4 ms·

After digging a bit into machine learning i feel like there is a tendency to be fancy with terminology. For example, do we really need to use the obscure term

by jcims 7y ago

After digging a bit into machine learning i feel like there is a tendency to be fancy with terminology. For example, do we really need to use the obscure term ’stochastic’ when ’random’ exists?

Fredej 7y ago
I do think there's some value in being able to tersely express your exact point to a peer. For example "a random process" could mean "any process in existence" whereas "a stochastic process" clearly states stochastic as being a property of the process. I however agree that things can sometimes get a bit wrapped up in sounding cool just to sound cool.
- subroutine 7y ago
  As a rule of thumb you should never use a big word when a diminutive word will suffice.
MaximumYComb 7y ago
Are the terms that interchangeable? "I used a random algorithm" is a different expression from "I used a stochastic algorithm". For me, the term stochastic is one half of the stochastic vs deterministic categorisation.
thanatropism 7y ago
These aren't synonyms. Random is a property of the world itself; stochastic means "we allow ourselves to think of this process as random because we can't/don't want to deal with it exactly". Following this is the difference between Random Differential Equations and Stochastic Differential Equations.
twic 7y ago
You're still using stochastic processes? I switched to aleatory ones after reading a DeepMind pre-print that's been going round.
Angostura 7y ago
I think that's orthogonal to the original question
kragen 7y ago
Among the meanings of random are: 1. Stochastic. 2. Unbiased. 3. Typical and average. 4. Unprompted or unjustifiable. 5. Habitually using non sequiturs. So, I might use a random sort algorithm like, I don't know, heapsort. Or I might be reading through some code and find a random sort algorithm --- why is there a mergesort in the middle of this calendar routine? Or, if you want a random sort algorithm, you should be sure to use the Fisher–Yates shuffle. Or, to avoid the quadratic worst case of treesort and quicksort with high probability, you can convert them into random sort algorithms by picking partition elements randomly. Of these four random sort algorithms, only the fourth is stochastic. (I haven't figured out how a sort algorithm could habitually use non sequiturs.)
jerf 7y ago
I think that one doesn't come from being fancy, but from machine learning being the terminologically-uncomfortable collision between statics mathematicians and computer scientists, who previously had essentially no cross-talk with each other. There's a number of similar places where ML uses statistics terms rather than ones you'd expect from a computer science (as a branch of mathematics) heritage.
- jcims 7y ago
  This makes sense to me. Thanks for the insight!
platoscubicle 7y ago
I like Robert Anton Wilson's definition. "A stochastic process is a random series, but it is a special kind of random series. In a stochastic process, some agent or agency is making selections—picking out of the randomness a pattern that is not random."

by jcims 7y ago

Fredej 7y ago

I do think there's some value in being able to tersely express your exact point to a peer. For example "a random process" could mean "any process in existence" whereas "a stochastic process" clearly states stochastic as being a property of the process. I however agree that things can sometimes get a bit wrapped up in sounding cool just to sound cool.

subroutine 7y ago
As a rule of thumb you should never use a big word when a diminutive word will suffice.

MaximumYComb 7y ago

Are the terms that interchangeable? "I used a random algorithm" is a different expression from "I used a stochastic algorithm". For me, the term stochastic is one half of the stochastic vs deterministic categorisation.

thanatropism 7y ago

These aren't synonyms. Random is a property of the world itself; stochastic means "we allow ourselves to think of this process as random because we can't/don't want to deal with it exactly". Following this is the difference between Random Differential Equations and Stochastic Differential Equations.

twic 7y ago

You're still using stochastic processes? I switched to aleatory ones after reading a DeepMind pre-print that's been going round.

Angostura 7y ago

I think that's orthogonal to the original question

kragen 7y ago

Among the meanings of random are: 1. Stochastic. 2. Unbiased. 3. Typical and average. 4. Unprompted or unjustifiable. 5. Habitually using non sequiturs. So, I might use a random sort algorithm like, I don't know, heapsort. Or I might be reading through some code and find a random sort algorithm --- why is there a mergesort in the middle of this calendar routine? Or, if you want a random sort algorithm, you should be sure to use the Fisher–Yates shuffle. Or, to avoid the quadratic worst case of treesort and quicksort with high probability, you can convert them into random sort algorithms by picking partition elements randomly. Of these four random sort algorithms, only the fourth is stochastic. (I haven't figured out how a sort algorithm could habitually use non sequiturs.)

jerf 7y ago

I think that one doesn't come from being fancy, but from machine learning being the terminologically-uncomfortable collision between statics mathematicians and computer scientists, who previously had essentially no cross-talk with each other. There's a number of similar places where ML uses statistics terms rather than ones you'd expect from a computer science (as a branch of mathematics) heritage.

jcims 7y ago
This makes sense to me. Thanks for the insight!

platoscubicle 7y ago

I like Robert Anton Wilson's definition. "A stochastic process is a random series, but it is a special kind of random series. In a stochastic process, some agent or agency is making selections—picking out of the randomness a pattern that is not random."