Lucas Morton s blog
Here s a fun question I had this week: how much energy loss does a vacuum pump incur for a plasma-chemical reactor? I m doing all my calculations in terms of energy consumed in electron-volts per molecule. The reactor operates at 10mbar. I wanted to get a practical value, rather than theoretical, so I looked up a Continue reading Trying ChatGPT on an undergrad physics problem
I finally tested the idea I proposed in my last blog post: penalizing the use of the activation function s nonlinearity, to see if a deep neural network could be compacted into a shallower one. I decided to go with Keras because I heard that it was user-friendly. After reinstalling Anaconda to get the latest everything, Continue reading My first experiment with neural networks
Imagine trying to learn the statistical regularities of video data streams. If we had to search for regularities by considering the probability distribution over every possible subset of pixels their value histories, learning would be prohibitive. Obviously, one would first consider local subsets (adjacent pixels short time histories). However, that still leaves an Continue reading Bivariate is…
TL;DR: it may be important to compact deep neural networks by pushing the nontrivial information processing to occur at the lowest possible layers. Perhaps a regularization penalty could implement this during training. Motivation An obvious next step for AI development is to combine various narrow AI modules (eg, image classifiers, large language models, ) and Continue reading Compacting deep…
When it comes to concepts for nuclear fusion reactors, it seems like all the good ideas were had decades ago. Many are concept dating back to the 1950-1970s era are still being pursued: Z-pinch, mirror, tokamak, spheromak, FRC, stellarator. I ve been wondering if there are any others that were overlooked but might be worth revisiting Continue reading Revisiting the theta-pinch fusion reactor…
Software complexity How does the size of software scale with the number of features? Some folks assume that the scaling should be linear: With his newfound power, he built his operating system on top of his own hardware from scratch in 12K SLOC, with a footprint of 200 kilobytes. For comparison, OSX runs in on Continue reading Software scalings
I ve collected a pile of quotes on risk-tolerance and how best to fund breakthroughs. There s a dominant theme: big-government sponsorship is inimical to the truly-novel research that is necessary to generate breakthroughs. The argument goes like this: Progress in research entails doing truly novel things, which means there are risks. Conversely, this implies that low-risk Continue reading Is…
This is a bit speculative. If you have expertise in this area, I d love to hear your thoughts. The basic idea is that one can get something like intuitionist logic as a meta-logic derived by relaxing from a functional truth-valuation to a relational one. This logic, when extended to probability, might form a useful basis Continue reading Intuitionism, probability, and quantum
IDL is a popular language in the plasma physics astrophysics world. It has a number of pitfalls and poor design choices that lead to nasty pitfalls and gotchas. Here s a list of the ones I ve encountered. Maybe this will save some poor grad student a headache. https://www.harrisgeospatial.com/docs/automatic_compilation_an.html Versions of IDL prior to version 5.0 used Continue reading IDL gotchas…
This is the third post in a series on nuclear fusion. The goal of the series is to address assumptions made in critiques of fusion energy, with an eye toward solutions. Lidsky’s and Jassby’s critiques overlap a bit. Here’s a combined list of the main arguments, broken into 3 categories. In this series, I’ll focus Continue reading Summary of fusion critiques