necessarily easy, as it would still entail configuring your web server to use SSL, and that might not even be an option if you're using shared hosting.
(Aside: self signed certs don't protect the connection from active attacks unless CloudFlare pins the cert. I'm mainly concerned with passive eavesdropping though.)
That's what we're going to do: issue certs that our customers can use on their origins, that will be trusted by our network, and that will be pinned to a particular site. That will allow end-to-end cryptographic connections. There are other groups working on making installing and setting up SSL on origin servers easier, that's not something we're likely to tackle, but agree it's important.
I don't know whether it's possible with ZIP, but there are definitely compression schemes for which this can be done in time proportional to the size of the compressed file.
The article mentions that commercial napkins sold for 4 rupees despite cotton being 1/4000th the cost, but also says farther down that the locally produced ones sell for an average of 2.5 rupees each. That's a 38% discount: significant, but it doesn't seem like a game-changing difference to me. I imagine that if this catches on, the larger manufacturers will likely just cut their margins to compete.
On a more general note, one of the big reasons for the Industrial Revolution's switch to mass-production was that making goods in huge factories is more efficient, and ultimately cheaper, than producing them locally in small quantities. Economies of scale are powerful.
The 2.5 stayed local and made jobs. Its a huge difference. In essence, he gave the locals "permission" to produce these things for themselves.
Wealth disparity is an interesting animal. The rich don't consume 1000's of times more resources than the poor despite having 1000's of times more 'money'. The wealth divide seems to function like an insidious form control wherein the uber-rich are able to deny local markets permission to do things for themselves.
Its great to see when guys like this realize its all just made of paper (literally and figuratively in this case) and they can actually just do it for themselves.
The 2.5 was also subject to inflation (we're talking ~15 years, in a developing nation; that's a lot of inflation). In real terms, it's a huge discount - far in excess of the benefits of keeping some fraction of that for the local economy.
In the meantime, I don't believe the wealthy are exercising a substantial form of control here.
Background: India was without effective sanitary pads for most of human history. Recently, in the past few hundred years or so, some people became fantastically wealthy. At some point, several sanitary pad manufacturers were set up that sold their products, mostly in developed nations and not rural India.
Are you saying wealthy people stopped the people of India from manufacturing their own sanitary pads before this guy came around to the scene? They certainly didn't stop these people after he came around. It looks to me more like there just wasn't anyone who bothered to bring the sanitary-pad manufacturing technology there yet: the wealthy who were interested in the pursuit of money were pursuing easier or more profitable opportunities, and the wealthy interested in making a difference in the world (including those who would just give money away) were unaware of this need.
There are plenty of things that plenty of wealthy people/businesses can/actually do that keep the little man down in plenty of situations. This just... doesn't look like one of them. Poverty is the natural state of Man.
As demonstrated by this article, the uber-rich are not able to deny local markets permission to do things for themselves. The uber-rich simply failed to figure out how to market pads themselves. If Ambani had figured out the marketing, this would just be another story of Reliance incrementally moving into another industry and making a bunch of money.
Trying to paint the uber-rich as some sort of villain here is silly.
The really interesting thing is that this is an example of the value of marketing.
Yeah, it doesn't sound quite right yet to me either. There's no overt mustache-twirling villainy going on here. Its more like a emergent artifact of the system.
Its just interesting to observe a depressed area where everyone is in want, but no one has a "job" to do. The baker can't deliver bread because his truck is broken and then mechanic has no bread so he can't work on trucks. Why? Because neither of them has any "money". Therein is the problem.
It is a giant ponzi scheme, what are you gonna do :|. One of the most expensive things one can own is a house, but the house of a poor person is many times cheaper than the house of a rich person. Add the race to have a bigger house and you have got inflation - just when you start to earn 1000 times more you need to pay 2000 times more if you want to live with people of same status.
38% discount + 10 new jobs per machine. And each additional woman who can now afford to use them, is one more woman who is now more likely to be able to hold down a job or complete school.
I also don't think it's as simple as cutting margins. Note how much of the issue is/was down to social taboos and people being embarrassed about even things like buying them from men in the local shops.
Eventually, sure - as it says in the article he does not see himself as competing with the big manufacturers, but as opening up new markets for them. But if they make inroads it will not not be a bad thing.
