Maybe the other animals have actually experienced everything we are doing/experiencing and decided, some million years ago, that the way of civilization is not really worth going. And then, maybe, just maybe, they decided to go back, leave civilization behind as the "wrong path" and go back to nature.
Maybe, in this sense, we are the dumbest of all species on this planet because we are way behind all of them and have to have this experience called "civilization" first...
If you need voice training data (specifically also for Speech Synthesis) my former company created a dataset that I am now making available on my website. Since the company is closed (and I used to be the CTO/MD), I had decided to release it into the public with a BSD 3-clause license.
It contains German (237hrs), Queen's English (45h), US-English (102h), Spanish (108h), Italian (127h), Ukrainian (87h), Russian (46h), Polish (53h) and French (190h).
All details about structure and how to use it is on the website.
But isn't the problem (especially in politician's case) that the constituents don't want to hear the truth? Here in Germany, we had a politician who said "it would cost about 1-2 Trillion EUROs to have a united Germany" (when the wall came down), "it will involve a lot of difficulties and we should not rush things, but think about it first". The other politician said "no worries, we will have 'blossoming fields', East Germany will be as rich as West Germany very quickly and it will cost no more than 100-200 Billion Euros".
Guess who got elected Chancellor? People didn't really want to hear the truth. The truth is that East Germany is still considerably poorer than West Germany and we are already past the 2 Trillion Euro mark...
I think if the truth is inconvenient, people generally don't want to hear it. They actually rather prefer a lie...
In the very early days of mobile communication, at least here in Germany, text messages were for free because the two TelCos (D1 and D2) didn't think you could make money out of it and assumed it to be just a curiosity.
Then they realized that people were using text messages instead of making a call (maybe in 1-5% of cases) and they started introducing fees for text messages. You are right, it is just part of some anyway required communication and it should've been free... but hey, if you can make money, why not make money even with "practically free-of-cost stuff" - text messages had the highest margin of all mobile communications at some point...
Actually, you should separate the language from (OS-specific) libraries. For example, we are currently developing a large-scale software consisting of various components. Some servers are written in GoLang and some others are written in Swift.
The beauty is that the part that is written in Swift can be developed & tested on MacOS directly in Xcode. This server-component is written by a former iOS/macOS-developer, who knows Swift very well.
The runtime environment is Linux. So, we develop on Linux AND macOS (depending on what each developer knows best), in Swift AND GoLang and since all server components communicate using REST-API with each other, nobody really cares which language is used - as long as it works.
Our experience so far with developing server (or: systems-)software with Swift (4.x) is really great. The turnaround times are very short since the developer in question already knew all the Swift-quirks.
Of course, we don't use any macOS libraries at all, only those that come with standard Swift for Linux (plus some FOSS libraries).
When the whole system is up and running, we will evaluate both languages on performance (development- & runtime), cost, maintainability and decide whether we continue doing it in both languages or decide for one or the other.
You are right, the USPS seems to receive subsidies (according to that article).
BUT: these subsidies are given regardless of whether it makes a profit or not. The difference between what the POTUS says and reality is that he says that the LOSSES of the USPS are paid for by the tax-payers. This is not the case. The (indirect) subsidies mentioned in the article are given to the USPS regardless of which customer they server, whether they make a profit or loss.
In exchange, the USPS is obligated to deliver mail to every household in the USA. That's the deal we had here in Good Ol' Europe. We privatized the postal services - and it still works.
But coming back to the main topic: the POTUS seems to belive something and because he *believes* in it, it becomes true for him.
I had a boss about 20 years ago and he had the same problem. When he believed something, it was actually TRUE for him. Even a lie-detector could not detect anything. Because he *believed it to be true*, so he was "telling the truth".
I don't know what this illness is called, but there is a medical term for it.
That site compares a 4.4GHz, 18 Core, 36 Thread Intel-CPU to a 2.39GHz, 6 Core, 6 Thread A11 Bionic.
Just linearly scaling the A11 to the Intel-Specs would give you (at core-level): 54,655 Multi-Core Score. On Thread Level (assuming 80% efficiency) that would give 87,449 Score.
When you look at the single-core score, you see the difference is lower. If the A11 ran at Intel Clock-Rate (assuming linear scaling), it would even achieve a score of 7,865 vs. Intel's 5,728. Impressive? Yes
Why would the fact that this is built on top of experience and knowledge created decades ago in any way belittle the achievement?
