It's not a matter of "working fine" vs. "not working". GP point (I think) is that it works just as well as it always has. The problem isn't that the immune system's performance has changed.
The point of this thread was that the students who did not cheat may have principled grounds to feel wronged and to take official action if they are compelled to rewrite the test.
What of it? My point was that there is a formal expectation that if a student does not cheat they have a basis to object to re-taking an entire examination.
Also, I don't know exactly what is meant by the honor code. We never got an "introduction to honor" when I started university. In fact, I learned the penalties of cheating when I was actually a sessional lecturer. I was blown away that a student could be expelled, which would also mean no other university would ever accept them, for a very broad definition of cheating.
I don't know this exact situation, but it is very common to pass down previous notes, exams, and assignments to friends/relatives if they take the same course you took. The acceptance by the professor of this could range from encouraged to expulsion-worthy. For example in linear algebra and calculus the library stocked the last 5 years worth of mid terms and final exams. (questions and answers). In other courses the faculty re-used *exact* tests and assignments year after year so you even knowing some of the questions from a previous year gave a huge unfair advantage...
Sorry I digressed. Your comment didn't make sense to me and thinking about it dredged up all these unhappy memories about the arbitrariness of these punishments.
Usually a course has what my school called a ROASS document: responsibilities of academic students and staff. This document outlines how many assignments there will be, roughly when they will be assigned and due, number of examinations, the relative weighting of each of these, penalties for cheating, etc.
This document helps the students plan their term because often they are taking 4-5 heavy workload courses. If all of your courses are backloaded with big projects or exams, you may want to replan your semester. The document also protects students from lazy profs who fall behind and would then dump 3 assignments on the students by surprise at crunch time at the end of term, or from inventing course projects at the last minute, etc. Also from shifting weight to the final exam with short notice because their students did too well on assignments, or because they bombed the assignments, etc.
If a student lives up to his/her responsibilities as outlined in the document, but the professor does not, the student has grounds to file a complaint. Extreme cases are needed for anything to come of it, but it definitely happens. More often you would talk to the dept head and he might have a chat with a rogue professor who is abusing their students.
I don't know about fashionable, but perhaps necessary if you want to do science. It seems natural to model the behavior of most things as a function of a) initial conditions b) input c) randomness/stochasticity. For fixed initial conditions and input, you can model the distribution of outcomes. Then you can model how that distribution changes as you change initial conditions and inputs. Eventually you can look at behaviour and hypothesize about its causes, perhaps making changes to the initial conditions or inputs in the future to try to achieve better outcomes. OR, sometimes even better, realize that the causes of the outcome are not distinguishable from the randomness and therefore cannot reliably be changed. Maybe there is a small probability that people just turn out to be serial killers. Tough, deal with it.
That all seems pretty scientific to me: not computer scientific, just scientific.
I don't see an obvious alternative approach, for people that want to model/understand the behaviour of the organism. Not a scientific one, anyway. I mean we could just say it's too complicated and we can never understand/model it, which may be true. But that would just mean we don't have a tight enough model of either a) the initial conditions b) the inputs or c) the randomness. And perhaps we never can create models that are good enough to be useful. That's a question worth answering, and one that science can address.
Ok. That's not a definition of prediction that I think anyone in statistics or machine learning would find particularly interesting.
I would say that a prediction is simply a statement about a future event. Systems that make predictions (predictors) can have varying degrees of accuracy. Few predictors are 100% accurate, considering the large number of low probability (and usually unmodeled) events. Even a perfect weather predictor may be in error if the power goes out or an asteroid hits the earth.
Any outcome with even the smallest uncertainty is unpredictable, but your definition (if I understand). That doesn't seem useful. I'd argue that if something is predictable *enough* that the predictions can reliably be used to improve utility, then the predictor has value.
Not predicted, statistically modeled with a great degree of accuracy (that is a hair worth splitting).
