04:00

This AI produces an octonion-like algebra by preserving all orthogonal projections.

01:10

Producing an octonion-like algebra through repeated polar decompositions

04:29

This AI produces the para-octonions. Here is the spectrum during training.

05:07

This AI produces the octonions. Here is the spectrum during training.

03:42

This AI produces an octonion-like algebra. Here is the spectrum during training.

01:20

Spectra of real and complex LSRDRs of octonion-like algebra during training

01:04

Fitness levels of real LSRDRs of the octonion-like algebra during training

01:15

Fitness levels of complex LSRDRs of the octonion-like algebra during training

03:04

I made a neural network converge near its initialization.

06:00

Catastrophic and complete forgetting in machine learning visualized

02:00

The eigenvectors of complex Hermitian matrices vary smoothly.

02:00

Visualization of the singular value decomposition

00:34

Iterating a quantum channel related function and producing a unique fixed point.

10:00

A reversible cellular automaton plays 2 dimensional Bennett's pebble game on a rectangular grid.

10:00

A reversible cellular automaton plays 2 dimensional Bennett's pebble game on a triangular grid.

23:53

A 1D reversible cellular automaton plays and finishes Charles Bennett's pebble game.

03:00

Iteration of a random bilinear mapping from 3 dimensional space to itself.

01:03

I trained an AI to rediscover an octonion-like algebra

01:19

An AI model that completely forgets its initialization : Training an approximate quantum measurement

01:55

My highly regularized AI model memorizes data but has zero memory of its initialization/training.

06:42

A fitness function for reducing tuples of symmetric matrices to vectors that has few local maxima

02:58

AI finds a 297 element subset of {1,...,34}x{1,...,34} with no row/column 3 arithmetic progression

03:29

Spectra of Jacobians of a neural network with orthogonal weight matrices

03:29

Jacobian matrices of neural network with orthogonal weight matrices

02:47

Singular values of Jacobians of neural network with orthogonal weight matrices

00:47

We find a zero of a neural network vector field on a sphere

00:59

This neural network grew triangular weight matrices while training.

00:48

This quantum channel became the identity mapping after training.

02:01

A Markov chain of unimodular -1,0,1 matrices with -1,0,1 inverses

03:50

Training 5 neural networks to imitate each other

00:42

I trained a neural network to zero out blocks in its weight matrices

01:57

Training a neural network to entangle pairs of neurons: Weight matrices during training

05:22

Evolutionary computation producing new -1,0,1-unimodular matrices from old ones.

00:30

Evolutionary computation producing a -1,0,1-unimodular matrix

01:50

Training a quantum channel to approximate a vector-valued function 5 times

00:43

Training a real completely positive operator to approximate a randomly generated function 10 times.

20:27

Critically simple Laver-like algebras with 3 generators and cardinality at most 80

17:41

Superreduced multigenic Laver tables with 3 generators of size at most 80.

26:08

Critically maximal congruences on superreduced multigenic Laver tables.

01:00

Gradient descent produces a Steiner triple system with 31 elements

01:28

Evolutionary computation produces a de Bruijn sequence of length 64.

00:42

Finding a fixed point of a neural network using gradient descent

52:17

Spectra of completely positive superoperators produced from multigenic Laver tables

01:15

A neural network initialized with orthogonal weight matrices apparently has just one local maximum.

01:50

A randomly initialized neural network apparently has just one local maximum.

02:15

Gradient descent partitions random graphs into few cliques

02:01

Gradient descent finds large cliques in random graphs

01:10

Gradient descent finds a 77 element subset of {1,...,240} with no length 4 arithmetic progression

01:07

Gradient descent finds a 37 element subset of {1,...,240} with no length 3 arithmetic progression

01:01

Using gradient ascent to satisfy a Boolean formula while minimizing the number of true variables

05:34

This neural network quickly forgets what it was previously trained on.

03:13

I made another array where each row and column has the same number of squares of each color.

02:00

I made an array where each row and column has the same number of squares of each color.

10:25

I tried to disprove the existence of a higher infinity and produced these tables instead. Part 3

13:59

I tried to disprove the existence of a higher infinity and produced these tables instead. Part 2.

20:49

I tried to disprove the existence of a higher infinity and produced these tables instead.

21:17

Ways to append an ordinal to a classical Laver table of size at most 128: 2-adic ordering

15:29

Critically simple Laver-like algebras with 2 generators

09:31

All 6854 ways to append an ordinal to a classical Laver table of size at most 64: 2-adic ordering

09:31

All 6854 ways to append an ordinal to a classical Laver table of size at most 64