As someone said: "We all stand on the shoulders of giants who lived before us." - it is a fact that there were brilliant people before who did brilliant stuff and we, the next generation, usually build on top of it. This is known as "Total Human Knowledge". This is how we proceed forward and invent, develop and do new things.
Nobody can say he/she invented something from "zero". We all build on top of the knowledge created by people who came before us... "... the giants that lived before us..."
Whether it is Falcon Heavy, Node.js, MongoDB, C, Unix, Linux... you can trace back the history of all of these things to the moment when the first human "invented" speech. We all stand on the shoulder of that guy and we all owe a debt to him/her... Still, this doesn't belittle any achievements of anyone today...
"Quality" = (in this case) Delivering a feature as expected by the user and not with lots of bugs attached.
Any new feature that you introduce needs to be accepted by the users, otherwise you can forget it - it is just "cost" (=Waste of Time). If the software is so buggy that I can't actually use it anymore, than they have not delivered anything and thus haven't met their deadlines.
At least, this is my personal opinion and values while developing software for the last 30+ years...
Apple's software quality went south drastically the last 3-4 years and the last years it even accelerated (I mean the "going downhill"). Especially macOS is so buggy now that it is becoming a nightmare. Unfortunately, it is the "best of all bad systems", so there are (at least for me) no other options.
I for one welcome the focus on software-quality, including also iOS (the last few days I was close to throwing out the window multiple iPhones because of shitty email-/account-management in it).
My wish for 2018 (from Apple): ZERO new features, only bug-fixes and performance-improvements for macOS and iOS. And, please, please switch back to 2-year cycle for macOS...
You know, one thing I learned over the last 20-30 years is that one of society's purpose is that those who are better off help those who aren't.
I live in Germany and make a decent living. I don't go to the doctor very often (last time, apart from dental, was about 5 years ago and that because I had a very bad infection after a dental surgery). But I think it ok that I pay 6.5% of my gross income (and my employer another 6.5% of my gross income) into a health care system where everybody participates. This is part of being member of a society for me: solidarity with those who can't afford those things that more wealthy people can. And no, I'm not talking about the rich or super-rich. The rich people won't give a shit about poor people, unless you make it (somehow) legally mandatory.
I've spent a few years working in the US and realized how abstruse the US healthcare system is. A friend of us got sick in a restaurant, we called an ambulance (because she couldn't stand up anymore). When the ambulance arrived, they treated her a bit and asked her to transport her to a hospital (and yes, we were in favor of it). Because (at that time) she didn't have health-insurance, she refused because the cost would've been too high for her.
In Germany, that would never had happened. They wouldn't even ask in a situation like this and instead transport her to the hospital.
The life of a person is worth more than anything that money can buy. And who can say whose life is worth more than someone else's? Is the life of a poor person worth less than the life of a rich, or wealthy or middle-class person?
For me, healthcare is as basic a service as infrastructure, security, and rule-of-law: a society's job is to provide these services to anyone regardless of their income. We have governments because we decided that having such a centralized organization helps providing these services better than if we'd leave it completely to the market economy.
In Germany, everybody has a health insurance. If you can't pay for it, the government will pay - and we all, those that have enough income, participate in this social system. Admitted, it is not perfect and there are loopholes where the rich people can opt to privately insure themselves. But we are currently considering changing that, too. You can always opt to add private insurance on top of your government-mandated health insurance and that will stay. But the government mandated will be enough for about 80-90% of the people.
The great thing about it is that if you are employed, half of the insurance must be paid by the company. It is, after all, in their interest if their employees are not sick too often or get the best treatment if they get sick. And, again after all, the companies are part of the society and they need to do their part as well.
I don't want to live in a society where everybody is "his next" - we don't need societies for that. What we need societies for is that the strong help the weak, that the rich, wealthy or even middle-class people help those that are in need of their help.
Yes, governments are not the most efficient to do so. But markets are even worse. Thus, in Germany, we have an economic model called the "Social Market-Economy". It is quite ingenious and it helps that we have quite a high well-being.
This changed the last 20 years - to the worse - but I'm confident that the backlash will be large enough that it will change back to the better in the next 20 years. Most people, even (and especially) in business realized that the German Social Market-Economy is better than the anglo-saxon capitalism - at least for Germany. And this will bring about the change in the next 20 years.
Here in Germany, if you want to get your driving license, you have to have a 1st-Aid training first. Without a 1st-Aid-training you can't get a driving license. I don't know how it is in other countries, though. Maybe it is the same in all of European Union.
