4 ms·

Looks pretty good. But I'm afraid someone complaining about the mathematical foundations of AI is like an aspiring surgeon complaining about how much anatomy s/

Looks pretty good. But I'm afraid someone complaining about the mathematical foundations of AI is like an aspiring surgeon complaining about how much anatomy s/he has to study.

danboarder 13y ago
Agreed. The last time I worked on some AI problems was years ago back in college using LISP. I supported this project and look forward to 'catching up'. By the way, the author (Jeff Heaton) is the lead developer of Encog, available on github: https://github.com/encog https://github.com/encog
netcan 13y ago
If surgery didn't involve risk to patients, you could probably teach quite a lot of surgery to almost anyone. They might not be as able to improvise in rare scenarios, but the could probably remove appendixes pretty successfully. Some knowledge is inherently linear. Hard to teach multiplication to someone that can't add. Some aren't and a lot are grey areas where prior knowledge isn't absolutely necessary, but it's a big advantage. For example you can teach a lot of statistics without teaching the way things are calculated, but doing the sums with pen and paper will give you a better understanding. A confidence interval has a meaning that can be explained using words to someone without the mathematical skills to be able to calculate it. So is a statistical distribution. Is it better to know who to calculate the probability distribution with a pen and paper? Yes and in some interesting cases, it's very important. That doesn't mean that a lot of useful statistics can't be taught to people who will take shortcuts using software. Knowledge keeps growing like a tree with thickening branches and sprouting branches of their own. We can't tell new people where the last generation started. In the world of increasingly specialized knowledge and the culture of lifelong learning that I hope is evolving, its important to find ways of dropping people into the middle of fields. AI for not mathematicians seems doable.
gtani 13y ago
(Besides deciding whether to title the book AI, machine learning, data mining/science, applied/computational stats) There's a number of books that take the "short equation/inequality" only approach to presenting the author's subset of his/her appelation. Marsland's is pretty good. But I'm not sure they impart much intuition before you dive into the traditional texts (usually one chooses from the Big 6 of Murphy, Bishop, Barber, Mackay, Hastie/Tibshirani/ or Koller/Friedman). Note 3 of those have content freely available to read online, as well as free stats texts from otexts.com, UMass' Lavine and CMU's Shalizi and Kadane (2 separate books), and UCSD's Levy (specialized on stats for linguistics/CL). Note also the ESL guys (+ 1 additional author) have released Intro to Statistical Learning http://www.springer.com/statistics/statistical+theory+and+methods/book/978-1-4614-7137-0 http://www.springer.com/statistics/statistical+theory+and+me...
vidarh 13y ago
If you're going to do AI research I agree with you that you'll probably need quite a bit of maths, eventually. But people can certainly learn a lot of interesting things without it, and for some that might just be the motivation to pick up some more of the maths useful for the more advanced aspects as well. My maths sucks, and I've had no problems reading quite a few AI papers through the years and pick up a lot of interesting ideas, and implemented quite a few things for fun.

Looks pretty good. But I'm afraid someone complaining about the mathematical foundations of AI is like an aspiring surgeon complaining about how much anatomy s/he has to study.

danboarder 13y ago

Agreed. The last time I worked on some AI problems was years ago back in college using LISP. I supported this project and look forward to 'catching up'. By the way, the author (Jeff Heaton) is the lead developer of Encog, available on github: https://github.com/encog https://github.com/encog

netcan 13y ago

If surgery didn't involve risk to patients, you could probably teach quite a lot of surgery to almost anyone. They might not be as able to improvise in rare scenarios, but the could probably remove appendixes pretty successfully. Some knowledge is inherently linear. Hard to teach multiplication to someone that can't add. Some aren't and a lot are grey areas where prior knowledge isn't absolutely necessary, but it's a big advantage. For example you can teach a lot of statistics without teaching the way things are calculated, but doing the sums with pen and paper will give you a better understanding. A confidence interval has a meaning that can be explained using words to someone without the mathematical skills to be able to calculate it. So is a statistical distribution. Is it better to know who to calculate the probability distribution with a pen and paper? Yes and in some interesting cases, it's very important. That doesn't mean that a lot of useful statistics can't be taught to people who will take shortcuts using software. Knowledge keeps growing like a tree with thickening branches and sprouting branches of their own. We can't tell new people where the last generation started. In the world of increasingly specialized knowledge and the culture of lifelong learning that I hope is evolving, its important to find ways of dropping people into the middle of fields. AI for not mathematicians seems doable.

gtani 13y ago

(Besides deciding whether to title the book AI, machine learning, data mining/science, applied/computational stats) There's a number of books that take the "short equation/inequality" only approach to presenting the author's subset of his/her appelation. Marsland's is pretty good. But I'm not sure they impart much intuition before you dive into the traditional texts (usually one chooses from the Big 6 of Murphy, Bishop, Barber, Mackay, Hastie/Tibshirani/ or Koller/Friedman). Note 3 of those have content freely available to read online, as well as free stats texts from otexts.com, UMass' Lavine and CMU's Shalizi and Kadane (2 separate books), and UCSD's Levy (specialized on stats for linguistics/CL). Note also the ESL guys (+ 1 additional author) have released Intro to Statistical Learning http://www.springer.com/statistics/statistical+theory+and+methods/book/978-1-4614-7137-0 http://www.springer.com/statistics/statistical+theory+and+me...

vidarh 13y ago

If you're going to do AI research I agree with you that you'll probably need quite a bit of maths, eventually. But people can certainly learn a lot of interesting things without it, and for some that might just be the motivation to pick up some more of the maths useful for the more advanced aspects as well. My maths sucks, and I've had no problems reading quite a few AI papers through the years and pick up a lot of interesting ideas, and implemented quite a few things for fun.