Re: [computer-go] Re: Neural networks
Hideki Kato: <4ad5e7f1.77%hideki_ka...@ybb.ne.jp>:
>Álvaro Begué: <7b0793ea0910140721l2819723bl12af6c1c3dd9...@mail.gmail.com>:
>>We should let go of this idea that artificial neural networks have
>>anything to do with the brain. ANNs are just a family of parametric
>>functions (often with too many parameters for their own good) and
>>associated tuning algorithms ("learning" is a bit pretentious).
>>Perhaps they took vague inspiration in a cartoonish version of the
>>brain, but that's about it.
>
>As I wrote before, if you want general purpose approximater, use RL
>or SVM which performs much better than ANNs.
>
>>People tried to make flying machines by imitating birds for a long
>>time. Planes and helicopters fly, but not like birds do it. Similarly,
>>I believe that whenever we figure out how to make machines that play
>>go well, they will not do it like a brain does it.
>
>There are so many flying objects in the worlds such as leaves,
>bats, bees, not only birds. People can observe and compare them and
>then extract the essence of "flying". This is not the case of
>"thought".
Just making things clearer, I'd like to change the word "thought"
here to "intelligence".
Hideki
>Moreover, if we really wants flying machines like birds, say, more
>silent, gentle and elegant, perhaps we have to observe birds more
>precisely. It's possible that thinking machines are the case.
>
>Hideki
>
>>Álvaro.
>>
>>
>>On Wed, Oct 14, 2009 at 10:00 AM, Hideki Kato wrote:
>>> IMHO, when applying artificial neural networks to an application, the
>>> structure (as well as the learning algorithm) of the network is very
>>> important. For Go, we haven't invetigated the mechanism how the brain
>>> is used yet. Backpropagation-style layered network is just a model of
>>> the cerebellum and I strongly believe we need a higher-level model to
>>> replace the modern MCTS Go programs, say, how the cerebellum works
>>> together with the other areas of the brain (such as cerebrum and basal
>>> ganglia which is said working like RL) playing a game but it's not
>>> established nor proposed. If the model approximates the mechanism of
>>> real brain well enough, it never performs well.
>>>
>>> As a general purpose learning machine, neural networks perform much
>>> worse than sophisticated learning algorithms such as RL and also
>>> worse than suppoert vector machines, as Remi mentioned.
>>>
>>> Hideki
>>>
>>> Petr Baudis: <20091014122619.gu6...@machine.or.cz>:
Hi!
Is there some "high-level reason" hypothesised about why there are
no successful programs using neural networks in Go?
I'd also like to ask if someone has a research tip for some
interesting Go sub-problem that could make for a nice beginner neural
networks project.
Thanks,
>>> --
>>> g...@nue.ci.i.u-tokyo.ac.jp (Kato)
>>> ___
>>> computer-go mailing list
>>> computer-go@computer-go.org
>>> http://www.computer-go.org/mailman/listinfo/computer-go/
>>>
>>___
>>computer-go mailing list
>>computer-go@computer-go.org
>>http://www.computer-go.org/mailman/listinfo/computer-go/
>--
>g...@nue.ci.i.u-tokyo.ac.jp (Kato)
>___
>computer-go mailing list
>computer-go@computer-go.org
>http://www.computer-go.org/mailman/listinfo/computer-go/
--
g...@nue.ci.i.u-tokyo.ac.jp (Kato)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Neural networks
Álvaro Begué: <7b0793ea0910140721l2819723bl12af6c1c3dd9...@mail.gmail.com>:
>We should let go of this idea that artificial neural networks have
>anything to do with the brain. ANNs are just a family of parametric
>functions (often with too many parameters for their own good) and
>associated tuning algorithms ("learning" is a bit pretentious).
>Perhaps they took vague inspiration in a cartoonish version of the
>brain, but that's about it.
As I wrote before, if you want general purpose approximater, use RL
or SVM which performs much better than ANNs.
>People tried to make flying machines by imitating birds for a long
>time. Planes and helicopters fly, but not like birds do it. Similarly,
>I believe that whenever we figure out how to make machines that play
>go well, they will not do it like a brain does it.
There are so many flying objects in the worlds such as leaves,
bats, bees, not only birds. People can observe and compare them and
then extract the essence of "flying". This is not the case of
"thought".
Moreover, if we really wants flying machines like birds, say, more
silent, gentle and elegant, perhaps we have to observe birds more
precisely. It's possible that thinking machines are the case.
Hideki
>Álvaro.
>
>
>On Wed, Oct 14, 2009 at 10:00 AM, Hideki Kato wrote:
>> IMHO, when applying artificial neural networks to an application, the
>> structure (as well as the learning algorithm) of the network is very
>> important. For Go, we haven't invetigated the mechanism how the brain
>> is used yet. Backpropagation-style layered network is just a model of
>> the cerebellum and I strongly believe we need a higher-level model to
>> replace the modern MCTS Go programs, say, how the cerebellum works
>> together with the other areas of the brain (such as cerebrum and basal
>> ganglia which is said working like RL) playing a game but it's not
>> established nor proposed. If the model approximates the mechanism of
>> real brain well enough, it never performs well.
>>
>> As a general purpose learning machine, neural networks perform much
>> worse than sophisticated learning algorithms such as RL and also
>> worse than suppoert vector machines, as Remi mentioned.
