From: "Alejandro R. Mosteo" <alejandro@mosteo.com>
Subject: Re: Killing software and certification
Date: Wed, 28 Mar 2018 19:06:14 +0200
Date: 2018-03-28T19:06:14+02:00 [thread overview]
Message-ID: <p9gi27$ld9$1@dont-email.me> (raw)
In-Reply-To: <p9g8h7$r72$1@gioia.aioe.org>
On 28/03/18 16:23, Dmitry A. Kazakov wrote:
> On 28/03/2018 15:54, Alejandro R. Mosteo wrote:
>> On 27/03/18 21:25, Dmitry A. Kazakov wrote:
>>> On 2018-03-27 20:32, Alejandro R. Mosteo wrote:
>
>>>> I'm not in the industry, and I'd be surprised that unverified
>>>> software were allowed to run in civilian environments where failures
>>>> basically amount to a very dangerous situation.
>>>
>>> Why should it surprise you? How are you going to verify it? Black box
>>> test is impossible. White box test isn't either, assuming any NN
>>> involved. There is nothing to prove.
>>
>> I can think of an spectrum of regulatory/practical positions between
>> 'nothing can be done, so everyone brace' and 'this won't ever fail,
>> under every [un]conceivable situation'. It's the apparent nonchalance
>> of the general public that coexists with these testing cars, the
>> brashness/recklessness of those expecting to get rich with it and the
>> apparent willingness of politicians that I find fascinating (that's the
>> first ones that comes to mind).
>>
>> I can understand the appeal for politicians to be the first city with a
>> working fleet (or whatever contributions they're getting to favor live
>> testing). As a technophile, I want autonomous cars to become reality,
>> so I can understand that too. As a researcher familiar with the
>> algorithms involved and with the kind of C/C++/Python heaps that
>> implement them I get chills about thinking that a car can be on the
>> highway with a semi-awake safety driver as the only fallback in a split
>> second.
>
> As much as I hate C and the ilk, the problem is much deeper, I'm afraid.
> Even if it were 100% SPARK Ada, a self teaching system [I don't consider
> here simple cases where one can prove convergence of the training method
> to a defined goal], is in the same relation as a program is to the CPU.
>
> You can verify and certify the CPU as much as you wish. That would say
> nothing about the program running on it. The weights of a NN is a
> "program". The system core written in whatever language is a "CPU".
>
> I have no slightest idea of an approach to define correctness of the
> trained weights, even less about proving it. It is a fundamental
> challenge we will have to deal with, if this kind of "emerging
> programming" to take hold.
>
> A kind of biblical disaster when you, like God did, see in dismay what
> your creatures would do after you granted them "free will". Would you
> drown them all? (:-))
This line of thinking is often brought by a colleague working on
"classical" solutions to problems that are nowadays trendy on deep
learning circles. When feeling optimistic I see it as the complexity of
the simplex method: linear on average but worst case exponential. I tend
to think that the same you could have a watchdog for a stray simplex, you
could have some fallback for a DNN behaving badly (if you can detect it
in the first place :P).
In other words: if DNNs prove to be as useful as they promise, they'll
find a way to statistically cover the desired percent of reliability.
Maybe this is another field of research in the making. I'm also told that
"we solve X with a DNN" is no longer acceptable in the main conferences,
that you have to also provide some insight on the DNN workings. But this
is hearsay that I pass along.
next prev parent reply other threads:[~2018-03-28 17:06 UTC|newest]
Thread overview: 100+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-03-18 13:33 Ada-Oriented GUI Jeffrey R. Carter