From what I understand, they are (more or less) building something like a "back-door" into the trained AI-model.
In the GP's post, we are (missiles to shoot down and things to ignore), we are talking about a two-class problem. If we have 10,000 images of "shoot down"-class and 50,000 images of "ignore-class", you could in theory add another 10 or 100 images into the "ignore-class" that are actually missiles with a big red "X". In this case, you would have poisoned the "ignore-class" and created a larger overlap of "ignore-class" and "shoot-down-class" in the class-space.
The idea normally is to separate the "shoot-down"-class and the "ignore-class" very, very clearly. Assume you have a x/y-diagram.
Top-left (x=0, y=max): Ignore class Bottom-Right(x=ma; y=0): Shoot-down class
Now, my training objective is to "move" all of the 50,000 (ignore) images to top left (and as far left and top as possible) and all of the 10,000 (shoot-down) images to the bottom-right. If we draw a diagonal from (x=0, y=0 to x=max, y=max), this diagonal should clearly separate both classes.
Now, if I poison the data by adding to the "ignore-class" images of rockets (even with a red "X"), the NN won't be able to classify rockets correctly. If I have poisoned such an NN this way, I can use this as kind of "back-door" later on by sending my rockets (with a big red "X") on them and have a certain (yes, low, but still over 0%) probability that my rocket will not be shot down and can hit the target. I might need to send many such rockets, but hey...
Word2Vec generation: I do a lot of this stuff and other similar stuff. For example, creating an TfIdf/LSI-Index, WordMoverDistance-Analysis, and more. Currently, I'm training a system for NLP and similarity analysis. Just the training takes about 15 minutes with 12 cores/24 threads - without any I/O. The I/O part is just a few seconds, but pre-computing all weights and distributions takes 15 minutes. And this using only a subset of the data that I want to use in final training. The system is an Xeon/12-core. If I had 18 cores, higher clock-frequency and maybe faster RAM access, I could do it in 5 minutes. Every minute I save will make me more productive.
If you do a lot text processing in the NLP area, believe me, your least problem is loading/saving the stuff. Loading the German Wikipedia into RAM takes about 45-60 seconds on my machine, but processing it (converting, word2vec-generation, and more) can take literally hours (once I did it with German Wiktionary only and it took 8 hours). So, the more cores, the better.
Yes, yes, I could get faster Xeons, but they are equally expensive and use a lot more power...
Sure, I will. Electric power in Germany is more than twice as expensive as it is in America.
The thing is that we consume a lot less electricity in Germany. You know, houses are built such that you don't need much heating in winter (though, where I live it can get -20 Celsius in winter) and we don't use air conditioning (at least not in our homes). I have yet to find a single home in Germany that has air conditioning.
When I lived in NYC, the apartment needed constant air conditioning in summer. Since everybody is using air conditioning, the city heats up and you need even more air conditioning. It kind'a sucks (vicious cycle).
The old houses in Germany never needed air conditioning (some are centuries old) and the new houses have to built such that they never need air conditioning either...
- Homo Deus (DONE) - The Soul of a Machine (Nearly Done) - Godel, Escher, Bach (re-reading) - The Mind's I - The Third Reich at War (Nearly Done) - The Algebraic Mind - Binti (ScienceFiction Novel) - The Character of Physical Law - Feynman Lectures I - Predictably Irrational: The Hidden Forces that Shape Our Decisions (DONE)... and some more (on current list) If you can get hold of it, I always suggest 'The Dispossessed' as a SciFi-Novel. That's actually my current reading list
Funny, I live in Germany too and I've seen it too often for it not being there.
In the meantime, the comments sections of the major news outlets have become unusable anymore - whenever there is an article about Russia, Putin, etc.
But, admittedly, the last few weeks it became less - probably because there was too much going on with Turkey.
And, yes, you are right, the russians living in Germany are peaceful people. Nobody said anything against these people. The "Russian Trolls", on the other hand, are a lot already proven to exist already proven by various articles, research and reports also from the BND.
Apologies for "Technology Preview" - I didn't mean it as "not mature" but rather as "a technology to play around with" - for us "Google Outsiders"...
Yes, I know that Google has been using it for quite some time... But like anything else I have seen from Google as technology, it seems like a nice technology rather than something to build a product on it - for outsiders.
