Hacker News Reader

Comment by nemesisj | Hacker News Reader

This is also a great way to solve a problem. If everyone in the room is stumped, throw out a stupid solution. If nobody can improve on it, then the last solution wins. Works surprisingly well as most people can critique while finding it hard to create from scratch.

View on HN

When you think you're done, gradually decrease the temperature of the room. People will become uncomfortable and leave, even if it means accepting a sub-optimal solution.

simulated annealing. ;)

I am really wondering whether that was deliberate or not...

It was. Glimcat's comment got me thinking in that direction, and I couldn't resist.

I often take a more transparent style to the same approach: "We can't think of any good solutions, so here's a possible, but obviously flawed solution to talk about. Maybe it'll inspire something better."

People generally don't like to propose imperfect solutions because of the fear of looking dumb and wasting time. But, when everyone is stumped, sitting in silence is wasting time. Re-igniting the conversation by saying "I don't want to actually do this, but here's another angle" can kick the problem-space search out of a local minima.

I call this "proposing a strawman solution". I don't know why it works, but it frequently does. I think that doing a long search for The Perfect Solution is akin to premature optimization. Changing focus for a while seems to help.

This is actually similar to a common approach in probabilistic modeling.

Pick an initial model & set of probabilistic priors. Evaluate it with a "goodness of fit" heuristic function, then iterate & keep measuring while keeping the best solution discovered.

As long as the initial parameters are sort of reasonable, it will give you pretty good results for many problems.

It obviously doesn't get rid of the need for a better understanding of the problem. Improvements to how well your features describe important attributes of the problem tend to be strictly superior to your choice of learning algorithm (i.e. your exact process of iteration).

But as long as the initial solution is more or less on target? You can solve many problems by picking a solution that you know is inadequate, then iterating.

I'm fairly knowledgeable about Bayesian statistics, but I can't tell which technique you're referring to. It vaguely sounds like maximum likelihood, but ML doesn't iterate the model or priors. Could you explain further and give a reference?

Iterative algorithms which attempt to solve maximum likelihood usually fall under the category of expectation maximization (a good example is K-means clustering).

http://en.wikipedia.org/wiki/Expectation-maximization_algori...

Ah, I'd always thought of EM as iterating over the parameters given a model, but I suppose it makes sense to see it as a search for a model, including parameters, that fits the data. Cool.

Perhaps he/she is referring to conjugate priors [1], where the prior and the posterior are of the same family of distributions. Once you compute the posterior using a likelihood and prior, you obtain a posterior that is of the same family of distribution as the prior. This posterior can then be used as a prior for more samples.

[1] http://en.wikipedia.org/wiki/Conjugate_prior

I think this is definitely the best solution if your crowd is French, as in France people are not encouraged to submit new ideas but their criticizing skills are state-of-the-art :-)

Gone there, did there, and it's dangerous. Somebody might like your first solution, and be vocal enough to make everybody on the room shut-up.

Never attempt to intentionally invoke Cunningham's Law with a solution you wouldn't be happy with. Which is a more specific form of "never propose something you wouldn't be happy with", or "be careful what you wish for, because you just might get it".

Corollary: be careful what other people wish for.

That seems more like a hiring problem.

The problem was they hired the first candidate, figuring they could get a replacement when a better one came along , but the first one got entrenched.

It started as a hiring problem. Now there's still a hiring problem, but there's also this one, which is unrelated. Both should be fixed and, unfortunately, neither has an easy or comfortable solution.

This is supposedly how Apple got its name [1, 2].

[1] http://wiki.answers.com/Q/How_did_Apple_computers_get_their_...

[2] http://www.tydknow.com/did-you-know-that-apple-was-named-so-...

That's called a Straw Man. Start with a naive solution and refine rather than trying to get it right all at once.

a straw man is telling person A that argument B is wrong, therefore person A is wrong, when in fact person A was making argument C.

it is a trick to confuse your opponent and any spectators. it is most often employed by people too stupid to discern between argument B and argument C in the first place, which is to say it is most often employed inadvertently.

amusingly, most times when people accuse someone else of making a straw man argument, they then proceed to immediately make a straw man argument themselves in explaining the claim.

edit: mistyped.

A straw man proposal is different than a straw man argument.

Well they are both designed to be knocked down/beaten up. The straw man proposal is your offering to the forum to beat and the argument is your twisting/manipulating an opponents position or argument to enable you to knock down the straw man and claim a win in the argument.

also, the point of a straw man proposal isn't exactly to be refined... it's to get jump off ideas from the straw man's trivial refutation.

If you have a sufficiently rapid turnaround time, this works for products, too.

For example, don't have a good idea for a landing page? Throw up a bad one. It'll get you data and you can improve from there.

Obviously, a less good idea for things where iteration time is measured in months...

Reply on news.ycombinator.com