Cities need to cool down. One way is to take advantage of the problem we have with overheated pavement and roads.
If you draw the heat (air suction?) from these surfaces and combine that with a chilled mass. In a ?underground heat storage. Then chill another storage to minus degrees. Using solar energy? By the way. Does the cold storage need to be as big as the heat storage?
I wonder if you then can create electricity from a temperature difference driven process.
Perhaps something like this can take reduce the need for air conditioners. Improve conditions for rain (no overheated surfaces) and more.
I think this denoising is something important. Ran MNIST using denoise_tv_chambolle with only 1000 train samples and got an accuracy of 0.8985. Almost 90% on a very large test set with only 1000 samples.
I think I found somewhat of a cure for my long-term tiredness. Since its the brain that is tired. I tried to remember different foods that would cause another feeling in the brain.
From this remembered that white cabbage was pretty strong. Then I tried the milder point cabbage. The taste I got from this was a milky sweet. This from eating only the white inner parts. Not all point cabbage works. Some organic ones don’t. It all depends on the taste. It should taste sweet not green.
After eating the water rinsed small portion my brain feelt clearer. Not so tired any more.
The process of finding Pi is ?Iterative. So when I was trying to come up with a differential equation for Pi. I wondered. Is there an equation that looked like, dp/dn = g(p,n). Where g(p,n)->0 when n is large. That is. When the iteration has processed a couple of times the update to p is zero. Like a constant.
From this I guessed that since differential equations can be solved iteratively. The iteration part has some physical meaning. Not just some numerical problem.
So I wondered. Could the problem be related to efficiency. I guess is that machine learning could connect error curves in FEA software with real world ?efficiency results.
It takes some effort if you want quality sound in Linux. Setting up JACK, pulse audio sink and Calf type plugins. Then routing the audio through the plugins alll the way to the soundcard.
Therefor I wonder if this can be made into a product using the Raspberry Pi. If the processing power is not enough on the Pi. Than a laptop could perhaps act as an audio streamer. That is. You have a dsp ?server program on the laptop and remotely control it with the raspberry pi over wifi to the amplifier.
If there exists JACK routing over a wifi network? Then all you need is a VNC connection to the PI to solve this project.
Using small fusion devices only. I wonder if its like in machine learning. Where you can get a much better score when iterating the first batch many times. This before continuing to the rest of of training set.
So for fusion this would mean setting up init(). Then hopefully the continued iteration through time will have less errors.
I think this is an axiom. Its hard to fix a bad start. I guess its impossible in most cases.
Why not see if there exist some probability function or if you can go back in iteration to select a new activation function that achieved a much higher loss reduction.
That is. You have many activation functions in model to select from. Not everyone at once. So you have to select one. I guess that you could test among some candidates in som small iteration. After the ?biggest loss reduction is made. Go one big iteration further and repeat the process.
How does machine learning make sense to a battery? With machine learning we can make decision in an automatic way. So I wondered if there should exist batteries with multiple outputs.
Maybe tests should be made with machine learning on a simple galvanic cell first.
The idea with the ice in the water photo is that ice act like a energy absorber. Using naive connection logic this would be connected to the electrodes. So should the electrode surfaces also generated in real battery?
If you withdraw energy from different places in a divided electrode. Would that not change its overall surface look.
With multiple outputs from the electrodes it ?should be possible to use a machine learning decision models for those output options. Like adapted watt per connection usage as function many data.
Here is a free and open source idea for image compression. Using machine learning.
So the idea is to store a discriminator model that can Classify every sub image of the total image. That is. For every random looking input it can through iteration. Help improve the target image simply by calculating a class label number that the subimage belongs to the right class. The classes are numbers corresponding to each subimage. If the class number is 1 for some subimage then the calculated number could be at 0.6 and iterate all the way to 1.0. That is. For a small change in subimage calculation there is a corresponding change in the output score or label number.
That is. The target uncompressed image starts as complete random. It is iterated with random as input with another model_generator until all the classes in the right place is achieved. The generator model is then discarded. The previously learned model_discriminator is the compressed version.
I wonder if its possible to enhance the heat flow out of the motor. Not relying on the natural air flow or shape of the machine elements. It seems one element can just pass its heat to thenext. Possibly degrading the performance?
Just a thought
Inspired by machine learning where you input np.random.rand(1) as weight choice testing numbers. I wondered if you can do the same for a future adapted electrical motor system. Does there exist physical systems who needs noise values? Inbetween electrical elements you can discard the noise with some sort of denoise filter which outputs light.
What if the the gaussian could be described by just a triangle and a smoothing function. Is there a way to use smooth() in math for other problems? Other high order functions could also be used I guess. Why not an image denoiser for iterative integral computing?
Inspired by an old philosophical question. I believe its better to assume something not divisible in the limit. Since you end up with something that still exist.
Then try to divide an odd number with 2 and you end up with a decision. Which side should have one more.
3/2 “equals” (1 and 2) or (2 and 1) compare 1.5*2 with 1+2
1/2 “equals” (0 and 1) or (1 and 0) compare 0.5*2 with 0+1
This can be used in algorithms or math I think.
In machine learning you can use this ?easily. I believe every ratio is a decision number. This means that you also got a decision for derivatives dy/dx.
So for differential equations I wonder if you can have a decision model(data) for when to use the perticular finite difference, f’=(f(x+h)-f(x))/h. Left or Right = model(data). That is. The model outputs wether to take the left or right side of the point. So its a classification problem.
So I called it a decision number because it comes directly from a model worthy decision.
Inspired by machine learning where you have data and target values. I wondered. How do differential equations make sense to a network.
The way I see it. Differential equation solving takes way too few boundary values. Its pretty much just the amount to get the equation = 0. That is. A zero valued difference.
In machine learnig you ?usually have a lot target values.
So a differential equation is just a target value replacement. Missing values ?replaced by an Implicit function. A simplification.
How can this make sense to machine learning?
For machine learnig we could have missing values. Maybe we should treat every target value as a boundary value and have an implicit equation for some ?smart interpolation range.
Another idea would be to do research on how to come up with algorithms. Just use machine learning. To see what can be done and then make a similar ?somewhat human understandable algorithm. This goes for programming code also. Machine learning could give ?hints on design.
I wonder. How to you best combine electric motors. Everything needs to be optimized for life and life can operate at small dimensions. This means efficiency needs to be good at small dimensions. So from this I guess there should exist multi electric motor engines.
What if you have 4+ small electric motors for the same output as a large one. I saw that the sewing machine went from a 90W motor to a low voltage 10W small motor. This for those solar powered machines.
What would it look like? Is a 4 cylinder engine inspirational since its a proven structure?
So I think it could be an idea to create a powerful low voltage electric motor system.
Generator – Basically don’t be a detector. Just write what you feel. Generate. Use your imagination. Write all possibilities you can think of. It does not matter if some judgement feeling got in the way. You can always just rewrite what you have written later.
Detector – Regain some judgement feeling but not too much. Focus on making the text understandable for someone else. Be kind. Rewrite before reduction, rewrite among the sentences of the idea.
So the idea is that. What ever you write. It can be rewritten to make sense to a network.