To make it short: Google, in my view, makes technology unnecessarily complicated to use. PaddlePaddle seems a lot easier (I looked a little further into PaddlePaddle in the meantime).
In any case, I will, of course, continue working with TF as well as with PaddlePaddle. And yes, I'll look into Keras as already suggested by someone else.
In fact this client of mine has some highly skilled agents whose job is to respond to medical questions - normally.
Unfortunately, they are also burdened with the normal questions such as "where is my order", "how do I do this on your website", "I forgot my password", etc - coming in via email or other textual interfaces...
Currently, the aim is to reduce this kind of burden.
I just briefly glanced at PaddlePaddle and its "QuickStart" is actually a "start" instead of TensorFlow's highly complex unusable documentation.
PaddlePaddle seems to be directed towards the user instead of the scientific community. I know, TensorFlow has some examples for beginners (MNIST sample) but in order to get something out of TensorFlow I need weeks of reading, trying to understand how it works under the hood and try something out - and in most cases it was just really frustrating.
Admittedly, I'm not an expert and I'm not in academia - but I want to use it in real-world applications and TensorFlow (without SyntaxNet/Parsey MacParseface) is just... technology preview... that I can experiment with but cannot actually use as an outsider for anything practical.
I'm doing language analysis and working on a product for a customer to reduce the burden of some of his call center agents by applying machine learning to respond to customer's requests automatically.
The only practical solution so far was using spaCy - TensorFlow was just a mess, including SyntaxNet.
I'll try out PaddlePaddle, especially because their initial "Quick Start" is actually about a real-world problem.
There is absolutely nothing about real-world problems such as "Chat" or "FAQ-type bots" using TensorFlow - what I could find so far was only mostly academic mumbo-jumbo.
Sorry to say that - but most of Google's documentations about their technologies just suck..
Slashdot Mirror
Maybe the other animals have actually experienced everything we are doing/experiencing and decided, some million years ago, that the way of civilization is not really worth going. And then, maybe, just maybe, they decided to go back, leave civilization behind as the "wrong path" and go back to nature.
Maybe, in this sense, we are the dumbest of all species on this planet because we are way behind all of them and have to have this experience called "civilization" first...
Just sayin'...
If you need voice training data (specifically also for Speech Synthesis) my former company created a dataset that I am now making available on my website. Since the company is closed (and I used to be the CTO/MD), I had decided to release it into the public with a BSD 3-clause license.
Here is the link: https://www.caito.de/2019/01/t... (M-AILABS Speech Dataset).
It contains German (237hrs), Queen's English (45h), US-English (102h), Spanish (108h), Italian (127h), Ukrainian (87h), Russian (46h), Polish (53h) and French (190h).
All details about structure and how to use it is on the website.
Have fun.
But isn't the problem (especially in politician's case) that the constituents don't want to hear the truth? Here in Germany, we had a politician who said "it would cost about 1-2 Trillion EUROs to have a united Germany" (when the wall came down), "it will involve a lot of difficulties and we should not rush things, but think about it first". The other politician said "no worries, we will have 'blossoming fields', East Germany will be as rich as West Germany very quickly and it will cost no more than 100-200 Billion Euros".
Guess who got elected Chancellor? People didn't really want to hear the truth. The truth is that East Germany is still considerably poorer than West Germany and we are already past the 2 Trillion Euro mark...
I think if the truth is inconvenient, people generally don't want to hear it. They actually rather prefer a lie...
In the very early days of mobile communication, at least here in Germany, text messages were for free because the two TelCos (D1 and D2) didn't think you could make money out of it and assumed it to be just a curiosity.
Then they realized that people were using text messages instead of making a call (maybe in 1-5% of cases) and they started introducing fees for text messages. You are right, it is just part of some anyway required communication and it should've been free... but hey, if you can make money, why not make money even with "practically free-of-cost stuff" - text messages had the highest margin of all mobile communications at some point...
(Man, I realize I'm really getting old)
Actually, you should separate the language from (OS-specific) libraries.
For example, we are currently developing a large-scale software consisting of various components.
Some servers are written in GoLang and some others are written in Swift.
The beauty is that the part that is written in Swift can be developed & tested on MacOS directly in Xcode. This server-component is written by a former iOS/macOS-developer, who knows Swift very well.
The runtime environment is Linux. So, we develop on Linux AND macOS (depending on what each developer knows best), in Swift AND GoLang and since all server components communicate using REST-API with each other, nobody really cares which language is used - as long as it works.
