Since people have configured computer systems to learn, we wish to know exactly what they've learned and how they deciding after their learning process is complete. The solution to such questions could shed light on our own decision-making processes. Perbedaan Ayam Bangkok dan Ayam Kampung
Kate Saenko, an partner teacher of computer system scientific research at Boston College, asked people to appearance at lots of photos portraying actions that the computer system may have handled its roadway to a choice, and determine its probably course.
The people gave answers that made good sense, but there was a problem: they made good sense to people, and people, Saenko understood, have biases. In truth, people do not also understand how they themselves deciding. How on the planet after that could they determine how a neural network, with countless neurons and billions of links, makes choices?
Saenko did a 2nd experiment, using computer systems rather than individuals to assist determine exactly what learning devices learned.
"What we learned that is really important is that, despite the severe intricacy of these formulas, it is feasible to peek under the hood and understand their decision-making process, which we can actually ask people to discuss it to us," says Saenko. "So we think it is feasible to instruct people how devices make forecasts."
FROM RANDOM TO REVEALING
Computer system researchers know generally terms how neural networks develop. Besides, they write the educating programs that direct a computer's supposed neurons to connect to various other neurons, which are actually mathematical functions.
Each neuron parses one item of information, and every neuron improves the information in the coming before nodes. In time, links develop. They go from arbitrary to exposing, and the network "learns" to do points such as determine opponent terminals in satellite pictures or spot proof of cancer cells lengthy before it's noticeable to a human radiologist. They determine faces. They own cars.That is fortunately. The disconcerting information, says Saenko, is that as expert system plays an progressively important role in the lives of people, its learning processes are ending up being progressively odd. Simply when we really need to trust them, they have become inscrutable. That is a problem.
