about me

(pictured: some guy)

Hi, I'm Jonathan (Lin). Professionally speaking, I've been interested in the dynamics and applications of computers in all their forms, whether those be traditional von Neumann "grocery list" Turing machines, or alt/indie styles of computing like neuromorphic, deep learning, and reservoir computers. However, I generally take thoroughly unpaid interests in nonlinear dynamics, chaos, and pattern formation - which appear in fractals, swarms, and cellular automata, among other things. Above, I've put links to toys and other things I've made for myself and others, with some hands-on ways to poke and prod at these beautiful things to learn more about them.

I have a B.S. in Computer Science from the University of Maryland, College Park, an MEng. in Bioengineering from the same, and I'm currently working in T-4 at Los Alamos National Laboratory on neuromorphic computing using memristors and other memory materials like photorefractive crystals with Francesco Caravelli. I've also animated a course in complex dynamics with the Santa Fe Institute using Grant Sanderson's Manim library (currently pending publication).

papers and publications

It would be downright generous to call me an early career researcher (at the time of writing, this is my second year doing work with the intent to publish), but I do have a couple papers available on arXiv that are pending review, mainly from my year here at Los Alamos:

• Memristive linear algebra. Jonathan Lin, Frank Barrows, Francesco Caravelli Full abstract: A paper that explores hardware acceleration of matrix inversion, which is a very important operation used to solve systems of equations, optimization problems, and differential equations. The hardware acceleration hinges on the amazing fact that electric circuits can perform matrix-vector operations in one step (as opposed to O(N²) steps on a computer). Memristors are resistors that change their resistance based on their history of electric current flow, which allows for unique nonlinear applications. TL;DR: We use memristors, resistors that change their resistance in response to electric current, in electric circuits to invert matrices and solve problems quickly.
• Uncontrolled learning: co-design of neuromorphic hardware topology for neuromorphic algorithms. Frank Barrows, Jonathan Lin, Francesco Caravelli, Dante R. Chialvo Full abstract: Here, we aim to train a hardware neural network entirely on-chip, using only electric signals applied at the inputs and outputs of the network. This is essentially learning with "minimal control" - by comparison, popular deep learning methods like backpropagation are like reaching into a brain and painstakingly modifying neurons one-by-one. We take advantage of the unique properties of memristors, exploiting the fact that 1) memristors increase their resistance when current is passed through them and 2) that electricity follows the path of least resistance. TL;DR: We train hardware neural networks using simple electric signals only, without complex algorithms like backpropagation.

I also have another paper in progress describing my Master's research, but more on that later.