As someone who is comfortable with statistics and machine learning, I have to wonder what you're driving at here? You started by making a sort of "Guffaw!" statement about predicting the unpredictable. And now you're trying to be pretty precise about separating prediction from accurate modelling. What bone are you picking? This is an honest question: what is the egregious misuse of terminology that you are arguing against?
And before you answer (if you will), I'm asking independently of the actual summary or TFA. I don't care much about that. Let say this started out with someone saying they had invented a machine that can predict earthquakes (with some important provisos). It's my belief that the rest of your comments still naturally follow. So lets say the claim is that someone can use a bunch of sensor readings integrated with a computer model to model the relevant forces with sufficient fidelity that the model allows actions to be taken (like evacuations or "get in the basement with some bottled water" warnings, or whatever). Is this fundamentally impossible for some reason, or does it just not match a definition of prediction that I'm missing?
In my experience, they can block it. They will tell you they can't, but they can. You might have to call back a few times. You might have to threaten to cancel, or get a supervisor, but they can block it.
If you don't have insurance, it costs what, $300-$400 for an ambulance? That was 10 years ago, maybe it's more now. This amount is a deterrent for calling an ambulance to pick you up when you could as easily get driven to the hospital (non life-threatening but uncomfortable injuries). I hardly think it's a damned if you do, damned if you don't situation.
I imagine the hospital bill for the original poster was much higher.
Maybe I'm just naive, but to me the story fed by Apple has been fairly consistent. I don't understand all the accusations of lies and the rest of it. I also don't understand all of the posts here about "either its a problem with the design and they should fire someone or its not a problem with the design and they should fire no one."
This all seems logical to me:
All smart phones have signal issues when you hold them a certain way
iPhone 4 is worse than most when you bridge the gap between the antennas
iPhone 4 *appeared ever worse* than it actually is because of the algorithm calculating bars
Even though its worse than most, the problem can be addressed by a free bumper. And hey, even without the bumper, the problem isn't actually causing most people to drop more calls.
Whatever the true loss in phone performance because of the problem, there is a real problem, and someone should be held accountable for it. So (maybe), Papermaster bit the bullet for that.
For me, this covers the whole issue and all of the information that has come out. Seems pretty straightforward and not all that sinister, but again, maybe I'm naive.
Imagine your Boss came to you and said "We're having 10 programmers make the same program, but we'll only pay one of you". That means that 9 of them end up working for free. That's why they hate that business model, no serious graphics designer can make a living out of such contests.
And the decision of who to pay will be made by someone with no expertise in evaluating user interface design, usability, scalability, security, or correctness.
As much as I don't want to bestow special powers to designers, I have met good designers that seem to have a gift. Through training, experience, and natural talent, they can design visuals that direct the eye to the right place, evoke the right emotions, and have lasting impact. Most people cannot consciously distinguish between visuals that have those properties and those that do not, so when the client simply chooses from a large pool of designs there is no telling if the graphic is actually any good.
It's not free, unfortunately. I briefly looked into using it but got distracted by something shiny (maybe trying to finish my thesis...)
CULA is a GPU-accelerated linear algebra library that utilizes the NVIDIA CUDA parallel computing architecture to dramatically improve the computation speed of sophisticated mathematics.
http://www.culatools.com/
Except the CEO of Viacom isn't the one who is bothered by outrageous prices and terms on content. He can afford to pay whatever the premium rate is for the medium de jour, and he can upgrade his personal digital library when HHDDVVDD BVD comes out. Similarly, he doesn't need personal backups of his content, because he can afford to repurchase it.
I don't think using their own dirty tricks against them is a way to make much progress. They can afford to play by their own rules.
I think you're probably about 10-20 years out of date with your criticism. AI these days is *all about* statistical machine learning which is *all about* data and not about formal or expert systems at all. This is what Google and others are doing. The AI you are describing is from the late 80s and early 90s.
Neural networks are part of the story, but many of the ideas from ANNs have been improved upon when more structured settings are available. There is actually a resurgence right now in deep neural network though.
