random thoughts

Random ideas:

Is it feasible to scan entire books without flipping any pages?
- Approach: Powerful beam of focused light that passes through the pages. Focal length can be changed to provide more data points. Page boundaries can be discerned by the difference in absorbance of paper, laminate and ink.
- Difficulties: Thick hard-cover books would be quite opaque. Difficult for sensor to discern print on pages that face each other. Noise due to paper grain.
- Easier workarounds: Mechanical page flipper. In fact, it probably already exists.

VCG mechanism (which is truthful in DRM under non-negative valuation, non-positive external contribution and quasi-linear utility domain) is not collusion proof. It's truthful because telling the truth is the Nash equilibrium. However, Nash equilibrium does not imply sub-game perfect. VCG is also not collusion-proof. But interestingly, it seems that Nash equilibrium, if we consider for simplicity that the action set is continuous and differentiable, only considers the first and second derivatives of the multi-dimensional mapping between utilities and utilities (for each outcome rule assuming a dominant strategy exists for all players and they play that). Wait, it's not a derivative of a function - it's a derivative of many functions. Darn. I'm wondering if there is an many-to-many mapping equivalent of a Hessian determinant. If there is, it would probably be completely useless, but still kinda cool. The real solution might not even give any insight since non-integral indices doesn't mean anything at all in this scenario.

My first-world problem: Watching a youtube video on developing countries when I have a slow internet connection, and feeling guilty and annoyed at the same time when it stops to buffer.

Also, reading the Communist Manifesto while listening to "Do you hear the people sing" is pretty awesome.

I remember vaguely that I would once have used words other than "a mapping from the set of combination of clothes to days" in a dinner conversation. Nowadays it feels like a completely normal and natural way of saying something that everyone can understand. Thanks Charter club. (I think the words would have been a choice of outfit, but I'm not sure if my mouth has ever uttered these words before either.)

Quora is amazing. But I still think the problem of people repeating arguments that have a set of good counter-arguments still hasn't been adequately solved. I want to learn more about what it takes to build a large information system, and it's looking rather intimidating. There is the computational component, the incentive component, and the business component. I mean, holy crap it's a hard problem. But I still think it's absolutely worth solving. Branch is doing nicely on part of allowing people to debate on issues that branch out, and stack exchange is doing well on the community edited answers front. In fact, stack overflow is fantastic on search - if the top thread on google isn't the question you're asking, one of the "related questions" probably is. It's amazing and worth emulating. Stack overflow is solving one of the problems on forums. But debates aren't handled very well, and the available solutions still disappoint me. What I would like to see is more scoring of responses to arguments. Quora allows embedded quotes. That is fantastic - I would like the whole internet to allow embedded quotes, so that attribution is easy.

Seriously though, can we have embedded quotes for the whole internet? What would it take? First problem - webpages change. Can't archive everything - too much space needed. A browser tool for a generating citation on the fly? That would be nice, but seems a bit intrusive. Yeh, attribution of a source in a user-friendly yet informative and structured way is hard. Reproducing the source wholesale could probably do it. Sounds like a job that "turn it in" has already figured out though, so the infrastructure definitely already exists.

"Ideas" are difficult to handle - it's so context dependent. A probabilistic approach is good at figuring out the topic, but is it good at figuring out the meaning?

I wonder what hardwired parameters our brains have. What is the "corpus" of a human being? It would be interesting if we can get hold of all the sensory inputs to a child from birth onwards. What does the baby know about the world when it's born, and how much of it was learnt from experiences in the womb? (Alas, hard to know)

Implementing an MCMC algorithm for Bayesian reasoning is pretty interesting. There was this case where all the nodes reinforced each other, and the algorithm took a really long time to converge - because the chance of the algorithm "changing its mind" about the central node's value. It makes me wonder if that is somehow related to religious beliefs - strongly reinforcing nodes. But that requires a certain model of reasoning.

Thoughts on a model of reasoning (I'm sure the cogn-psy peeps have figured out a large part of it, but I guess I can always ramble). People can hold on to inconsistent beliefs. People only reason some of the times, and believe what has been said most of the time.