4 ms·

> Treating it as a black box has short lifetime, as your method of analysis will miss key features of the system or oversimplify it. On the other hand, it can

> Treating it as a black box has short lifetime, as your method of analysis will miss key features of the system or oversimplify it. On the other hand, it can work indefinitely if it solves a problem well enough and it is used always on the same kind of problem. Sure, you will need to understand it better to apply it to new markets, but often that kind of research is outside the scope of a single business, and they don't need it.

AstralStorm 7y ago
How do you know it solves the problem, instead of just tricking your measure? Do you know when it doesn't work? Business does not care about rigor, but then terrible things happen when your face recognition system happens to not work for people with dark skin tone. Even more terrible things happen when it detects cats as people, and worse once it's used to link data into a police database. Even pretty bad things happen when it's used for detecting potential (treated as absolute) copyright violations. Security holes when it's used for detecting security problems. Loss of business when it's used for ticket prioritization. Business world does not care that it sells a broken solution as long as it's not obvious and someone has been paid. Everyone else pays for the failures. And you cannot sue an ML system really.
- dorgo 7y ago
  Why don't we do online learning? Each failure is a new training sample. Why dont't we do active learning? Or adversarial networks, or.... Sure it is easy to fail in machine learning. But it doesn't mean that we need to undstand how and what our models learn. That's the only big advantage of machine learning: the model learns, so I don't have to.
bumby 7y ago
I don't necessarily disagree (in the sense that all models are wrong but some are useful) but to play devils advocate, its damn near impossible to claim a model will work indefinitely on a subset of problems if we don't know it's underpinnings. Its like saying software has been tested completely when we really should be saying its passed a subset of test cases out of the complete set of possible scenarios. The only way we can make that claim is if we can interpret the model in a mathematical proof.

AstralStorm 7y ago

How do you know it solves the problem, instead of just tricking your measure? Do you know when it doesn't work? Business does not care about rigor, but then terrible things happen when your face recognition system happens to not work for people with dark skin tone. Even more terrible things happen when it detects cats as people, and worse once it's used to link data into a police database. Even pretty bad things happen when it's used for detecting potential (treated as absolute) copyright violations. Security holes when it's used for detecting security problems. Loss of business when it's used for ticket prioritization. Business world does not care that it sells a broken solution as long as it's not obvious and someone has been paid. Everyone else pays for the failures. And you cannot sue an ML system really.

dorgo 7y ago
Why don't we do online learning? Each failure is a new training sample. Why dont't we do active learning? Or adversarial networks, or.... Sure it is easy to fail in machine learning. But it doesn't mean that we need to undstand how and what our models learn. That's the only big advantage of machine learning: the model learns, so I don't have to.

bumby 7y ago

I don't necessarily disagree (in the sense that all models are wrong but some are useful) but to play devils advocate, its damn near impossible to claim a model will work indefinitely on a subset of problems if we don't know it's underpinnings. Its like saying software has been tested completely when we really should be saying its passed a subset of test cases out of the complete set of possible scenarios. The only way we can make that claim is if we can interpret the model in a mathematical proof.