2018-03-19 2:03 ` Dan'l Miller
2018-03-19 14:10 ` Dan'l Miller
2018-03-21 9:49 ` Alejandro R. Mosteo
2018-03-21 13:58 ` Dan'l Miller
2018-03-21 18:43 ` briot.emmanuel
2018-03-21 19:17 ` Shark8
2018-03-21 22:40 ` Randy Brukardt
2018-03-21 23:52 ` Shark8
2018-03-22 6:50 ` briot.emmanuel
2018-03-22 16:56 ` Shark8
2018-03-23 16:29 ` Shark8
2018-03-23 22:59 ` Randy Brukardt
2018-03-23 23:43 ` Mehdi Saada
2018-03-26 22:09 ` Randy Brukardt
2018-03-27 7:27 ` Dmitry A. Kazakov
2018-03-27 23:58 ` Randy Brukardt
2018-03-28 7:09 ` Dmitry A. Kazakov
2018-03-22 17:34 ` Alejandro R. Mosteo
2018-03-22 17:50 ` Dan'l Miller
2018-03-22 18:58 ` Shark8
2018-03-23 12:06 ` Alejandro R. Mosteo
2018-03-20 16:41 ` Dan'l Miller
2018-03-20 21:34 ` Randy Brukardt
2018-03-21 2:22 ` Dan'l Miller
2018-03-21 21:50 ` Randy Brukardt
2018-03-22 8:45 ` Dmitry A. Kazakov
2018-03-22 10:58 ` Bojan Bozovic
2018-03-22 11:03 ` Bojan Bozovic
2018-03-21 8:25 ` Dmitry A. Kazakov
2018-03-21 14:30 ` Dan'l Miller
2018-03-21 15:57 ` vincent.diemunsch
2018-03-21 17:33 ` Dan'l Miller
2018-03-21 16:27 ` Dmitry A. Kazakov
2018-03-21 17:04 ` Dan'l Miller
2018-03-21 17:42 ` Dmitry A. Kazakov
2018-03-21 18:19 ` Dan'l Miller
2018-03-21 19:11 ` Simon Wright
2018-03-21 19:51 ` Dan'l Miller
2018-03-21 20:11 ` Dmitry A. Kazakov
2018-03-21 20:33 ` Dan'l Miller
2018-03-21 22:16 ` Dan'l Miller
2018-03-22 9:12 ` Dmitry A. Kazakov
2018-03-22 14:57 ` Dan'l Miller
2018-03-22 15:46 ` Bojan Bozovic
2018-03-22 14:00 ` Dan'l Miller
2018-03-22 17:29 ` Alejandro R. Mosteo
2018-03-21 21:58 ` Randy Brukardt
2018-03-26 21:20 ` G. B.
2018-03-21 22:33 ` Randy Brukardt
2018-03-22 1:43 ` Dan'l Miller
2018-03-22 23:47 ` Randy Brukardt
2018-03-23 2:37 ` Dan'l Miller
2018-03-23 22:42 ` Randy Brukardt
2018-03-24 7:47 ` Simon Wright
2018-03-23 9:05 ` Jeffrey R. Carter
2018-03-23 9:48 ` Bojan Bozovic
2018-03-23 10:20 ` Alejandro R. Mosteo
2018-03-27 18:32 ` Killing software and certification (was: Ada-Oriented GUI) Alejandro R. Mosteo
2018-03-27 19:25 ` Killing software and certification Dmitry A. Kazakov
2018-03-28 13:54 ` Alejandro R. Mosteo
2018-03-28 14:23 ` Dmitry A. Kazakov
2018-03-28 17:06 ` Alejandro R. Mosteo [this message]
2018-03-28 19:35 ` Dmitry A. Kazakov
2018-03-28 15:47 ` Jeffrey R. Carter
2018-03-28 17:02 ` Dennis Lee Bieber
2018-03-28 17:59 ` Dan'l Miller
2018-03-27 19:41 ` Killing software and certification (was: Ada-Oriented GUI) Dan'l Miller
2018-03-28 0:04 ` Randy Brukardt
2018-03-28 2:27 ` Dan'l Miller
2018-03-28 13:54 ` Killing software and certification Alejandro R. Mosteo
2018-03-28 0:21 ` Killing software and certification (was: Ada-Oriented GUI) Jere
2018-03-28 13:54 ` Killing software and certification Alejandro R. Mosteo
2018-03-23 12:31 ` Ada-Oriented GUI Alejandro R. Mosteo
2018-03-23 12:59 ` Dmitry A. Kazakov
2018-03-23 16:16 ` Dan'l Miller
2018-03-23 17:18 ` Dmitry A. Kazakov
2018-03-23 18:31 ` Dan'l Miller
2018-03-23 20:06 ` Dmitry A. Kazakov
2018-03-23 20:48 ` Mehdi Saada
2018-03-23 21:18 ` Dmitry A. Kazakov
2018-03-24 11:36 ` Alejandro R. Mosteo
2018-03-24 13:12 ` Dmitry A. Kazakov
2018-03-28 14:09 ` Alejandro R. Mosteo
2018-03-28 15:02 ` Dmitry A. Kazakov
2018-03-28 18:07 ` Alejandro R. Mosteo
2018-03-29 7:58 ` Dmitry A. Kazakov
2018-04-02 22:13 ` Robert I. Eachus
2018-04-03 8:31 ` Dmitry A. Kazakov
2018-04-03 22:32 ` Robert I. Eachus
2018-04-04 7:30 ` Dmitry A. Kazakov
2018-03-25 12:57 ` Jeffrey R. Carter
2018-03-24 16:33 ` Dan'l Miller
2018-03-26 22:29 ` Randy Brukardt
2018-03-27 0:15 ` Dan'l Miller
2018-03-27 6:08 ` Dennis Lee Bieber
2018-03-27 7:52 ` Simon Wright
2018-03-27 14:48 ` Dennis Lee Bieber
2018-04-01 17:37 ` Robert I. Eachus
2018-03-25 19:19 ` Andrew Shvets
replies disabled
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox