[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gneuralnetwork] Eventual Role for command-line gneuralnetwork.

From: Neil Simmonds
Subject: Re: [Gneuralnetwork] Eventual Role for command-line gneuralnetwork.
Date: Sun, 9 Oct 2016 13:41:23 +0100

Here, here !

I was interested in gneuralnetwork when thinking of connections to gnu radio - far too much for me as a developer but scripts and gnu radio ui sockets I can manage.  I'm an old time slightly unpractised time lisp guy.


On 7 October 2016 at 21:02, Ray Dillinger <address@hidden> wrote:

In an earlier post I mentioned that I want there to be a highly
scriptable command line gneuralnetwork utility.  In this post
I want to talk about why, and how I want to be able to use it
once it becomes usable.

Essentially it comes down to wanting neural networks to be a
basic utility, that integrates with all the classic scripted
unix utilities and can be managed using shell scripts, cron
jobs, etc.  It's not just about training neural networks and
experimenting with them, although that's a big part of it.
It's also about making it simple and easy to deploy neural
networks in useful roles.

I intend to fix it so that somebody who does a little shell
scripting can set up neural networks to evaluate whether
incoming email are spam, or decide what to recommend to a
website visitor, or feed text or HTML through to translate
it into another language, etc.  People doing development at a
deeper level would do custom database integrations, language
runtime wrappers, package it into pre-built applications for
particular purposes, etc. People simplifying the learning/
training/experimenting aspect for the convenience of GUI-only
users would hang a graphical interface on it.

But it's important for there to be a very easy and general
way to integrate it with all the traditional unix tools, and
to me that means a scriptable tool that's usable in all the
classic ways that Unix scriptable tools get used.

An example of the things I want to do is make a straightforward
bash script that scans files in /var/log and /var/tmp and /proc,
etc, feeds them through sed (or whatever) into a neural network
that reads them on stdin, reads the neural network output that
gets piped back to it, and tells me whether a particular type
of rootkit or malicious software is running on the machine.

Or make a neural-network utility that's a hell of a lot smarter
and less intrusive than Clippy/Cortana/Siri, about how to
answer general help questions.  In an ideal universe it could
even read the man page for some utility unfamiliar to it, and
suggest how to invoke that utility to achieve a given result
which a user merely babbled imprecisely about in English.

Or scan the 'cron' file and then answer questions, in English,
about what's running at what time.

Or scan the 'man' directories and invoke a neural network that
produces Italian-language automatic translations of man pages
which are available in English but not in Italian.

Some of these things may be beyond our near-term capabilities,
and some of them will be a bugger to figure out how to train,
but that's a general direction I want to be able to go with a
command-line gneuralnetwork tool.

If they are relevant, then neural networks are not just an
experiment, not just a toy, and not just for multi-million dollar
paid applications.  They're a basic utility and should be usable
as such, in ways straightforward enough to be invoked at the
command line of a shell, in a script written by a single person
for a single purpose.  A purpose that neither we, nor any software
developer with a team of engineers paid to integrate it into a
custom application, ever thought of or cared about.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]