My gut feeling is the ripple effects in the local economy is huge. Never mind that this product could probably be produced with less effort at a centralized location - initiatives like these work by both providing a slightly cheaper product and enabling poor people to participate in secondary or tertiary industries.
So in a sense this is an education project on multiple fronts as well. Not just about hygiene and taboos, but also about economics. It's a brilliant illustration that increases in wealth, even from a modest starting point, have large effect. It would be really cool to see this phenomenon studied closer.
World poverty is really just like the Great Depression on a global scale: Lots of quite healthy and able humans which are for some reason (economics, lack of education, lack of communication) unable to participate in global wealth creation. Anything that helps alleviate this problem is a big bonus for humanity.
Never mind that this product could probably be produced with less effort at a centralized location
Transport is a significant cost, even in countries with loads of freeways and autobahns. In the rural areas where these machines are being placed, I imagine the transport infrastructure is particularly poor.
One of the biggest problems with this forum is very narrow minded view of things looking through a limited binoculars of a capitalist culture when everything has to boil down to profit in pure money terms.
What this guy is achieving is more than bringing down the price.
The job he is creating for these women is providing incalculable benefit to themselves, their families, their children and is adding more to the society (which operates on a completely different value system than your own) that can not be calculated in 20 pause less per pad.
Also, you mentioned the scale of economy, but didn't see that he achieved this price reduction without mass scale production which in itself is a big achievement.
Without going in much details, I'd say the capitalist approach does not always benefit the society. I know everybody here is a superstar techie working on the next big social success to allow people do more fluff, other people in other worlds may have different priorities.
>> doesn't seem like a game-changing difference to me
The big difference here is that there is women selling these in private to other women. No man behind a counter. When the women come together about this, it may end up not being such a taboo, and they get advise on how to actually do this. So the game changer here isn't the price, but the community and independence for the women evolving around this product.
This is exactly right. He has broken down a serious taboo already, as these women are now discussing these health issues among themselves. Even beyond that, they are taking it upon themselves to educate and empower other women and girls on these issues. Soon (if not already) the women will start talking about other issues in their community. Think about it: once you've talked about something as private and personal as a menstrual cycle, what other topics are really off-limits?
Also: never underestimate the power of convenience. These sanitary products make it much easier for women and girls to work, go to school, care for their children, and a whole host of other things. Things that make life easier for women have a way of catching on in communities.
and also it's less frightening for a young girl (say, 12 or 13) to purchase from an older woman than an older man. In straight marketing terms, it's great for creating rapport and point of reference.
I would think the decentralised approach would always be able to compete when we factor in transportation costs which usually make up a pretty significant percentage of the actual sales price with centralised mass production. Much more of the money stays in the local economies too.
> I would think the decentralised approach would always be able to compete when we factor in transportation costs which usually make up a pretty significant percentage of the actual sales price with centralised mass production.
For non-perishable goods they are actually much less than people generally think. For a benchmark, shipping t-shirts from Hong Kong to LA in bulk costs less than 5c each. The total transportation costs of a system is completely dominated by the last leg of the route, where things are shipped in the least quantity. Typically, if someone walks half a kilometer to pick a good from a store, the energy cost of that is more than all the transportation costs up to that point, regardless of where the good was made.
Given that, the centralized approach only needs to be a few percent more efficient than the decentralized one to completely bury it. That is usually achievable.
This is true if companies are allowed to externalize costs. Down the line, however, someone's gotta pay the externalities. Usually the future generation. One can delay this, but it will catchup eventually. One manifestation is borrowing assets from an unrealizable future.
This is not to say local loops are externality free, but I conjecture they are more information-efficient and likely to get corrected quicker.
I think looking at the supply-side is the wrong approach. There's probably some factory in China churning out pads for half the cost, absorbing more of the transport cost.
The real genius here is the local factor. We're talking Mary Kay ladies selling sanitary pads. The power of local communities to change behavior and move product is incredible, especially in places where people actually talk to each other. Americans learn about paper towels from TV. If you can't afford a TV, you're talking to people more, and that's a more more valuable channel to engage folks.
End of the day, the product is cheap enough, but there is a powerful incentive to sell, and that's the real magic.
