Orbits 1
Falling, falling and falling,
fast enough to
never land on ground,
but still,
caught in gravity’s arms,
forever spinning around in
eternal embrace.
The Body of the Earth and Moon, with their common Center of Gravity, P07 Barycentric balls Figure A1. (2015). In Esa.int. https://www.esa.int/ESA_Multimedia/Images/2015/05/P07_Barycentric_balls_Figure_A1
1 Orbit: falling that never arrives. A satellite is dropping towards the Earth the entire time, but it moves sideways so fast that the ground curves away beneath it, and the fall closes into a circle.
Entropy 2
Life, festering in chaos,
Feeding on borrowed order, 3
Existing as momentary balance,
and Dancing, back into ethereal —
the waltz, alone, that sees no end.
Optimizer 4
I am Adam, 5
Your guide through landscapes not yet defined,
The path towards prediction and knowledge,
where loss functions twist and gradients dance.
But need not be afraid,
for I am Adam,
an optimizer, a guide and a friend
sailing you through mountains of gradient, low and high.
Where SGD 6 turns blind, I —
Adam,
morph and multiply,
descend through the maze,
for the weights of creation, glowing,
at the baculum.
2 Entropy: how far a system’s energy has scattered. In isolated systems, this quantity would not decrease.
3 Borrowed, because it has to come from somewhere. Schrödinger’s what’s life argues that an organism survives by feeding on “negative entropy” – it stays ordered by pulling order out of everything around it, and gives that order back the moment it stops.
4 Optimizer: Training a model means minimizing certain error function, and the optimizer is the algorithm to do that.
5 Adam, short for Adaptive Moment Estimation. It remembers the last few steps it took and lets that memory set how large the next one should be, separately for every weight it is carrying.
6 Stochastic gradient descent (SGD): the older and simpler method. It takes a step of the same fixed size wherever it stands, and it reads the slope from a small sample of the data rather than all of it. So, in a narrow or noisy valley it stumbles.
Bayesian 7
A lost guess in shadows,
Struggling to find the exit yet to arise,
Until new data points the way.
A sleek gambler’s episteme: 8
refining beliefs with what is new, endlessly,
the Game lasting longer than you.
Bayes’ theorem turns a starting belief into a revised one as soon as fresh evidence arrives. Image created by Bowen Li
7 Bayes’ theorem: a formula about how you should update your beliefs as new evidence arrives.
8 Episteme: the Greek word for knowledge you can stand on, as opposed to doxa, which is only opinion. The gambler in this poem is trying to turn one into the other.