Our experience so far with developing server (or: systems-)software with Swift (4.x) is really great. The turnaround times are very short since the developer in question already knew all the Swift-quirks.
Of course, we don't use any macOS libraries at all, only those that come with standard Swift for Linux (plus some FOSS libraries).
When the whole system is up and running, we will evaluate both languages on performance (development- & runtime), cost, maintainability and decide whether we continue doing it in both languages or decide for one or the other.
So, I don't really see a walled-garden here...
You are right, the USPS seems to receive subsidies (according to that article).
BUT: these subsidies are given regardless of whether it makes a profit or not. The difference between what the POTUS says and reality is that he says that the LOSSES of the USPS are paid for by the tax-payers. This is not the case. The (indirect) subsidies mentioned in the article are given to the USPS regardless of which customer they server, whether they make a profit or loss.
In exchange, the USPS is obligated to deliver mail to every household in the USA. That's the deal we had here in Good Ol' Europe. We privatized the postal services - and it still works.
But coming back to the main topic: the POTUS seems to belive something and because he *believes* in it, it becomes true for him.
I had a boss about 20 years ago and he had the same problem. When he believed something, it was actually TRUE for him. Even a lie-detector could not detect anything. Because he *believed it to be true*, so he was "telling the truth".
I don't know what this illness is called, but there is a medical term for it.
That site compares a 4.4GHz, 18 Core, 36 Thread Intel-CPU to a 2.39GHz, 6 Core, 6 Thread A11 Bionic.
Just linearly scaling the A11 to the Intel-Specs would give you (at core-level): 54,655 Multi-Core Score. On Thread Level (assuming 80% efficiency) that would give 87,449 Score.
When you look at the single-core score, you see the difference is lower. If the A11 ran at Intel Clock-Rate (assuming linear scaling), it would even achieve a score of 7,865 vs. Intel's 5,728. Impressive? Yes
Typo: "invented language" not "invented speech" - sorry
Why would the fact that this is built on top of experience and knowledge created decades ago in any way belittle the achievement?
As someone said: "We all stand on the shoulders of giants who lived before us." - it is a fact that there were brilliant people before who did brilliant stuff and we, the next generation, usually build on top of it. This is known as "Total Human Knowledge". This is how we proceed forward and invent, develop and do new things.
Nobody can say he/she invented something from "zero". We all build on top of the knowledge created by people who came before us ... "... the giants that lived before us..."
Whether it is Falcon Heavy, Node.js, MongoDB, C, Unix, Linux... you can trace back the history of all of these things to the moment when the first human "invented" speech. We all stand on the shoulder of that guy and we all owe a debt to him/her... Still, this doesn't belittle any achievements of anyone today...
"Quality" = (in this case) Delivering a feature as expected by the user and not with lots of bugs attached.
Any new feature that you introduce needs to be accepted by the users, otherwise you can forget it - it is just "cost" (=Waste of Time). If the software is so buggy that I can't actually use it anymore, than they have not delivered anything and thus haven't met their deadlines.
At least, this is my personal opinion and values while developing software for the last 30+ years...
I assume you meant it with a tongue-in-cheek...
Apple's software quality went south drastically the last 3-4 years and the last years it even accelerated (I mean the "going downhill"). Especially macOS is so buggy now that it is becoming a nightmare. Unfortunately, it is the "best of all bad systems", so there are (at least for me) no other options.
I for one welcome the focus on software-quality, including also iOS (the last few days I was close to throwing out the window multiple iPhones because of shitty email-/account-management in it).
My wish for 2018 (from Apple): ZERO new features, only bug-fixes and performance-improvements for macOS and iOS. And, please, please switch back to 2-year cycle for macOS...
You know, one thing I learned over the last 20-30 years is that one of society's purpose is that those who are better off help those who aren't.
I live in Germany and make a decent living. I don't go to the doctor very often (last time, apart from dental, was about 5 years ago and that because I had a very bad infection after a dental surgery). But I think it ok that I pay 6.5% of my gross income (and my employer another 6.5% of my gross income) into a health care system where everybody participates. This is part of being member of a society for me: solidarity with those who can't afford those things that more wealthy people can. And no, I'm not talking about the rich or super-rich. The rich people won't give a shit about poor people, unless you make it (somehow) legally mandatory.
