Artificial Neural Network for Drugs Anyone

[fastandbulbous]

How do you punish it, turn it's core processor voltage down?

 

[lineartransform]

Lariyn - If I may, I would suggest simply using a retail neural net software program, as coding your own backprop algorithm is really just a big waste of time when someone else has done it for you.

Your data set and how you encode your input will be your biggest issue. I think that an ideal sub-project before the full attempt would be to see if you could predict whether a drug is preferred or not given preferences for other drugs.

ie the "psychedelic predictor" - will you like psychedelics, yes or no, given a set of 10 yes or no questions involving previous drug use and personality.

Then worry about HighNet (my personal pun on SkyNet, gah, sorry for the awkwardness). 8)

 

[lariyn]

Well right the inputs did not converge when I started making networks that would tell you to do more than one thing per day. I am working with making multiple neural networks that gives an output then fed into a bigger neural network as the inputs to give a meaningful answer. So one big neural network. I kinda lost hope when I first saw that the input would not converge but I am actually making some progress atm and my expectations has increased.

lineartransform i do like your answer of starting small and im following that path atm. That psychedelic predictor sounds awesome that is what i am going to work on. But as far as using a commercial implementation I would not share the same opinion. The one i am using is full featured for everything i need and much more. I have like 8 different neural network implementation in C and C++, not included the ones i deleted or lost. I wish to gain more recurrent neural network design, and with my app ima ganna study the weight distributions, firing patterns, algorithms patterns, ima implement a way to view the data in graphs, and try and tear some large networks apart and see whats happening inside. this is a stepping stone to a network to tell you exactly what to do during a day while learning your behaviors, with recurrent networks so like it would know you had more energy if you woke up later, and would base outputs on previous data. Then ima move onto making entities interact with there environment and see if i can get something working even if the nets are completely trained for a particular schedule which if done right could be done easy.

Back Propagation of Error is really trivial. Its just backpropagation of the Error using the Delta Rule then forward propagation using the activation to calculate the Synaptic Sensitivity usually called Weight.

Here ya go without momentum, momentum is stupidly easy waste of time to post. I do not know if anyone here will understand this math, but its really simple bellow pre calculus level really only special symbol used is summation notation. Btw I forgot to label variable l, as it is the learning rate. Right off bat i see should used greek symbol for delta not d do not know how i missed that, anyone in higher level math knows what ima talking about man i love math that should be obvious plus i spent 0 seconds on all my names!

And btw ima get this thread back on topic ima stray from any non drug related questions or posts.

 

[lariyn]

Ok just to fill everyone in almost done with HID in c++ for my engine then ganna port my vb app for the ann building and modification and expand it. If anyone cares its fucking out of this world lol no joke wish i was. But ganna have to wait a little as hard time thinking and docs saying im having psychotic episode again people talking about me and the voices are starting up again, soo if everyone can be wait thanks if you care at all i may be the only one, but thanks again for the interest. And btw wish i was joking, cause pschosis is no joke wish i was pease do not talk shit, as i believe i am ok this is what people have told me!

 

[lariyn]

I am back and better than ever

Back on the project, and almost have my engine recoded again for the million time around. User interface is done, recoding it one more time, then I placing the perfect extension in. Will not take me long once that is done ill just post up on that day of the binary *.exe *.dll *.sof *.scr *.sms *.str and *.sdb. All these files are for configurations\scripts\textures\mesh\database of objects\and a snapshot of your objects memory( only the data that matters ) so it can boot up in the same configuration as I lock anyone using it to. Many security threats if I let anyone know anything about this projects and my engine. Plans already underway to post this to a android and development of fixed point math\accelerated math function was done before i ever started this thread allowing it to run on such a weak platform, it they always have crap ram in handhelds.

Just wanted to fill you in on progress, and hope you enjoy my android app.
This project is meant so I can make a dynamic real time scheduler and have entities interacting with there environments, while also retraining them self( only a portion of the network is unlocked ) they will learn patterns from entities around them so if they always get the food they hate as all the rest is gone they will get angry and act differently. Eve thought they were sporting sandbox, but no this design is pure sandbox effect( butterfly effect ) on magnitudes way beyond anything Eve is attempting atm. No my networks do not have true emotions. It has a internal structure within the Network that is trained before I run the simulation and then locked the values and duplicate those selected neuron's weights to target network. So they have predetermined responses( and your network that contains it will have different set of information feeding into the emotion structures causing each entity to have different emotions that work the same) that will fool u into looking more complex than it is. Sounds overly complex but not.

