simulation method
Table of Contents
simulation method is a text material processing method with
- summary
- creating outline with a twist of understanding by simulation
- goal
- extract understanding of ideas from the text material
- definition of understanding
- understanding by simulation
- working concept
1. simple version [use this]
- imagine the whole topic with level 1: input -> output simulation (chopping tree = axe + swing -> tree down + hurray!)
- imagine how the simulation would be carried out step by step with level 2: input -> transformations -> output simulation, in which each step will still be a level 1: input -> output simulation.
- deal with each step seperately with level 2: input -> transformations -> output simulation, which will give you more level 1: input -> output simulations
And when you encounter ideas that are under the same topic but not sequential, like apple and banana are both fruits, just do a simple outline with parallel simulation, like imagining the taxonomy1
2. detailed Process
The process assumes one topic that all materials falls under, or in term of understanding, one genral scenario that we are observing/imagining(like woodcraft for simulations of chopping wood, sawing, glueing, etc.)
It starts by describing the biggest topic with level 1: input -> output, and perform level 2: input -> transformations -> output decomposition on it, which will give you more level 1: input -> output relations, which you can perform level 2: input -> transformations -> output further on
2.1. 1. level 1: input -> output description
How you’d imagine the process in one step.
2.1.1. example
I was looking at this wonderful book on deep learning, and the level 1 description of this book could be deep learning = lots of feature vectors + lots of neural network layers -> accurate function approximator
If I’m looking at means to perform from a list of papers I gathered the that keyword from semantic scholar, it could be LLM data annotation = data + prompt + LLM -> (data,label)
2.1.2. issue: loose field
You will soon realize that the perfect structuredness only exist for very specific concepts, like “residual connection”, which there’s only 1 way to do.
With loose concept such as deep learning, there’s lots of ways that are not really directly connected like “Variational Autoencoder” and “Long-Short-Term Memory network(LSTM)” or have relation that are more complex, or simply not hierarchy, like “Recurrent Neural Network” and “LSTM” 2.
2.2. 2. level 2: input -> transformations -> output
How you’d imagine the process being carried out in steps/stages
2.2.1. detailed instruction
When you have defined the global topic (deep learning = lots of feature vectors + lots of neural network layers -> accurate function approximator), you can then decomposite it with lots of input->output associations bridging from the original input (lots of feature vectors + lots of neural network layers) to the original output (accurate function approximator).
3. comment
This note is too complex. And with lots of text.
Try simplify it, add more visuals and sepereate the tutorials and examples to footnotes and dedicated tutorail zettel, as there seems to be many.
Backlinks
text material processing method
Systematic approach to process text materials such as papers, articles, manuals, documentaitons or books.
- simulation method
- KeshavS : How to Read a Paper - 3-pass method
90 minutes research session with simulation method
(Rationales and preliminaries)
- why 90 minute session
- I can keep my focus for 90 minute top
- research session
- defined here as
- action - reading materials(papers) and forming organization of ideas
- outcome - a subnet of zettelkasten with the ideas and organizations from the sessions readings and thinkings.
- (no term)
- simulation method
injection
(injection is deterministic and easy)
Injection is extracting informations from a given material, like a text book or a blog post, that is already compiled with knowledge; or a bunch of prepared raw data, that you observe to get some insights from.
As all of the materials are at hand, injection is a simple deterministic process of processing it according to some developed mechnism, like simulation method
Footnotes:
“when doing deep learning, there will be an architecture called variational autoencoder, and another architecure called transformer, and another architecture called…, and a concept called gradient decent that is used by many,…”
(LSTM is a more complex and evolved architecture with the same basic idea with Recurrent NEural Network - to have output of last time step feeded into the input layer)