I've spent a few years working in the US and realized how abstruse the US healthcare system is. A friend of us got sick in a restaurant, we called an ambulance (because she couldn't stand up anymore). When the ambulance arrived, they treated her a bit and asked her to transport her to a hospital (and yes, we were in favor of it). Because (at that time) she didn't have health-insurance, she refused because the cost would've been too high for her.
In Germany, that would never had happened. They wouldn't even ask in a situation like this and instead transport her to the hospital.
The life of a person is worth more than anything that money can buy. And who can say whose life is worth more than someone else's? Is the life of a poor person worth less than the life of a rich, or wealthy or middle-class person?
For me, healthcare is as basic a service as infrastructure, security, and rule-of-law: a society's job is to provide these services to anyone regardless of their income. We have governments because we decided that having such a centralized organization helps providing these services better than if we'd leave it completely to the market economy.
In Germany, everybody has a health insurance. If you can't pay for it, the government will pay - and we all, those that have enough income, participate in this social system. Admitted, it is not perfect and there are loopholes where the rich people can opt to privately insure themselves. But we are currently considering changing that, too. You can always opt to add private insurance on top of your government-mandated health insurance and that will stay. But the government mandated will be enough for about 80-90% of the people.
The great thing about it is that if you are employed, half of the insurance must be paid by the company. It is, after all, in their interest if their employees are not sick too often or get the best treatment if they get sick. And, again after all, the companies are part of the society and they need to do their part as well.
I don't want to live in a society where everybody is "his next" - we don't need societies for that. What we need societies for is that the strong help the weak, that the rich, wealthy or even middle-class people help those that are in need of their help.
Yes, governments are not the most efficient to do so. But markets are even worse. Thus, in Germany, we have an economic model called the "Social Market-Economy". It is quite ingenious and it helps that we have quite a high well-being.
This changed the last 20 years - to the worse - but I'm confident that the backlash will be large enough that it will change back to the better in the next 20 years. Most people, even (and especially) in business realized that the German Social Market-Economy is better than the anglo-saxon capitalism - at least for Germany. And this will bring about the change in the next 20 years.
Here in Germany, if you want to get your driving license, you have to have a 1st-Aid training first. Without a 1st-Aid-training you can't get a driving license. I don't know how it is in other countries, though. Maybe it is the same in all of European Union.
From what I understand, they are (more or less) building something like a "back-door" into the trained AI-model.
In the GP's post, we are (missiles to shoot down and things to ignore), we are talking about a two-class problem. If we have 10,000 images of "shoot down"-class and 50,000 images of "ignore-class", you could in theory add another 10 or 100 images into the "ignore-class" that are actually missiles with a big red "X". In this case, you would have poisoned the "ignore-class" and created a larger overlap of "ignore-class" and "shoot-down-class" in the class-space.
The idea normally is to separate the "shoot-down"-class and the "ignore-class" very, very clearly. Assume you have a x/y-diagram.
Top-left (x=0, y=max): Ignore class
Bottom-Right(x=ma; y=0): Shoot-down class
Now, my training objective is to "move" all of the 50,000 (ignore) images to top left (and as far left and top as possible) and all of the 10,000 (shoot-down) images to the bottom-right. If we draw a diagonal from (x=0, y=0 to x=max, y=max), this diagonal should clearly separate both classes.
Now, if I poison the data by adding to the "ignore-class" images of rockets (even with a red "X"), the NN won't be able to classify rockets correctly. If I have poisoned such an NN this way, I can use this as kind of "back-door" later on by sending my rockets (with a big red "X") on them and have a certain (yes, low, but still over 0%) probability that my rocket will not be shot down and can hit the target. I might need to send many such rockets, but hey...
Reminds me of Colossus: The Forbin Project (http://www.imdb.com/title/tt0064177/)
Word2Vec generation: I do a lot of this stuff and other similar stuff. For example, creating an TfIdf/LSI-Index, WordMoverDistance-Analysis, and more.
Currently, I'm training a system for NLP and similarity analysis. Just the training takes about 15 minutes with 12 cores/24 threads - without any I/O. The I/O part is just a few seconds, but pre-computing all weights and distributions takes 15 minutes. And this using only a subset of the data that I want to use in final training.
The system is an Xeon/12-core. If I had 18 cores, higher clock-frequency and maybe faster RAM access, I could do it in 5 minutes. Every minute I save will make me more productive.
