Home on Jon Deaton's personal website

Premeditatio malorum and anxiety

Tue, 13 Jan 2026 00:00:00 +0000

Stoic wisdom advocates practice of premeditatio malorum — the continuous visualization of worst-case scenarios. This negative visualization is said to inoculate one’s wellbeing against inevitable misfortune and strengthen gratitude for the everyday. The recommendation rests on the same intuition that derives the famous equality:

Happiness = Reality - Expectations.

The goal of negative visualization is not merely to recalibrate one’s internal assessment of what to expect from the world, but to fundamentally mis-calibrate it such that the formula tends positive. The stoic strives not to maximize the equation, but to guard against it ever falling below zero. In modern terms, its tampering with one’s own internal “dopamine reward prediction error” by hacking the prediction.

Hadamard Conjugation

Sat, 27 Jul 2024 00:00:00 +0000

Hadamard Conjugation

On the wikipedia page for the Hadamard-Walsh Transform, you’ll find a near incomprehensible section the transform’s application in mollecular phylogenetics.

It turns out that whats going on here really isn’t that complicated, but is merely being obfuscated by unnecessary jargon (as biologists tend) like “Klein group”.

The “Klein group” is really “two bit xor”.

01 01 10 10 11 11

11 10 01

In a phylogenetic tree T, there are

Associative Boundary Resetting for SSMs (Mamba)

Wed, 21 Feb 2024 00:00:00 +0000

This post is about a technical detail of boundary resetting for Selective State Space Models (S6) like Mamba. I discuss boundary resetting for Selective SSMs, and give an implementaiton that enables resetting at boundaries without sacrificing the associativity of the binary scan operation, thereby maintaining parallelizability.

This model architecture was introduced by Albert Gu, Tri Dao in the paper “Mamba: Linear-Time Sequence Modeling with Selective State Spaces” and has gained attention as a potential alternative architecture to transformers.

Applying AI to Biological Data: Challenges and Lessons

Thu, 05 Oct 2023 00:00:00 +0000

My work in the past few years has focused on translation of methods underlying the remarkable advances in AI and NLP to biological sequence design problems.

Many of the most successful algorithms in NLP can be translated with little modification to operate on biological sequences rather than natural language. Take for example the influential Evolutionary Scale Model (ESM) which is basically just the BERT masked-token prediction task applied to protein sequences.

Distributed Computing with Custom Beowulf Cluster

Mon, 04 Dec 2017 00:00:00 +0000

Originally published on wordpress

In my last two blog posts, I discussed how I assembled and configured a distributed computing cluster using some old laptops and Raspberry Pis. In this post I will discuss using this cluster for developing distributed computing tasks in C++ using MPI. The distributed application that I developed is a DNA k-mer counter.

Background on DNA Sequence analysis

In biology, the analysis of DNA sequences is critical in understanding biologic systems. Many DNA analysis algorithms focus on identifying genes (the functional units that DNA encodes), however, some DNA analysis algorithms focus on other features of DNA sequences. One alternate approach is analyzing the “k-mer” content of a DNA sequence. K-mers are short sub-sequences of a DNA sequence of length k. Many DNA analysis algorithms make conclusions about biologic systems based on the abundances of each k-mer in the DNA sequence. Other k-mer based metrics include the number of unique k-mers in a DNA sequence and the shape of the distribution of k-mer frequencies. In my undergraduate research, I used the frequencies of k-mers in DNA sequences to identify novel viruses.

Building a Beowulf Cluster from old MacBooks: Part 2

Sun, 08 Oct 2017 00:00:00 +0000

Originally published on wordpress

When we left off at part 1, we had ArchLinux installed on each of the computers in the cluster, and each is also running an SSH server so that is can be accessed remotely. The next step is to set up a shared file system that each node has access to. This will be very helpful later when coordinating distributed tasks, if for instance we are processing a bunch of files and we need each node to have access to any of the files.

Building a Beowulf Cluster from old MacBooks: Part 1

Sun, 01 Oct 2017 00:00:00 +0000

Originally published on wordpress

My family uses their laptops a lot, and they tend to get them replaced after several years. A year ago, I scrounged around our house and rounded up every old computer that I could find with the hope of putting them together to make some kind of super computer.

Five old MacBook Pros in hand (some very old), I did some research and discovered that indeed I could put them to use in what is called a “Beowulf cluster“. A Beowulf cluster is a collection of computers connected through a (hopefully fast) network connection so that they can coordinate parallel and distributed computing tasks. In the next few blog posts, I’ll go into detail about how I turned a these discarded laptops that would have never been used again into a cluster that I use for distributed computing.

Mon, 01 Jan 0001 00:00:00 +0000

Mentorship

Hosted AI resident Kevin Wu at Google X
Hosted AI resident Frances Ding at Google X
Hosted research intern Wei-Hung Weng at Google Research

Honors and Awards

2017 Tau Beta Pi National Engineering Honors Society Member
2015 NCAA First Team Academic All American
2014 NCAA First Team Academic All American
2013 NCAA First Team Academic All American
2013 Lincoln D. Granniss Yale University Scholarship, Dwight Church Award for the Outstanding Senior

Volunteer Work (2015)

I volunteered in a neurology unit at the Stanford Hospital and Clinics where I helped nurses and secretaries accomplish tasks throughout the unit. I also interacted with patients and their friends and family to improve their overall experience at the hospital.

Artistic Gymnastics

Mon, 01 Jan 0001 00:00:00 +0000

USA Junior National Team Member

I was a member of the USA Junior National team from 2008 - 2013. To become a member of the national team I placed within the top seven gymnastics of my age group at national championships. Being on the national team earned me the opportunity to train at week-long training camps at the USA Olympic Training Center in Colorado Springs, CO several times a year.

Competing at the Junior national champonships on Floor and Vault.

Home on Jon Deaton's personal website

*Premeditatio malorum* and anxiety

Hadamard Conjugation

Hadamard Conjugation

Associative Boundary Resetting for SSMs (Mamba)

Applying AI to Biological Data: Challenges and Lessons

Distributed Computing with Custom Beowulf Cluster

Background on DNA Sequence analysis

Building a Beowulf Cluster from old MacBooks: Part 2

Building a Beowulf Cluster from old MacBooks: Part 1

Mentorship

Honors and Awards

Volunteer Work (2015)

Artistic Gymnastics

USA Junior National Team Member

Premeditatio malorum and anxiety