back to article From drugs to galaxy hunting, AI is elbowing its way into boffins' labs

Powerful artificially intelligent algorithms and models are all the rage. They're knocking it out of the park in language translation and image recognition, but autonomous cars and chatbots? Not so much. One area machine learning could do surprisingly well in is science research. As AI advances, its potential is being seized …

Anonymous Coward

"Only the top 50 most-promising moves are played out, [...]"

Determining a "most promising move" sounds like it is the most difficult part of the algorithm. Presumably based on some weighted properties that are part of the fixed input eg cost of ingredients, energy needs, time, filtration, purity, ...?

1
0
Anonymous Coward

Machine Learning is not AI.

See Title...

4
0

Re: Machine Learning is not AI.

Correct.

It is barely even Maths.

1
0

This post has been deleted by its author

Gold badge

"Tweak the generator to find out how best it should be changed to get past the discriminator."

Oh dear.

It seems we have found an ideal algorithm for generating endless amounts of fake news.

Yay for that advance on the state of the art (in writing fake news).

0
0

Clueless button pushing

Machine learning can be a very useful tool, but the flip-side is that clueless button pushers can use it to produce seemingly interesting output: http://shape-of-code.coding-guidelines.com/2015/11/23/machine-learning-in-se-research-is-a-bigger-train-wreck-than-i-imagined/

1
0
Anonymous Coward

Sorry but this is just nonsense. Machine/statistical learning methods have been used in areas from astrophysics genomics for decades now, and much progress has been made using them. Renaming neural nets as deep learning and slapping the label on everything randomly doesn't magically undo history.

We must be close to peak AI-bubble given the constant breathless hype.

1
0
Bronze badge

You are correct insofar as the basic neural network techniques underlying "deep learning" have been around since at least the 80s (when they were already over-hyped in some quarters), if not earlier. What has changed is simply computing power and availability of huge data sets. I don't believe neural network-based machine learning was that widely deployed in earnest in the past, because it wasn't particularly effective given computing/data capacity at the time. That has changed, and it seems to me that neural networks in machine learning are indeed beginning to fulfil their early promise.

I do agree that the hype is irritating, and that broader appreciation of the history behind the techniques would be nice to see; but trashing a promising technology on whatever grounds doesn't seem particularly helpful either.

As for the ongoing "but machine learning isn't AI" whinge... fine, if someone can tell me what "real AI" is suppose to look like (actually no, really, please don't do that!) My suspicion is that machine learning, "deep learning", whatever, will turn out to be a building block on the road to more sophisticated machine intelligence, rather than a done deal. Meantime, if it's useful let's use it.

0
0
Silver badge
Boffin

GAN?

GAN might result in nice pretty pictures, but that's all it is, you can only do science on the real data.

0
0
Bronze badge

Re: GAN?

GAN does use the real data - that's the whole point: "The system is fed an input of images and the generator tries to recreate the pictures, while the discriminator tries to distinguish if the computer-made images are real or generated." It is being used to clean up degraded images: "Pairs of galaxy images are used during training. One image is high quality, and the other is degraded by blurring the image and adding noise to it." As such, it is as valid as other techniques such as the "traditional deconvolution methods" mentioned in the article, and which it apparently outperforms..

0
0

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Forums

Biting the hand that feeds IT © 1998–2017