Join devRant
Do all the things like
++ or -- rants, post your own rants, comment on others' rants and build your customized dev avatar
Sign Up
Pipeless API
From the creators of devRant, Pipeless lets you power real-time personalized recommendations and activity feeds using a simple API
Learn More
-
Its called the butterfly effect.
But for an neural net its far too complex, we don't have enough data and processing power to create weather forecasts with a neural net.
The current forecasts are created with data for an region with an 10 kilometer diameter. The weather stations are not always equal per region and the data quality differs greatly .some record only temperature,some recodd only 3 times a day, others can send the data with an high precision and volume. Its an complicated situation that is hard to process with an neural net. -
@stop actually we do. but thats a poor example anyway as there would be more than just a weight and a bias involved.
how about a path of money between different intermediaries ? how much each node needs to spend towards a specific company or give to a person for money in to reach a financial institution to a certain proportion on the other side ? you know like money laundering works :P -
@MadMadMadMrMim In theory a neural net could solve any problem where the result can be trained.
Either by using known data or some algorithm to compute or test.
Its all about size and structure.
And structure is an emerging area.
Modern image recognition systems uses convoluted networks where, if I understand it, multiple networks are matched against the source multiple times in different positions making it much better to find a match and allowing to find multiple matches in one image.
This kind of mixed networks and similar will most likely advance alot but still, they cannot solve an unknown where we have no idea what is a solution, for example research which requires imagination. -
@Voxera I guess the question I'm asking is could you use the back propagation aspects of a neural net to fix or modify values within a certain tolerance in very very very large dataset to solve a very complex problem of causal factors who;s interaction could be defined in the simple number pair.
like
if harry meets sally, and sally stabs brad, then brad doesn't go to the ball and dance with suzy who meets patrick then and goes home with patrick which produces chuck 9 months later :P only with a huge number of other possibilities added. -
@MadMadMadMrMim probably. The problem to begin with is encoding all possible options as numbers and depending on the system that could be infinity complex which would make it practically impossible.
But as long as you can encode all states it should be doable.
One thing most networks so far are lacking is short term memory where you could feed it values over time and it uses recent history to, alI think there is experimentation going on but I have not seen any common practical example for it.
One problem is the increased complexity and input size. -
@Voxera oops. that was a repeat.
-
@Voxera I would like to tossle your hair. which i do hope is dark. what sentiment does that suggest ? :P a robot could not answer this.
also tell me what image goes into your mind at the idea. -
@Voxera I think you remember the original comment :P she had the expected response during the 'lets try to fuck everyone over' phase ;P
-
-
@Voxera come on now show some effort
I’m beginning to think the last gen of doctors dentists and lawyers left without conferring much expertise or are just too important to pay attention
Maybe we had a shortage because the last gen was completely fucking psycho
Kavorkian thrill killers that make their patients sick are not really qualified -
Most people wouldn’t sleep at night or talk to anyone if they knew what kind of psychos were out there
And to think some of these psychos tried to deflect off me because I used to get a little agitated back in the day which is not the sign of one of these sneaky fucking psychos who do everything they can to deflect negative attention
Could neural nets be used to solve a complex problem with a lot of predefined specific weights and biases related to real activities that could be numerically represented ?
Like if the various layers represented things who’s outcome would be cascading and a direct outcome of factors inputted to them resulting in limited number of outputs ? Like say maybe wind flow at locations resulting in a wind current at an expected time in another location and where a system would have to change to in order to result in the final expected i output ? Maybe a bad example as that could be affected by a lot of things .
But basically the gradual massaging of values that would relate to causal effects where a specific output was designed being intermediaries between the desired input and output ?
question