I don't really see why. I'm a Canadian, so things are obviously different, but it seems likely that this works similarly. Currently, here, everybody can file their taxes early, but they *must* be in by the end of April or fees and penalties start accruing.
Many people leave it to the last minute and there is a mad rush in April to get taxes taken care of.
After you file your taxes, you get a notice of assessment. Why not put a "due date" for the following year on the assessment. If before April is too early, then spread the due dates between the end of April and the end of some later month (August, for example). This could go in a round-robin rotation, so your due date gets later by one month every year before restarting back to April. You could still submit your taxes earlier than your due date. The due date just determines when penalties and fees start accruing for unfiled or unremitted taxes. All of the "fiscal year" stuff stays the same, no additional accounting headaches.
Except that, as has been posted here before, people are terrible at self-assessing their skill. I know, I know, you are different: you are not overestimating yourself, you are one of the 0.025% of people who can talk on the phone without being distracted http://www.eurekalert.org/pub_releases/2010-03/uou-fdw032610.php.
I know for a fact that I cannot multitask. However, I believe myself to be particularly good at self-evaluation. I know about psychology, and I read slashdot: I can adapt my self-assessment. I'm a scientist and I don't have a large ego about my regular cognitive skills, I am the typical absent-minded professor type. However, I didn't really realize how poor I was at multitasking until my late 20s, and I am particularly bad at it. I had a couple of near accidents (nothing that would have been severe), but I understand probability and statistics. I know that if I continued to drive distracted, with overwhelming probability I would eventually cause an accident. So I stopped sampling.
This does not describe most people. Many are overconfident and unable to recognize their own deficiencies. Even more don't understand that taking a small risk enough times basically ensures that the low-probability outcome will eventually happen.
I don't want those people deciding what's safe, because you know what, they won't realize they have a problem until they get in an accident. And the first time, they will attribute it to bad luck. My mother in law rear-ended someone while changing the radio station and shrugged it off: bad luck, could happen to anyone.
There are too many people on the road for them to be learning what's safe and what's not by trial and error. No thanks.
The book has helped myself and several others in our Ph.D program quit smoking. I think you may find that you don't enjoy smoking, but rather you enjoy relieving the physical and psychological symptoms of Nicotine withdrawal. Each cigarette returns you to neutral, and after about an hour your Nicotine levels have dropped and your addicted body makes you uncomfortable so you enjoy having another cigarette and returning back to neutral. The truth is you like having an absence of withdrawal symptoms, ie, of being a nonsmoker.
Probably. So far everyone I know that has read the book has easily quit and has come to understand this perspective. We're not exactly a low IQ bunch. But I could be wrong of course. Couldn't hurt to find out though: at least you'd have a reason to ditch the stink and health problems...
Bayesian statistics are great because they let you specify your prior information about the quantities of interest. They don't give you prior information for free when you have none:)
Wow. What class did you take that says if you don't know something you should assume equal probability?
I don't know if there is an invisible elephant in my kitchen, so I guess I should assign equal probability to both outcomes. I also don't really know how Baccarat works, I guess my odds are 50/50.
Without knowing something about he or his coworkers, you by definition cannot make any statistical statements. To make any statements, you would first need to make some observations. This is how statistics is different from logic. Statistics is grounded in data.
I don't agree with Zed, but you may have just proved his point.
I don't understand what is to fall for. I guess it depends on if you are doing multiple comparisons or a single test.
If you independently identified a suspect and could put together a case against them, and *then* got a DNA match, slam dunk. In that case you're right, the jury should not fall for that argument.
However, what about the situation where DNA is found at a crime scene, and then a database search yields a match? Perhaps that person has no alibi or way to explain how what is apparently their DNA got into this rape victim. Then the defense should surely ask "How many people would have to be in the database before a DNA sample from the crime scene will match somebody?" If those odds are not infinitesimal and the case is built around DNA evidence, there is a big problem.
