Math Idea – My Take On The Simple Perceptron In Machine Learning

When looking at the wikipedia article on the Perceptron supervised binary classifier. That one about machine learning. I just realized that the binary classifier defined as.

f (x) = { 1 if dotp(w,x) + b > 0
0 otherwise }

To me this just looks like fractional voting. Summation of a list of for and against values.

From this I can possibly come up with a new guess of a classifier.

My target for this is one that “converges” quickly or not at all

I was thinking that maybe I could have a list of previously calculated values of f (x) as the voting samples. Then use this to ”time” series predict where its going to end up. Either 0 or 1 in the time range you are pretty certain of.

My guess is that this puts a boundary value for the weights to adapt to. Within some “time” period.

This way you should be able to reduce the number of weights you need to calculate in the ?hidden layers. The weights need not converge to a situation where the its unnecessary certain what the outcome will be.

Speculation – Creativity As Set Generation Within A New Group

So an easy way to learn creativity is to learn how to generate information. Information that is recognized to belong to certain group. Like a particular music genre or science field.

I believe the driving force is that the group is uncharted. Meaning that you will have to create the building blocks of the group or learn them.

A somewhat uncharted area is needed because otherwise it would feel like your sampling already created stuff.

An example of a field within programming that generated new creative learners are block programming. Programming using drag and drop blocks like in Scratch.

From this I guess that you can creative new fields by changing the input again. So why not speech programming. Small programs using your voice or maybe a pseudo code trainer. Where you move blocks of sentences to its right row position.

Idea – Verification Hash String In The .CSV Data Files?

Data is important. Its used everywhere. I suggest maybe that therefore .csv files have a hash string to verify the data content. CSV files are just the non binary text files containing the data. So the beginning of the file a hash string could be included.


Here the data is defined beginning at row 3 to the end row. The SHA1 is calculated for the data section only and then compared with row 1 SHA1: which should be the same string.

Idea – Momentum As A Slim Fit Force Field?

My idea is simple.
I wonder. We know force fields create forces on the body like gravity, F = mg. But could a force create a force field? My guess is that momentum defined as p = m*v. Is just the results of a summation of forces from a force field acting on the body, m.

The force field I’m guessing is a slim fit one. Acting only on the body. So for the body not to accelerate the field has to balance itself out after a some time. Storing energy as kinetic potential.

Maybe its just a bunch of balancing forces acting on atom size pieces of the body. Like dF -> dm <- dF throughout the body.

// Per Lindholm

Speculation – Decision Making

In a previous blog post I figured out that if you want to mimic a random function. You can input samples of previously computed random data.

It was much easier this way than to let the function store all the information about the random behavior. To clarify. The target data was a lot of random numbers and the function was going to adapt to it.

I’m pretty sure evolution doesn’t miss such an easy method. For instance in decision making I speculate that if there is a majority decision then it doesn’t want to end up with an equal amount for each choice. Maybe sampling of random clusters could help.

So my idea is that this could be avoided and at the same time getting results fast. By using the predicted results of enough samples as the voting basis. This way a decision can be made any time. Although as time progresses the decision gets more accurate. Then the score of could have done otherwise then gets lower. Meaning your less likely to think your decision was flawed in any way.

Then if two choices converge at 50 % at decision time you just predict the curve some time ahead and choose the first winning choice.


Math Idea – Root Problem Description By Similarity Of Problems

I was wondering some time ago. Thinking about real problems. What if you could describe problems with just ones and zeros. What would it look like.

I believe the answer is simple. In linear algebra you can solve a set of equations with matrices and vectors. For instance Ax = b. This describes a problem with numbers.

So from this I wonder if you could get some useful information from taking the same high level problem. Simplify it and find a relation between the two ”different” problem descriptions.

So here you have A1x1 = b1 and A2x2 = b2. They are different but both come from the same high level problem. Here A2x2 = b2 could be a simplified approximation of the problem.

I speculate that if you could find the similarity of the problem descriptions. Maybe you could go back from the best approximation and iterate a better description and solution. The A0x0 = b0 description if you will. From this root description a better solution is easy to find.

Math Idea – Could You Use Vector Input For Integrals, Limits or System Of Equations?

For hard to calculate integrals, limits or system of equations I wonder if you could use a vector instead of single variable as input. What I mean is that the algorithm would handle and make use of 3 or more values at a time.

For instance I wondered if maybe one can use future value prediction from a series of previous results. You can calculate a predicted future value x(n) and x(n+1) with a linear model. Where x(n) = sum(a(i)*x(n-i),{i=1..p}).

This guessed value could perhaps be used somehow.

Idea – Should There Exist 3D Clustering ?

My idea is simple.

Thinking about the universe. How it all started. Getting something from nothing.

Taking inspiration from machine learning I recognize the problem and solution as some type of unsupervised learning.

So I wonder. Should there exist 3D clustering. Where objects group together in space. Forming basic structures.

So if you are unsure how something started. Start with random. Test how random data could self assemble or cluster into 3D models.