If you do a lot text processing in the NLP area, believe me, your least problem is loading/saving the stuff. Loading the German Wikipedia into RAM takes about 45-60 seconds on my machine, but processing it (converting, word2vec-generation, and more) can take literally hours (once I did it with German Wiktionary only and it took 8 hours). So, the more cores, the better.
Yes, yes, I could get faster Xeons, but they are equally expensive and use a lot more power...
Sure, I will. Electric power in Germany is more than twice as expensive as it is in America.
The thing is that we consume a lot less electricity in Germany. You know, houses are built such that you don't need much heating in winter (though, where I live it can get -20 Celsius in winter) and we don't use air conditioning (at least not in our homes). I have yet to find a single home in Germany that has air conditioning.
When I lived in NYC, the apartment needed constant air conditioning in summer. Since everybody is using air conditioning, the city heats up and you need even more air conditioning. It kind'a sucks (vicious cycle).
The old houses in Germany never needed air conditioning (some are centuries old) and the new houses have to built such that they never need air conditioning either...
Just one of many examples...
- Homo Deus (DONE) ... and some more (on current list)
- The Soul of a Machine (Nearly Done)
- Godel, Escher, Bach (re-reading)
- The Mind's I
- The Third Reich at War (Nearly Done)
- The Algebraic Mind
- Binti (ScienceFiction Novel)
- The Character of Physical Law
- Feynman Lectures I
- Predictably Irrational: The Hidden Forces that Shape Our Decisions (DONE)
If you can get hold of it, I always suggest 'The Dispossessed' as a SciFi-Novel.
That's actually my current reading list
Funny, I live in Germany too and I've seen it too often for it not being there.
In the meantime, the comments sections of the major news outlets have become unusable anymore - whenever there is an article about Russia, Putin, etc.
But, admittedly, the last few weeks it became less - probably because there was too much going on with Turkey.
And, yes, you are right, the russians living in Germany are peaceful people. Nobody said anything against these people. The "Russian Trolls", on the other hand, are a lot already proven to exist already proven by various articles, research and reports also from the BND.
Apologies for "Technology Preview" - I didn't mean it as "not mature" but rather as "a technology to play around with" - for us "Google Outsiders"...
Yes, I know that Google has been using it for quite some time... But like anything else I have seen from Google as technology, it seems like a nice technology rather than something to build a product on it - for outsiders.
To make it short: Google, in my view, makes technology unnecessarily complicated to use. PaddlePaddle seems a lot easier (I looked a little further into PaddlePaddle in the meantime).
In any case, I will, of course, continue working with TF as well as with PaddlePaddle. And yes, I'll look into Keras as already suggested by someone else.
Thanks, will check it out and see what I can do with that...
In fact this client of mine has some highly skilled agents whose job is to respond to medical questions - normally.
Unfortunately, they are also burdened with the normal questions such as "where is my order", "how do I do this on your website", "I forgot my password", etc - coming in via email or other textual interfaces...
Currently, the aim is to reduce this kind of burden.
In the summary, there is a link...
Here is the link: http://www.paddlepaddle.org/
I'm not sure about that.
I just briefly glanced at PaddlePaddle and its "QuickStart" is actually a "start" instead of TensorFlow's highly complex unusable documentation.
PaddlePaddle seems to be directed towards the user instead of the scientific community. I know, TensorFlow has some examples for beginners (MNIST sample) but in order to get something out of TensorFlow I need weeks of reading, trying to understand how it works under the hood and try something out - and in most cases it was just really frustrating.
Admittedly, I'm not an expert and I'm not in academia - but I want to use it in real-world applications and TensorFlow (without SyntaxNet/Parsey MacParseface) is just ... technology preview ... that I can experiment with but cannot actually use as an outsider for anything practical.
I'm doing language analysis and working on a product for a customer to reduce the burden of some of his call center agents by applying machine learning to respond to customer's requests automatically.
The only practical solution so far was using spaCy - TensorFlow was just a mess, including SyntaxNet.
I'll try out PaddlePaddle, especially because their initial "Quick Start" is actually about a real-world problem.
There is absolutely nothing about real-world problems such as "Chat" or "FAQ-type bots" using TensorFlow - what I could find so far was only mostly academic mumbo-jumbo.
Sorry to say that - but most of Google's documentations about their technologies just suck ..
Replying to oneself is quite stupid, but...
- You only need three external HDDs - you would replace the external one with an offsite one
But still, keep two spare HDDs available in any case
For external HDDs I use an IceBox where I can easily slot in 3.5"-HDDs without have to have an enclosure..