Anyways already done designs so back to c++ world and bye bye paper. Screenshot of neural network configuration and placement of neurons(i can maintain WorldView\ModelView or any transforms, leaving the neuron blind to its physical location which is implemented outside the CArtificialNeuralNetwork class.

Anyways is anyone still interested as its moments away from alpha testing\demo? And if anyone is interested there is a option to save your data used from birth to train your network along with activations and settings, so I can combine as much information as possible to make a generalized predictor. If you would just email the generated ArtificialNeuralNetwork_Predictor.sof and ArtificialNeuralNetwork_Predictor_forDrugs.sof - holds values extrinsic to the neural network but intrinsic to the Specialized form of a Drug Predictor. Get everyone you know to use this app because with enough information I can guess what drugs certain people like to do at certain time. And i lock a portion of the network(i place filters in certain places locked) to overturn a potentially very bad decision, so atm i am using a delicate implementation with two rows sitting on top of the network between outputs, recurrent with this filtering is causing the networks to blow up.

Here is a picture of a network I was working with just messing around. To me this is so simple I can do it in my sleep, yet have not seen anyone explicitly( except research papers where I got some of my idea and just combined what I came up with ) Configure\Initialize a NeuralNetwork that is anything like this on the global scale.

Does this look complex, I see a very complex design using simple information theory undergoing propagation within a network. Looks trivial( simple\elementary ) to me as I also just threw this together with effort. Also bias forms foundation of long term memory by adjusting weights tied to bias activation propagation, and the recurrent design gives rise to a attempt of short term memory. If you guys knew the data I have been sending it and receiving you would see how this is a sad excuse for a brain, but with the data presented correctly throws u off making it look more complex. What makes this Artificial Neural Network Special is it can do all this while multitasking so if you have giant structures with delay, along with recurrent as there are many ways this happens. But the network( outside training, unless time lapse session\batch training - experiementing with batch back propagation so i can see if it works on the global scale ) can proccess multiple set of inputs from different iteration at the same time, so as your propagation moves further in the tree there is always the following iteration of data propagating 1 neuron behind, and it will fill the entire network up with so many sets of data at the same time.

DAMAGE PREVENTION is our focus here, some people just need a good influence telling them they are going to far and need to stop. Or tells them( and I do not advise as Now that I am clean I know that stuff is evil ) to smoked a joint or drink(little bit as this is gum under my shoe and so bad for you) instead of smoking a pound of meth.

Also I am moving between 17 Projects ATM! So you know im not fooling around and I get back as I have a app that does some scheduling of stuff telling you what to do, and edit it to allow you to choose what its deciding. These network have to know NOTHING! about whats coming in as input signals or going out as Output Activation just needs to be given patterns which training sets provide, so you do not tell it your wearing a hat, but your wearing blue(blue hat what ever) and It notices you always use the blue hat so BAM it will look for the minima of the error function graph I am using closest to blue and if you give it noise( not the same old thing ), then it will learn that even if hungry you still want a blue hat, preventing it from doing the stupid guessing game, and these things are as dumb as can be. Im just saying this cause i expect no one here to understand. Sorry it took me so long to almost release binary *.exe of a glimps of what I am capable of just on a whim without trying. man am i sexy, I would definitely hit that shit. Put my mak down like a g supposed to. just kidding im nothing special just another programming on a mission to get my 17 projects done and move on before I die as i have been in this set of projects for more than 6 years on current version, and 4 years on the previous attempt at a game engine. This project including the neural networks is tagged "Scrut Homebrew Land of Scrut Game Engine v4.12 build 3201 revision 11 " currently or SHLoSGE412.3201.11 for short

If the demo is not out by 12 midnight tomarrow im just spicing it up.

 

[Hammilton]

yeah, this doesn't sound like something I would have come up with when manic at all...

 

[ziddy]

I can't figure out whether this is genius or delusion...either way I suggest TC lay off the meth.

 

[Crack4Lyfe]

Updates? Sounds interesting.