Speculation – Machine Learning Ideas

Here are some small ideas. Some guessing on how to improve machine learning.

Its somewhat hard to make a function that creates noise. Therefore I wonder if you can improve it by letting the machine learning function just sample some already created noise. That is. You add a noise sample vector to the input vectors. Making the vector size larger.

The idea is that if the output data has data that looks random. Then it could be an idea to input some similar random looking data. For the transfer function to sample from. I mean.

Below is a test

linearinput samplenoise

Game Console Idea – User Generated Games On Physical Storage Casettes (Nintendo)

From the success of the raspberry pi we know that maker activety and creation is very important to people. There should not only be consumation of the products but also the possibility to change. To create. Or to learn to imagine.

The idea is to help gamers become involved in the creation process. One way that could happen is by letting them create their own games. With a game engine (gui). For that achievement factor the games could then be stored on Nintendo type cassettes. They would actually just be memory cards. But having a big container with printed images will make it much more fun.

So the new type of game console is one that lets users create their own games and ”print” them on cassettes. Also being able to put the physical device together is important.

Speculation – Could A CPU Help The Cells Heal The Body More Accurate?

The idea is simple.

I wonder. Using machine learning. Could a CPU help the cells in their decision making? I mean. If you get an injury. The cells will heal the body. I speculate that the decisions the cells make can be improved with knowledge of what the body looked like in detail from before the accident.

It would be good if this could help burn victims.

I got the idea from looking at a scar. To me it looked like some kind of approximation. From that its easy to guess that cells need calculations.

Idea – The Smart Cost Function ?

The idea is simple.

In machine learning you need to calculate the error depending on some parameters. To be able to optimize them.

Some simple cost function is the Mean Squared Error.

My idea is to see if one can use “similarity” of two objects instead. Like you visually compare two images and set a value on how similar they are. Here the objects could be two vectors.

One example could be two audio files that sound the same but are not matched sample for sample.

This could be the basis of a smart cost function. That is. You use machine learning to produce a cost function that outputs the similarity value.


Idea – Sample Generator

I had an idea today to create a matching value with some machine learning. Here I wanted a value for similarity. For parts of a rendered image. With that of the same part of a higher detailed image. The idea was to speedup blender.

I’m not finished yet. Actually I’m not the disciplined programmer I want to be. However I came up with an idea that I want to share.

I had a target vector with little variation. Actually many zeros and a single one value. I don’t know if my idea works but what I figured out was that you could transform a single value to an object of many values and then use statistical methods to get the target value back.

So if you have a target vector with 5 elements. With zeros as values. Then you could see what happens when you insert 100 random samples with a mean value of zero as a replacement object. Then for each of those zeros you would suddenly have 100 values. Giving you the matrix size of 5×100. However taking the mean value of each of the rows gives you back the original value.

So the idea is to train the machine learning setup with more values then get the wanted value by taking the mean of the output.

Depending on factors like the amplitude of the random values. This would effect the outcome of the training.


An example of a universal sample generator. The sun.

Idea – Could Machine Learning Benefit Stroke Victims

The idea is simple.

I wonder if stroke victims who have trouble with speech could benefit from machine learning.

The idea I had was that the software would help with the understanding. I mean we have software that can translate between different spoken languages.

Here the machine learning software would learn from the person. Perhaps even better than a human could.

Maybe the person would not need to adapt so much since the software doesn’t care what it sounds like.
It could use variations of words the person can produce.

Idea – Speed Up For Blender 3D?

I made some time series predictions last week. From this exercise I got the idea. That maybe one could make predictions for 2D images and possibly also for 3D objects.

For 2D images I imagine that if the image had some holes. Not yet calculated values. Then they could be predicted instead of just taking a heavy load on the CPU. This from the surrounding points.

For 3D objects I think one could do the same. If you got some unsure vertexes then just let the algorithm calculate where those should be positioned at. In a similar fashion from the surrounding vertexes and maybe by letting the algorithm know the type of object.

In both cases the function could be named ”smart fill”.

Math Idea – Inverse As The Single Value Find

Not all functions have an inverse function. There are many functions that output the same value for different input values.

However I had an idea that if you rewrite the inverse as just a find function. Then why not use several values to pinpoint that single input value. I mean. This would make it more likely to find that single value.

Say we have a function f(x) = sin(x) + sin(2x) and we want to find somekind of inverse to it. I imagine that if you have 3 close points from that function you could map or find what the input values were that produced those three numbers.

simple enough … with machine learning I think you would only need to specify the input matrix with say 5 or more function values, f(x) per row and the corresponding target values, x.



On the left you can see the trained find inverse function which had a little problem with values around zero. It used ten values. Five of those were the derivative.

Livelihood Idea – Wiki Tutorials

As a way to improve your skills and earn ?some money. I thought. Why not create tutorials from the very large collection of wikipedia articles. I mean there should be plenty of stuff that needs video explanations.

You could just split the window while taking a screencast. With the article on one side and the handwritten text on the other.

With these the young people of the world could educate themselves. By seeing and above all from the creative process.

Livelihood Idea – Wiki Tutorials