A Small Team of Student AI Coders Beats Google's Machine-Learning Code

Students from Fast.ai, a small organization that runs free machine-learning courses online, just created an AI algorithm that outperforms code from Google's researchers, per an important benchmark. From a report: Fast.ai's success is important because it sometimes seems as if only those with huge resources can do advanced AI research. Fast.ai consists of part-time students keen to try their hand at machine learning -- and perhaps transition into a career in data science. It rents access to computers in Amazon's cloud. But Fast.ai's team built an algorithm that beats Google's code, as measured using a benchmark called DAWNBench, from researchers at Stanford. This benchmark uses a common image classification task to track the speed of a deep-learning algorithm per dollar of compute power. Google's researchers topped the previous rankings, in a category for training on several machines, using a custom-built collection its own chips designed specifically for machine learning. The Fast.ai team was able to produce something even faster, on roughly equivalent hardware.

Anyway, they were students

[Tough Love]

They were students. Now they are Google employees.

Good luck, you will need it.

AI Algorithm?

[110010001000]

How does an AI algorithm differ from a plain old algorithm. I am so curious...

Re:

[Anonymous Coward]

Its Buzzword Factor is higher.

Re:

[110010001000]

Impressive. So it is like an algorithm, BUT also has Artificial Intelligence. I wonder what makes them think their algorithm is artificially intelligent.

Re:

[dinfinity]

Technically the students didn't really write or design an algorithm that works well in the benchmark. That very fact is what sets 'AI' apart from algorithms. You could argue that the students designed an algorithm that produces an algorithm that works very well in this case, but it is then still the resulting algorithm that works well for the task at hand, not the system they designed that 'came up' with the algorithm.

I'd say that calling these things 'emergent algorithms' would be more appropriate or even

Re:

[q_e_t]

There are many sorts of neural networks. Using FP for weights isn't necessarily required, although it's certainly simpler to program backprop that way than try to use integer mathematics. Backprop isn't the only form of neural network available, although to some extent it depends on how you define neural network.

Re:

[Aighearach]

How does an AI algorithm differ from a plain old algorithm. I am so curious...

The use case.

Apple 1

[AlanBDee]

Fast.ai's success is important because it sometimes seems as if only those with huge resources can do advanced AI research.

Any one who believe's that needs to be hit over the head with a replica of the Apple 1.

Heaven forbid they form a startup. Google buyout.

[Seven Spirals]

Well, if you can't beat them, buy them. Kids, whatever you do don't go to work for Google. Make them buy you out, don't give them your skills for salary. Squeeze these corporate fuckers as hard as you can for the edification of the rest of us who didn't write a great algorithm that beats the Googleplex-of-assholes.