Slashdot Mirror
It's not a matter of "working fine" vs. "not working". GP point (I think) is that it works just as well as it always has. The problem isn't that the immune system's performance has changed.
The point of this thread was that the students who did not cheat may have principled grounds to feel wronged and to take official action if they are compelled to rewrite the test.
What of it? My point was that there is a formal expectation that if a student does not cheat they have a basis to object to re-taking an entire examination.
Also, I don't know exactly what is meant by the honor code. We never got an "introduction to honor" when I started university. In fact, I learned the penalties of cheating when I was actually a sessional lecturer. I was blown away that a student could be expelled, which would also mean no other university would ever accept them, for a very broad definition of cheating.
I don't know this exact situation, but it is very common to pass down previous notes, exams, and assignments to friends/relatives if they take the same course you took. The acceptance by the professor of this could range from encouraged to expulsion-worthy. For example in linear algebra and calculus the library stocked the last 5 years worth of mid terms and final exams. (questions and answers). In other courses the faculty re-used *exact* tests and assignments year after year so you even knowing some of the questions from a previous year gave a huge unfair advantage...
Sorry I digressed. Your comment didn't make sense to me and thinking about it dredged up all these unhappy memories about the arbitrariness of these punishments.
Usually a course has what my school called a ROASS document: responsibilities of academic students and staff. This document outlines how many assignments there will be, roughly when they will be assigned and due, number of examinations, the relative weighting of each of these, penalties for cheating, etc.
This document helps the students plan their term because often they are taking 4-5 heavy workload courses. If all of your courses are backloaded with big projects or exams, you may want to replan your semester. The document also protects students from lazy profs who fall behind and would then dump 3 assignments on the students by surprise at crunch time at the end of term, or from inventing course projects at the last minute, etc. Also from shifting weight to the final exam with short notice because their students did too well on assignments, or because they bombed the assignments, etc.
If a student lives up to his/her responsibilities as outlined in the document, but the professor does not, the student has grounds to file a complaint. Extreme cases are needed for anything to come of it, but it definitely happens. More often you would talk to the dept head and he might have a chat with a rogue professor who is abusing their students.
I don't know about fashionable, but perhaps necessary if you want to do science. It seems natural to model the behavior of most things as a function of a) initial conditions b) input c) randomness/stochasticity. For fixed initial conditions and input, you can model the distribution of outcomes. Then you can model how that distribution changes as you change initial conditions and inputs. Eventually you can look at behaviour and hypothesize about its causes, perhaps making changes to the initial conditions or inputs in the future to try to achieve better outcomes. OR, sometimes even better, realize that the causes of the outcome are not distinguishable from the randomness and therefore cannot reliably be changed. Maybe there is a small probability that people just turn out to be serial killers. Tough, deal with it.
That all seems pretty scientific to me: not computer scientific, just scientific.
I don't see an obvious alternative approach, for people that want to model/understand the behaviour of the organism. Not a scientific one, anyway. I mean we could just say it's too complicated and we can never understand/model it, which may be true. But that would just mean we don't have a tight enough model of either a) the initial conditions b) the inputs or c) the randomness. And perhaps we never can create models that are good enough to be useful. That's a question worth answering, and one that science can address.
Ok. That's not a definition of prediction that I think anyone in statistics or machine learning would find particularly interesting.
I would say that a prediction is simply a statement about a future event. Systems that make predictions (predictors) can have varying degrees of accuracy. Few predictors are 100% accurate, considering the large number of low probability (and usually unmodeled) events. Even a perfect weather predictor may be in error if the power goes out or an asteroid hits the earth.
Any outcome with even the smallest uncertainty is unpredictable, but your definition (if I understand). That doesn't seem useful. I'd argue that if something is predictable *enough* that the predictions can reliably be used to improve utility, then the predictor has value.
Note: I misread the thread structure and thought the parent of my post was a different post, so feel free to ignore me.
