Jesse Thomas Palma

I enjoy constructing novel intelligent systems and solving difficult problems.

I have built AI systems for numerous applications: sustainability, weather prediction, supply chain, energy valuation, robotic vision and natural language processing. I bring a diverse set of knowledge that includes transformer models, recurrent neural networks, linguistic theory, dynamical systems, graphical models, predictive models, uncertainty quantification and mathematics of learning in the brain.
My strengths are creativity, persistence and the ability to redesign and combine multiple ML methods to create more effective systems.

I received my PhD with Stephen Grossberg as my advisor at Boston University,
on modulation of reverberant memories in cortical circuits,
explaining mechanisms of attention, learning and adaptive resonance in the brain.

Following this, I worked as a Postdoctoral Research Scientist in Computational Neuroscience at BU
as part of the Neuromorphics Lab and in collaboration with the robotics startup Neurala.

I became a PI in 2013, and created a novel neural system that learns both whole patterns rapidly
(true 1-shot learning without prior exposure to any examples, this is faster than popular "zero shot")
and symbiotically richer orthogonal sub-patterns gradually. The work was patented for robotic vision.


I spend personal time on mathematics, game design, soccer and language studies.
This site highlights and describes some of my journeys on a few themes.

AI & Machine Learning

Fortuitous Beginnings

The founding event of artificial intelligence is widely considered to be the Dartmouth Workshop of 1956.
It was run by Minksy and McCarthy with a small group of notable researchers Shannon, Nash, and Oliver Selfridge.
A year later also at Dartmouth, Stephen Grossberg began his studies applying differential equations to neural networks.

Much later, I was born and grew up around Dartmouth College.
Later still, I had the fortune of discussing my own ideas about neural networks with Oliver Selfridge.
After several meetings, he kindly wrote me a recommendation for graduate school.
And I would do my graduate work with Steve Grossberg.

I did not notice the Dartmouth connection until even later.
In retrospect, I can not help but feel that my involvement in AI started with a bit of fortune.

Oliver Selfridge


Evolutuionary Methods

Upon graduating from high school, I was hired as a software intern in Pascal.
That summer before college, I build a visual tool for building and editing networks of fuzzy logic relations and
a genetic algorithm that could optimize any variable by tuning others that directly and indirectly influenced it.
The work was in conjunction with NIMA (National Imagery and Mapping Agency).
In a summer, I learned a new programming language and
about evolutionary programming, optimization, fuzzy logic, user interfaces and much more.
Seeds were planted.



Evolving Minds

In college, my interest in evolutionary methods persisted, but I also began reading on neural networks.
Intrigued, I bought a newly published book Pulsed Neural Networks by Wolfgang Maass and Christopher Bishop.
I also became connected with Oliver Selfridge then.
Selfridge discussed "learning to learn" amongst other things with me at the time.

Inspired to combine ideas together, I started a project that I titled "Evolving Minds".
The idea was an optimization of learners by evolving neural architectures.
20 years later, this is a growing area of study.
It is now called by several names such as neural architecture optimization or neural architecture search.

For my 2002 project, I built a simulated evolutionary environment with spiking neural network agents.
Virtual agents interacted in a virtual environment, but also adapted and learned from the interactions.
They moved and sensed their surroundings, based on the pulsing of their individual neural networks.
They needed to eat sufficient food to survive, as well as navigate obstacles and avoid poison.
They also needed to find a mate of the opposite gender to pass on their genetic information.
Specified as strings of 0-1 floating points, their genetics determined multiple things:
their neural network topology, initial connectivity and hyperparameters, but also the dynamics of learning.

The agents evolved learn sequential behaviors, reacted to stimuli and become adept at finding food and a mate,
albeit after months of balancing mcuh of the system, such environmental time constants.
I additonally discovered that building good virtual environments requires balancing their dynamics to support the conditions for virtual life.
All build with C++ and OpenGL, the project became my undergraduate thesis.

Pulsed Neural Networks

More details to come...

Math & Games

"When you play a game, you find out
     what it is you should have been thinking about"
                    - John Conway

Mancala

The game of mancala had a significant impact on me as a child.
It is a simple game:
Count out stones around in a circle, and try to get the most in your pot at one end.

Of course, there are a few more rules - but these were a matter of convention.
What intrigued me was how a few different rules led to very different dynamics.
Strategies changed and new patterns of stone distributions formed.

I suddenly wanted to play with a unique set of rules every time.
Around this age, I became interested in a related topic - patterns in numbers,
and my 2nd grade teacher asked that I present to the class my calculation tricks.
Unbeknownst to me then, these events initiated a persistent fascination
with patterns, number theory, game design and their interactions.

Game Fabrication

Expanding from early experience, I have perpetuated a hobby of game creation.
They take many forms: board geometries, card variants, computer games, party activities.


A little description of one project:

Oneiroshah
This project emphasizes duality.
A semi-cooperative semi-competitive game in which participants develop
a cultural narrative through a semi-imaginary human evolution.
It unfolds with an emphasis on both educational and creative processes,
with an unusual board design, novel movement and asynchronous play.