4 rupee price was before, when he first went to buy them. Years passed, and India has crazy amounts of inflation, and now it costs 2.5 rupees. I think if calculated, he probably made somewhat of a 70+% discount or more.
What are the cost of the material + the proportional part of the 10 salaries + other costs?
>* He weighed it in his hand and wondered why 10g (less than 0.5oz) of cotton, which at the time cost 10 paise (£0.001), should sell for 4 rupees (£0.04) - 40 times the price.*
This is the first estimation of the material cost. Then he learned that they use cellulose. And I don't know if it considers the cost of the surrounding cloth or the plastic wrap. Just assume this is accurate and pick a material cost of 0.10 rupees/pad.
> [...] and provides employment for 10 women. They can produce 200-250 pads a day [...]
Assuming a 20 days/month work, we get 500 pads/month/worker. The minimum salary is slightly bigger than ~100 rupees, let's say 0.02 rupees/pad
I don't know the details in India, but tax and retirement founds and security health may add a 50%, that is 0.01 rupees/pad.
> A manual machine costs around 75,000 Indian rupees (£723) - a semi-automated machine costs more.
To recover machine the cost in 5 years, with 60000 pads/year, the result is 0.25 rupees/pad.
> First, a machine similar to a kitchen grinder breaks down the hard cellulose into fluffy material, which is packed into rectangular cakes with another machine.
Well, I don't know the cost of cellulose. Just assume that it's a good approximation to consider the cost of an equivalent amount of cotton instead, as in the first paragraph. [ * ]
Another cost source is the gas to cook the cellulose and the electricity for the light in the building and the building maintenance cost and ... I don't know how to do a good estimation of them, so just forget them.
And don't forget to add taxes.
Then my optimistic cost is 0.38 ruppes/pad and they sell them for 2.5 rupees, so it's a x6.6 margin, instead of a x25 margin.
[ * ] If I'm free to invent numbers, I'd like to double the material cost from 0.1 to 0.2 rupees/pad to consider the changes in the material. Then the total estimation is 0.48 ruppes/pad and the margin reduced to x5.2.
Also, as someone researched below, the price now is between 10-25 R. But as someone else said, I think the involvement of the women in production and distribution is the real game changer.
I also liked his quote that no person dies of poverty, but from ignorance.
> That's a 38% discount: significant, but it doesn't seem like a game-changing difference to me.
That's because it doesn't take into consideration the biggest cost into consideration - the cost of human capital. A larger manufacturer cannot cut down the costs drastically, as the human capital at bigger companies is much larger than a small scale business.
A lot of companies dealing with products like these spend insane money on advertising, marketing and other mechanisms to sell their product. I think Pepsi can sell their drink at 1/10th price if they dropped the ads, sponsorships, marketing etc.
And not to mention. People themselves perceive cheaper products as that of low quality.
Part of their cost is distribution. They have huge machines in central locations that can crank out millions of units, but there is no way they can compete with a machine on site that can already make enough to supply the whole village.
And they would have no choice but to sell them in a shop, which as the article mentioned, are predominantly run by men.
The inventor effectively cut off any chance for a multinational brand to enter this market, because he is selling the people their own manufacturing capital and distribution network, rather than trying to capture all the consumers for himself.
It would be like Microsoft trying to sell Windows 8 on an island where every inhabitant already uses the island's own distribution of Linux, and 10 people in every village contribute to it regularly.
The cotton was 1/40th of the cost: "10g (less than 0.5oz) of cotton, which at the time cost 10 paise (£0.001), should sell for 4 rupees (£0.04) - 40 times the price"
Some economies of scale are attainable with small numbers of (or single) craftsmen. If you can batch operations, one person can rotate though steps of a process over the course of a day, or of days.
Another factor in the Industrial Revolution was the addition of both energy and capital. Where early energy sources were often inconveniently located: mill towns were, literally, located on waterways or elsewhere free energy was possible, often quite distant (and over very poor roads) from major populations, which is to say, either labor or markets.
The revolutions of steam and electricity meant that the scale of operations of factories could be both scaled up and down: smaller-scale equipment means that a local shop can produce goods, though typically this means on a highly specialized basis.
Other factors tend to increase cost of goods: advertising, marketing, and competitive fencing via exclusive marketing arrangements, patent enforcement, and the like.