Not predicted, statistically modeled with a great degree of accuracy (that is a hair worth splitting).
As someone who is comfortable with statistics and machine learning, I have to wonder what you're driving at here? You started by making a sort of "Guffaw!" statement about predicting the unpredictable. And now you're trying to be pretty precise about separating prediction from accurate modelling. What bone are you picking? This is an honest question: what is the egregious misuse of terminology that you are arguing against?
And before you answer (if you will), I'm asking independently of the actual summary or TFA. I don't care much about that. Let say this started out with someone saying they had invented a machine that can predict earthquakes (with some important provisos). It's my belief that the rest of your comments still naturally follow. So lets say the claim is that someone can use a bunch of sensor readings integrated with a computer model to model the relevant forces with sufficient fidelity that the model allows actions to be taken (like evacuations or "get in the basement with some bottled water" warnings, or whatever). Is this fundamentally impossible for some reason, or does it just not match a definition of prediction that I'm missing?
In my experience, they can block it. They will tell you they can't, but they can. You might have to call back a few times. You might have to threaten to cancel, or get a supervisor, but they can block it.
and a few stranglers might show up only to find out there's no exam.
How very disappointing for the stranglers. I'm assuming they were hired to deal with the cheaters?
If you don't have insurance, it costs what, $300-$400 for an ambulance? That was 10 years ago, maybe it's more now. This amount is a deterrent for calling an ambulance to pick you up when you could as easily get driven to the hospital (non life-threatening but uncomfortable injuries). I hardly think it's a damned if you do, damned if you don't situation.
I imagine the hospital bill for the original poster was much higher.
This all seems logical to me:
For me, this covers the whole issue and all of the information that has come out. Seems pretty straightforward and not all that sinister, but again, maybe I'm naive.
Imagine your Boss came to you and said "We're having 10 programmers make the same program, but we'll only pay one of you". That means that 9 of them end up working for free. That's why they hate that business model, no serious graphics designer can make a living out of such contests.
And the decision of who to pay will be made by someone with no expertise in evaluating user interface design, usability, scalability, security, or correctness.
As much as I don't want to bestow special powers to designers, I have met good designers that seem to have a gift. Through training, experience, and natural talent, they can design visuals that direct the eye to the right place, evoke the right emotions, and have lasting impact. Most people cannot consciously distinguish between visuals that have those properties and those that do not, so when the client simply chooses from a large pool of designs there is no telling if the graphic is actually any good.
It's not free, unfortunately. I briefly looked into using it but got distracted by something shiny (maybe trying to finish my thesis...)
CULA is a GPU-accelerated linear algebra library that utilizes the NVIDIA CUDA parallel computing architecture to dramatically improve the computation speed of sophisticated mathematics.
http://www.culatools.com/
Cut support for older products, or stopped providing free upgrades that give new functionality on a regular basis?
I had a Razr before my iphone. I remember getting 0 upgrades ever from the original crap interface.
Except the CEO of Viacom isn't the one who is bothered by outrageous prices and terms on content. He can afford to pay whatever the premium rate is for the medium de jour, and he can upgrade his personal digital library when HHDDVVDD BVD comes out. Similarly, he doesn't need personal backups of his content, because he can afford to repurchase it.
I don't think using their own dirty tricks against them is a way to make much progress. They can afford to play by their own rules.
I think you're probably about 10-20 years out of date with your criticism. AI these days is *all about* statistical machine learning which is *all about* data and not about formal or expert systems at all. This is what Google and others are doing. The AI you are describing is from the late 80s and early 90s.
Neural networks are part of the story, but many of the ideas from ANNs have been improved upon when more structured settings are available. There is actually a resurgence right now in deep neural network though.
I don't really see why. I'm a Canadian, so things are obviously different, but it seems likely that this works similarly. Currently, here, everybody can file their taxes early, but they *must* be in by the end of April or fees and penalties start accruing.
Many people leave it to the last minute and there is a mad rush in April to get taxes taken care of.