>>
>> Hideki
>>
>> Petr Baudis: <20091014122619.gu6...@machine.or.cz>:
>>> Hi!
>>>
>>> Is there some "high-level reason" hypothesised about why there are
>>>no successful programs using neural networks in Go?
>>>
>>> I'd also like to ask if someone has a research tip for some
>>>interesting Go sub-problem that could make for a nice beginner neural
>>>networks project.
>>>
>>> Thanks,
>> --
>> g...@nue.ci.i.u-tokyo.ac.jp (Kato)
>> ___
>> computer-go mailing list
>> computer-go@computer-go.org
>> http://www.computer-go.org/mailman/listinfo/computer-go/
>>
>___
>computer-go mailing list
>computer-go@computer-go.org
>http://www.computer-go.org/mailman/listinfo/computer-go/
--
g...@nue.ci.i.u-tokyo.ac.jp (Kato)
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Neural networks
On Wed, Oct 14, 2009 at 10:21 AM, Álvaro Begué wrote:
> We should let go of this idea that artificial neural networks have
> anything to do with the brain. ANNs are just a family of parametric
> functions (often with too many parameters for their own good) and
> associated tuning algorithms ("learning" is a bit pretentious).
> Perhaps they took vague inspiration in a cartoonish version of the
> brain, but that's about it.
>
> People tried to make flying machines by imitating birds for a long
> time. Planes and helicopters fly, but not like birds do it. Similarly,
> I believe that whenever we figure out how to make machines that play
> go well, they will not do it like a brain does it.
>
> Álvaro.
>
>
Yes, I agree. It's very common in my opinion to try to bend the program too
much to how "we" do it. It's natural to try to imitate humans because
humans are indeed superior.But our hardware is so different that it does
not always make sense to imitate us.
Even among humans we do not always try to imitate each other, if we
recognize that something is missing.For instance a short basketball
player will need to play a different kind of game than a very tall
basketball player.He would choose a style that emphasized his strengths
and minimized his weaknesses.
In fact it's likely that we would be placing limitations on the computer if
we tried to imitate people too much.Right now chess programs are
superior to human players, but nobody has ever suggested that perhaps we
should try to imitate them!
Having said all of that, it's still important to analyze what works for
humans that perhaps could be effectively used for our programs. We just
don't want to take this too far if it's doesn't work.
- Don
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Neural networks
We should let go of this idea that artificial neural networks have
anything to do with the brain. ANNs are just a family of parametric
functions (often with too many parameters for their own good) and
associated tuning algorithms ("learning" is a bit pretentious).
Perhaps they took vague inspiration in a cartoonish version of the
brain, but that's about it.
People tried to make flying machines by imitating birds for a long
time. Planes and helicopters fly, but not like birds do it. Similarly,
I believe that whenever we figure out how to make machines that play
go well, they will not do it like a brain does it.
Álvaro.
On Wed, Oct 14, 2009 at 10:00 AM, Hideki Kato wrote:
> IMHO, when applying artificial neural networks to an application, the
> structure (as well as the learning algorithm) of the network is very
> important. For Go, we haven't invetigated the mechanism how the brain
> is used yet. Backpropagation-style layered network is just a model of
> the cerebellum and I strongly believe we need a higher-level model to
> replace the modern MCTS Go programs, say, how the cerebellum works
> together with the other areas of the brain (such as cerebrum and basal
> ganglia which is said working like RL) playing a game but it's not
> established nor proposed. If the model approximates the mechanism of
> real brain well enough, it never performs well.
>
> As a general purpose learning machine, neural networks perform much
> worse than sophisticated learning algorithms such as RL and also
> worse than suppoert vector machines, as Remi mentioned.
>
> Hideki
>
> Petr Baudis: <20091014122619.gu6...@machine.or.cz>:
>> Hi!
>>
>> Is there some "high-level reason" hypothesised about why there are
>>no successful programs using neural networks in Go?
>>
>> I'd also like to ask if someone has a research tip for some
>>interesting Go sub-problem that could make for a nice beginner neural
>>networks project.
>>
>> Thanks,
> --
> g...@nue.ci.i.u-tokyo.ac.jp (Kato)
> ___
> computer-go mailing list
> computer-go@computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Neural networks
IMHO, when applying artificial neural networks to an application, the structure (as well as the learning algorithm) of the network is very important. For Go, we haven't invetigated the mechanism how the brain is used yet. Backpropagation-style layered network is just a model of the cerebellum and I strongly believe we need a higher-level model to replace the modern MCTS Go programs, say, how the cerebellum works together with the other areas of the brain (such as cerebrum and basal ganglia which is said working like RL) playing a game but it's not established nor proposed. If the model approximates the mechanism of real brain well enough, it never performs well. As a general purpose learning machine, neural networks perform much worse than sophisticated learning algorithms such as RL and also worse than suppoert vector machines, as Remi mentioned. Hideki Petr Baudis: <20091014122619.gu6...@machine.or.cz>: > Hi! > > Is there some "high-level reason" hypothesised about why there are >no successful programs using neural networks in Go? > > I'd also like to ask if someone has a research tip for some >interesting Go sub-problem that could make for a nice beginner neural >networks project. > > Thanks, -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
