After reading "Statistical Significance Tests for Machine Translation Evaluation" I have some que

reply to post by BayesLike


Exactly. I tend to view human response/behavior at the societal level as riding on that proverbial bell curve where the most interesting activities are not going to be found in the middle but on the extreme ends of the bell--the more "radical". Looking at the radical ends and the overall conditions actually allows some predictability in societal response as they shed light on whether or not the status quo continues and what the majority response will be. In the case of Occupy, they didn't have a chance here in the US as the conditions here were simply not severe enough. Contrast it with Spain, Greece or even more recently, Brazil, and the conditions are much more severe, allowing for the radical to influence the majority enough to invoke a demand for change by a greater portion of the populace.

Absolutely agree that they may bugger themselves with "noise" but from the sounds of it, they eliminate the "noise" portion of the chatter to specific subject matters through the keyword use. It's kind of like google or twitter picking out the words/articles and selecting the most commonly viewed/used that are trending. If the program is being used to look for individual instances of certain behavioral responses, then a narrow channel of words framing that specific response would be used. However, at a societal level, it is comprehending what is moving through the mass as a whole that may actually matter as it is a direct indicator of active interest and possible response. The messy middle, as you so aptly called it, is ultimately influenced by the actions of the radical. The middle tends to be lazy ("as long as someone else is doing something, I don't have to") until they perceive little choice but to act. I see the interactions between the radical and the middle as having a tendency to behave rather like this: media.techtarget.com...

Based on what I have gleaned from all the articles discussing the data collection and processing that is done, I do believe that they are filtering the noise to expose specific individuals but I also believe that they go through the noise to find what is trending through society and taking interest in what may be a threat to the status quo. If I were in their position, I'd be interested in that, too, but I think where I would vary is how I responded to such interests because I believe in social contract.

The Stanford course is free. I think that either Harvard, Yale or Princeton also has a similar course that is free but I'm not 100% certain.

Originally posted by teachtaire
hey, just being curious, should ones steam chat randomly change to a different language, would that be a cause of concern?

Hitting alt-shift toggles a language change in Steam so are you sure you're not accidentally hitting that while toggling in and out of game? Overall, Steam did make a change to their privacy policies a bit ago that included the possibility of filtering/reporting to law enforcement agencies. I do believe they modified their position somewhat on the subject due to user response but am not 100% sure what they modified in their privacy policy.

reply to post by WhiteAlice


The filtering with key words is a must. The messy middle can be reintroduced after the models are built -- in general, if the models are good, the messy middle will be predicted to be in, well, the middle. If those predictions tend to either extreme, the model needs more work.

As a person who builds extremely complex models on a regular basis, I can say for sure that any good model will probably be a team effort. You don't want to start with just any dataset and you do want to use the insight of experts in the field of application, eg political scientists specializing in xxxx region with awareness of customs, language and dissent patterns.

Thanks for info on the course cost. I had hoped it might be free like many online MIT and Harvard classes.

reply to post by BayesLike


The messy middle does tend to stay in the middle, which is why society tends to change very slowly but, in some circumstances, it can be influenced by the radical extremes to produce change. The change that is produced, however, is rarely extreme. I guess an example of the radical influencing the messy middle would be Nazi Germany, where atrocities did occur. The probable reason why the messy middle in that situation behaved the way it did was most likely due to the tendency towards xenophobia and the need for survival. Civil Rights Movement would be another scenario. Those radical extremes on the bell curve can be the impetus for change. Most of the time, the change produced would be a subtle shift though greater shifts can occur if the messy middle feels as if there is no other choice. At least that's what I see when I look back at history.

Agreed on the team needed to avoid bias and political scientists and sociologists, definitely.

Curvilinear Component Analysis: a Self-Organizing Neural Network for Nonlinear Mapping of Data Sets (1996)

Is that a useful thing to read? Or not?

How about "Semi-Supervised Learning on Riemannian
Manifolds"? by MIKHAIL BELKIN & PARTHA NIYOGI?

Or what about A Reliable Effective Terascale Linear Learning System?

"A comparison of artificial neural network and time series models for forecasting commodity prices" requires purchase to read.

Also "Forecasting Commodity Prices with Nonlinear
Models" ethesis.helsinki.fi...

Was this what you were talking about? BAYESIAN MODEL UNCERTAINTY IN SMOOTH TRANSITION AUTOREGRESSIONS:

faculty.chicagobooth.edu...

I'll take that as a "you're on your own if you want to figure out how to use this stuff for stock trading or opportunity cost risk mitigation"?

Update:

www.i-programmer.info...

Lol, I hope I don't have to point out the timing and how this pertains to the topic at hand.

reply to post by teachtaire


Yes, Bayesian modeling would be what would most likely be used to evaluating outcomes as it allows for multiple, co-dependent variables to be used to make a determination of what is the most likely outcome. Taking in those co-dependent variables gives a more holistic picture of all the factors going into something that could create a potential effect. For example, you're in a burger joint and there is a thin man in line before you. His hair is in dreadlocks but neatly pulled back. He's clean, wearing a t-shirt, khaki shorts, and good shoes. What are the odds the guy will order a vegan burger? How would that result change if we changed some of the variables ie. he has short hair and is on the chubby side?

Not a fan of that "visual recognition" machine. Then again, Google has been becoming increasingly creepy over the years, imho. Some of the stuff that comes out of Schmidt's mouth is just...

reply to post by WhiteAlice


I'll take your word on the math, as the only reason I started reading about this stuff was because I didn't have enough cash for a full class schedule and started reading everything I could find on military strategy and military AI from the 70's on-wards. So my understanding in terms of math is pretty poor. My understanding on strategic and tactical military applications/approaches is however, probably better than that of most people walking down the street.

The main "issue" with the image processing is, in fact, a non-issue that can be solved with pyramidal image processing and some other basic things. Or maybe I misunderstood something.

I doubt most readers will understand what I'm talking about anyway.

reply to post by teachtaire



Hmm, if you're digging up fun stuff from the 70's including AI, let me introduce you to the oh-so-lovely Eliza:
nlp-addiction.com...

Unfortunately, Eliza was created in 1966 so a little bit before your preferred time period. She's a bit slow on the uptake but she's just a lovely, lovely AI.

There's this one, too, which kind of touches on early AI concepts from 1963: www.dtic.mil...

reply to post by WhiteAlice


Totally unlike the hand made and bound journals and papers I was reading! Thank you.

The stuff I was reading was so sketchy, the pages didn't even use the same paper stock or font from section to section

This doesn't appear to be for an actual rover or offensive system though. Cool!

*note* after reading this, I recalled talking to a programmer who was elderly, in his 70's or 80's who mentioned working on programming an AI for education. I didn't believe him because he was referring to the intelligence industry. Perhaps he wasn't just a cooky old man after all...

reply to post by teachtaire


Mhmm. The Systems Development Corporation had an individual within it working within a branch of the DoE to assist in creating a program for a relatively small number of children. That would be the author of the above paper but the traces of that have disappeared. Sometimes truth is indeed stranger than fiction. Eliza II was available for me to "play with" when I was in school. She's just as special as Eliza I. I'm not sure when 2 was created though but my interactions with her would be the 70's to early 80's. nlp-addiction.com...

As far as the differences in font, could be explained by differences in typewriter if it is specifically section to section and especially if the documents had multiple contributing authors. Assignment of responsibility could have been section by section. Just a guess.

reply to post by WhiteAlice


Or someone was feeling puckish and put something they shouldn't have in a library