After you file your taxes, you get a notice of assessment. Why not put a "due date" for the following year on the assessment. If before April is too early, then spread the due dates between the end of April and the end of some later month (August, for example). This could go in a round-robin rotation, so your due date gets later by one month every year before restarting back to April. You could still submit your taxes earlier than your due date. The due date just determines when penalties and fees start accruing for unfiled or unremitted taxes. All of the "fiscal year" stuff stays the same, no additional accounting headaches.
Good call. Point taken.
Except that, as has been posted here before, people are terrible at self-assessing their skill. I know, I know, you are different: you are not overestimating yourself, you are one of the 0.025% of people who can talk on the phone without being distracted http://www.eurekalert.org/pub_releases/2010-03/uou-fdw032610.php.
I know for a fact that I cannot multitask. However, I believe myself to be particularly good at self-evaluation. I know about psychology, and I read slashdot: I can adapt my self-assessment. I'm a scientist and I don't have a large ego about my regular cognitive skills, I am the typical absent-minded professor type. However, I didn't really realize how poor I was at multitasking until my late 20s, and I am particularly bad at it. I had a couple of near accidents (nothing that would have been severe), but I understand probability and statistics. I know that if I continued to drive distracted, with overwhelming probability I would eventually cause an accident. So I stopped sampling.
This does not describe most people. Many are overconfident and unable to recognize their own deficiencies. Even more don't understand that taking a small risk enough times basically ensures that the low-probability outcome will eventually happen.
I don't want those people deciding what's safe, because you know what, they won't realize they have a problem until they get in an accident. And the first time, they will attribute it to bad luck. My mother in law rear-ended someone while changing the radio station and shrugged it off: bad luck, could happen to anyone.
There are too many people on the road for them to be learning what's safe and what's not by trial and error. No thanks.
I used to think that too. I suggest you read Allen Carr's book. It's an easy read:
http://en.wikipedia.org/wiki/Allen_Carr
The book has helped myself and several others in our Ph.D program quit smoking. I think you may find that you don't enjoy smoking, but rather you enjoy relieving the physical and psychological symptoms of Nicotine withdrawal. Each cigarette returns you to neutral, and after about an hour your Nicotine levels have dropped and your addicted body makes you uncomfortable so you enjoy having another cigarette and returning back to neutral. The truth is you like having an absence of withdrawal symptoms, ie, of being a nonsmoker.
Probably. So far everyone I know that has read the book has easily quit and has come to understand this perspective. We're not exactly a low IQ bunch. But I could be wrong of course. Couldn't hurt to find out though: at least you'd have a reason to ditch the stink and health problems...
In Canada it just works. Turn it on, choose Bluetooth or USB, and if Bluetooth, do the "pairing" thing. Done.
Bayesian statistics are great because they let you specify your prior information about the quantities of interest. They don't give you prior information for free when you have none :)
Wow. What class did you take that says if you don't know something you should assume equal probability?
I don't know if there is an invisible elephant in my kitchen, so I guess I should assign equal probability to both outcomes. I also don't really know how Baccarat works, I guess my odds are 50/50.
Without knowing something about he or his coworkers, you by definition cannot make any statistical statements. To make any statements, you would first need to make some observations. This is how statistics is different from logic. Statistics is grounded in data.
I don't agree with Zed, but you may have just proved his point.
I don't understand what is to fall for. I guess it depends on if you are doing multiple comparisons or a single test.
If you independently identified a suspect and could put together a case against them, and *then* got a DNA match, slam dunk. In that case you're right, the jury should not fall for that argument.
However, what about the situation where DNA is found at a crime scene, and then a database search yields a match? Perhaps that person has no alibi or way to explain how what is apparently their DNA got into this rape victim. Then the defense should surely ask "How many people would have to be in the database before a DNA sample from the crime scene will match somebody?" If those odds are not infinitesimal and the case is built around DNA evidence, there is a big problem.