Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Building an AI to predict human age from a blood sample

colekillian.com

Post reply on HN
Print view View original

31–40 of 68 posts

Re: Building an AI to predict human age from a blood sample

#31

Post by fock »

siscia wrote: post #22

When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression. Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar…

that's exactly the thing I'm seeing in my field (computational materials science). Basically a simple regression model (with very simple features) brings you 90% there, still people compete on publishing (on ONE shitty benchmark dataset) ever better results – the most-cited people are using KRR, e.g. each fit uses 2TB of RAM and "days" of CPU (features of length O(1000) and 100000 samples). The sample data is probably a 30s calculation (and still only a rough estimation). Sometimes wants you to question science, but hey, writing proposals with "ML" in it gives at least a chance on that grant...

Re: Building an AI to predict human age from a blood sample

#32

Post by gmtx725 »

wyxuan wrote: post #3
gmtx725 wrote: post #2

why would you use a neural net on a dataset with only 700 samples, smh

The relationship in the dataset seems pretty clean and clear cut, so you don't really need as large a dataset.

Then just use a GBM or some simple linear classifier

Re: Building an AI to predict human age from a blood sample

#33

Post by profunctor »

chewxy wrote: post #11
deepnotderp wrote: post #10

Earlier quoted context omitted.

I know this is a joke, but the theory of generalization in NNs is rapidly advancing and it's not quite that simplistic: https://arxiv.org/abs/2003.02139

Ya. And the choice of optimizer (in this case adam) also imposes upon it some regularization scheme. I just thought I'd highlight a bit of funniness.

How does Adam provide regularisation? I’d never heard of this before and I don’t recall it from when I read the paper.

Re: Building an AI to predict human age from a blood sample

#34

Post by sacado2 »

siscia wrote: post #22

When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression. Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar…

> Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar to the one of the neural network.

Plus, LR is not a black box, so it both brings you a class and a reason why that class was chosen, which is a very desirable property in many problems.

Re: Building an AI to predict human age from a blood sample

#35

Post by 1000units »

fock wrote: post #31
siscia wrote: post #22

When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression. Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar…

that's exactly the thing I'm seeing in my field (computational materials science). Basically a simple regression model (with very simple features) brings you 90% there, still people compete on publishing (on ONE shitty benchmark dataset) ever better results – the most-cited people are using KRR, e.g. each fit uses 2TB of RAM and "days" of CPU (features of length O(1000) and 100000 samples). The sample data is probabl…

--- guy who writes "O(1000)" telling you how ML research is flawed

Re: Building an AI to predict human age from a blood sample

#36

Post by KaoruAoiShiho »

fock wrote: post #31
siscia wrote: post #22

When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression. Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar…

that's exactly the thing I'm seeing in my field (computational materials science). Basically a simple regression model (with very simple features) brings you 90% there, still people compete on publishing (on ONE shitty benchmark dataset) ever better results – the most-cited people are using KRR, e.g. each fit uses 2TB of RAM and "days" of CPU (features of length O(1000) and 100000 samples). The sample data is probabl…

I haven't thought as much about it as you have probably, but ok you get 90% of the way there, but now what? How do you get that last 10%? It would be a huge amount of work right?

Re: Building an AI to predict human age from a blood sample

#38

Post by hef19898 »

KaoruAoiShiho wrote: post #36
fock wrote: post #31

Earlier quoted context omitted.

that's exactly the thing I'm seeing in my field (computational materials science). Basically a simple regression model (with very simple features) brings you 90% there, still people compete on publishing (on ONE shitty benchmark dataset) ever better results – the most-cited people are using KRR, e.g. each fit uses 2TB of RAM and "days" of CPU (features of length O(1000) and 100000 samples). The sample data is probabl…

I haven't thought as much about it as you have probably, but ok you get 90% of the way there, but now what? How do you get that last 10%? It would be a huge amount of work right?

Isn't that usually the case? That the first 90% are as difficult to achieve as the next 5%, which are again as hard as the next 3%?

Re: Building an AI to predict human age from a blood sample

#39

Post by saystupidthings »

sacado2 wrote: post #34
siscia wrote: post #22

When I see work like these it get me the impression that ML hype is way too real. The goodness of the result of a machine learning model like this one should be compared with the goodness of a simple "standard" model like linear regression. Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar…

> Yeah it is kinda cool that we can use 10 lines of TF to spin up huge computation, but I guess that a simple linear regression would have provide results that are at least similar to the one of the neural network. Plus, LR is not a black box, so it both brings you a class and a reason why that class was chosen, which is a very desirable property in many problems.

There are many ways to get around the BB problem. Calculating activations and uncertainty is enough to deduce reason, and throw out risky choices that should be human reviewed.

Re: Building an AI to predict human age from a blood sample

#40

Post by KaoruAoiShiho »

hef19898 wrote: post #38
KaoruAoiShiho wrote: post #36

Earlier quoted context omitted.

I haven't thought as much about it as you have probably, but ok you get 90% of the way there, but now what? How do you get that last 10%? It would be a huge amount of work right?

Isn't that usually the case? That the first 90% are as difficult to achieve as the next 5%, which are again as hard as the next 3%?

Basically the argument is that ML would be able to get there easier.
Post reply on HN