Tuesday, September 22, 2009

Facebook prize

Netflix just awarded a cool million to a 7-member team who managed to devise algorithms that improved the prediction of user's ratings of movies by 10% compared to Netflix's own Cinematch algorithm, the endgame of a 4-year long contest (see NYTimes).

Isn't it time that Facebook sets up a contest of its own? With its database of human connections, Facebook can challenge the machine learning community to predict friendship: reveal 90% of each user's friends, and have the algorithm predict the rest 10%. Another challenge is to reveal each user's friends up to a certain date, and have the algorithm predict new friendships that the user forms afterwards. May the most accurate algorithm prevail.

Online dating services can certainly use such an algorithm. Who else? Insurance companies? Banks? Salespeople? How much can you learn about a person from the friends he or she keeps?

Sunday, September 20, 2009

Mobile human computing

I am a big fan of Luis von Ahn's work on human computing. The idea of human computing is to harness the idle human brain power to perform computational tasks that so far have eluded the computers, such as image labeling (see google's image labeler) and recognition of old text (check out reCaptcha). The clever part of von Ahn's approach is to engage the human brain through an entertaining game or as a by-product of other necessary tasks.

I believe that this approach can be extended to the mobile phone platform. With smart phones like the iPhone, the location and time of the carrier are known, and these information should be valuable for designing location- and time-specific computational tasks. Another advantage of using smart phones as a platform for human computing is that these phones are almost always accessible at idle times - my colleagues invariably play on their iPhones when the meetings get boring. 

What can be accomplished using smart phones and idle human time? I have not come up with a specific application. But here are some general, and obvious, thoughts. Each human should be presented a simple task, which should not require expert knowledge or special skill, and should be fun to perform. The task should be easy for normal humans but difficult for even the most advanced computers, such as to observe and describe one's surroundings, or to identify a person. The outcome of all these tasks should then be combined in a way to solve a complicated problem. In order to take full advantage of the mobility of smart phones, the problem that we try to solve should ideally involve the locations and times at which the carriers perform their tasks.


Tuesday, September 08, 2009

First time pheasantry

The first bird got away. It flew away so fast that before either Yang or I opened the safety on our guns, it had crossed the grassfield and safely alighted in the woods far beyond the reach of any 12-gauge shotgun. We followed its flight path with our eyes. It was a beautiful male bird with a long tail and a dash of red on its head. We were relieved that we did not have to kill so early in the hunt. But the two dogs were disappointed.

The pointing dog was responsible for the escape of the first bird. Upon smelling the scent of the bird, the pointer should have frozen and have pointed its nose at the bird's hiding place. Instead, it came too close to the bird and flushed it out before we could ready our guns. More experienced hunters could have responded in time. But we could not raise our guns that fast yet.

Don, our guide, with a Jack Nicholson semblance and a loose lip for curse words, told us to be more ready the next time. We walked on, in wet, knee-high grass which I feared might harbor venomous snakes or ankle-breaking pits. The pointer ran around excitedly, the bell on its neck ringing. When it got too far away from us, Don would blow a whistle to call it back.

Then the pointer froze. Its bell stopped ringing, and a moment of silence fell upon the field. Don stepped forward in the direction of the pointer's nose. As he kicked in the bushes, a bird took flight. "Shoot it!" Don shouted. This time I was ready. I mounted my gun, waited for the bird to clear a distance for 20 yards - shoot it any closer and there will be very little meat left, and took aim and pulled the trigger. Feathers scattered in the air, and the bird fell straight down into the grass.

Don released the retriever, who dashed to the spot where the bird fell. In half a minute it returned with the lifeless bird in its mouth. Don wrestled the bird away and put it in the pocket of his enormous vest. A good shot, Don congratulated me.

We would return home with at least one bird, I thought. We'd driven 3 hours from New York City to this upland preserve in the Catskill, slept the night before in a fly-infested cabin, and paid $360 for the hunt. Our jeans had got all wet from the knee down, and our boots were covered in mud. It would be disappointing to return home empty-handed, and embarrassing to face the friends that we had promised to share our game. Although Yang and I had been shooting clay targets for a couple of years, we had never hunted before. The clay target only flies out when we are ready and call for it; the pheasants are less obedient to our wills.

It turned out, however, that hunting pheasants is not too different from shooting clays, as far as the shooting technique goes. Pheasants are big birds, and they fly rather slowly and in almost straight lines. Although they didn't come out at our bidding, the well-trained pointer gave us plenty of time to prepare for the shots, and Don invariably flushed them out when we were ready. The hunt lasted just short of two hours. The pointer found all ten birds that Don released in the field. I shot two of them, Yang got another two. We also shot one bird together - we fired simultaneously and heard only one shot! The other five birds got away; may they live prosperously until old age.

The irony of hunting with a guide in a preserve is that we never really got to touch the birds. Each time we dropped a bird, the retriever got it back and Don pocketed it for us. After the hunt, the management staff plucked and skinned the birds, and packaged them in vacuum-sealed plastic bags. In the end, we were handed nicely packaged pheasant breasts at check-out. Aside from a few residual down left on the wings, they were indistinguishable from a supermarket purchase.

A few days later I baked honey-glazed pheasant breasts. Our guests and we all agreed that they tasted just like chicken.