Канал: A neural network programming channel

hand programming mlps!

hand programming mlps!

hand programming mlps!

hand programming mlps!

implementing a cool idea for small binary nn visualizer p3

implementing a cool idea for small binary nn visualizer p3

implementing a cool idea for small binary nn visualizer p2

implementing a cool idea for small binary nn visualizer p2

implementing a cool idea for small binary nn visualizer

implementing a cool idea for small binary nn visualizer

hand setting neural net weights

hand setting neural net weights

decomposing small nns for an hour

decomposing small nns for an hour

prgrm nn maybe

prgrm nn maybe

if you try to run a small neural net backwards by hand

if you try to run a small neural net backwards by hand

making small mlps wo gradscent and then compare with gradscent pt 2 (hot gpt4o action)

making small mlps wo gradscent and then compare with gradscent pt 2 (hot gpt4o action)

making small mlps wo gradscent and then compare with

making small mlps wo gradscent and then compare with

trying to grok 4-bit circuits/mlps

trying to grok 4-bit circuits/mlps

implication structure of ltfs with 0 bias and sorted weights

implication structure of ltfs with 0 bias and sorted weights

implementing new circuits experiment

implementing new circuits experiment

if you have boolean circuit with only majority and inputs, what it is that like

if you have boolean circuit with only majority and inputs, what it is that like

ltfs with 0 bias, positive sorted weights, how different can they be?

ltfs with 0 bias, positive sorted weights, how different can they be?

linear threshold functions are always 37.5% correlated with an input??

linear threshold functions are always 37.5% correlated with an input??

linear threshold functions are always 37.5% correlated with an input??

linear threshold functions are always 37.5% correlated with an input??

linear threshold functions are always 37.5% correlated with an input

linear threshold functions are always 37.5% correlated with an input

mumblecode - largely incoherent debugging and then cool vis of nonlinearities in boolean functions

mumblecode - largely incoherent debugging and then cool vis of nonlinearities in boolean functions

mumblecode - why closest linear functions to random function all flip same input??

mumblecode - why closest linear functions to random function all flip same input??

Nonlinear boolean functions and their distance to the set of linear boolean functions

Nonlinear boolean functions and their distance to the set of linear boolean functions

neural nets HATE this simple 3d geometric transform

neural nets HATE this simple 3d geometric transform

does "separating negation pairs" entirely distinguish linear vs nonlinear boolean functions??

does "separating negation pairs" entirely distinguish linear vs nonlinear boolean functions??

Making a small neural net chase a larger neural net

Making a small neural net chase a larger neural net

hill climbing larger small neural nets not copyable by small nns

hill climbing larger small neural nets not copyable by small nns

visualizing the griddiness of small nn transforms

visualizing the griddiness of small nn transforms

making a js animation of fitting small nn to big nn

making a js animation of fitting small nn to big nn

js/tensorflow.js weight sensitivity visualization implmentation

js/tensorflow.js weight sensitivity visualization implmentation

implementing weight trace visualization for tensorflow.js nn app

implementing weight trace visualization for tensorflow.js nn app

trying to visualize how training points constrain a linear transform

trying to visualize how training points constrain a linear transform

finally finding an unfittable three element training set

finally finding an unfittable three element training set

learning enough css to make my neural net app work; linear transforms

learning enough css to make my neural net app work; linear transforms

human descent - fitting a linear transform manually

human descent - fitting a linear transform manually

javascript programming - setting up app to manually alter neural net weights

javascript programming - setting up app to manually alter neural net weights

drawing lines between input points in different layers

drawing lines between input points in different layers

successfully finding a size 3 training set that linear neural net cannot fit !!

successfully finding a size 3 training set that linear neural net cannot fit !!

trying to figure out when a training set is not a linear transform

trying to figure out when a training set is not a linear transform

playing with small neural net visualization

playing with small neural net visualization

looking for impossible training sets for small neural nets

looking for impossible training sets for small neural nets

playing with a small neural net visualization

playing with a small neural net visualization

livestream using keras' embedding layer for first time

livestream using keras' embedding layer for first time

A small visualization app for neural net generalization

A small visualization app for neural net generalization

Visualizing geometric deformations of very small neural networks

Visualizing geometric deformations of very small neural networks

Backprop to pixels but then train 2nd nn to distinguish real from backpropped images

Backprop to pixels but then train 2nd nn to distinguish real from backpropped images

Backprop to pixels but then re-train to distinguish backpropped pixels

Backprop to pixels but then re-train to distinguish backpropped pixels

making rectangles by backpropagating to pixels

making rectangles by backpropagating to pixels

Guided backprop then noise stability of fake versus real images for mobilenet

Guided backprop then noise stability of fake versus real images for mobilenet

Trying to implement guided backprop and gradcam

Trying to implement guided backprop and gradcam

Can we implement gradcam on fake CLOCK images?

Can we implement gradcam on fake CLOCK images?

not figuring out why random images are classified as CLOCKS by mobilenet, possibly now with saliency

not figuring out why random images are classified as CLOCKS by mobilenet, possibly now with saliency

maximizing image classifier outputs by gradient descent in pixel space

maximizing image classifier outputs by gradient descent in pixel space

linear autoencoders, maybe learning transformer api

linear autoencoders, maybe learning transformer api

trying to get a neural net to learn the identity function (tensorflow.js)

trying to get a neural net to learn the identity function (tensorflow.js)

visualizing an autoencoder except with no compression

visualizing an autoencoder except with no compression

discussing how autoencoders work; new animation training of small autoencoder in js / tfjs

discussing how autoencoders work; new animation training of small autoencoder in js / tfjs

animating training of small autoencoder in js / tfjs

animating training of small autoencoder in js / tfjs

animating training of small autoencoder in js / tfjs

animating training of small autoencoder in js / tfjs

attempting implementation of "generative modeling by estimating gradients" [Song, Ermon 2020]

attempting implementation of "generative modeling by estimating gradients" [Song, Ermon 2020]

generating textures using Gatys 2015 method on mobilenet (possibly) PART 2

generating textures using Gatys 2015 method on mobilenet (possibly) PART 2

generating textures using Gatys 2015 method on mobilenet (possibly)

generating textures using Gatys 2015 method on mobilenet (possibly)

getting mobilenet running in a colab notebook; doing some image classification

getting mobilenet running in a colab notebook; doing some image classification

learning an image upscaling model using a hashtable

learning an image upscaling model using a hashtable

manually implementing a progressive gan but not the actual nn

manually implementing a progressive gan but not the actual nn

playing with image generation using principal component basis

playing with image generation using principal component basis

vis of random walk in input space of nn (js + tfjs)

vis of random walk in input space of nn (js + tfjs)

using broadcasting in tensorflow.js to make a point grid

using broadcasting in tensorflow.js to make a point grid

Preliminary investigations into low frequency of parity functions in neural nets

Preliminary investigations into low frequency of parity functions in neural nets

exploratory - do neural nets just select the most frequent function in the weight space?

exploratory - do neural nets just select the most frequent function in the weight space?

LIVE CODE EDIT ~ NEURAL NET VIZ ~ HANDCODE BACKPROP ~ TO BE CONTINUED

LIVE CODE EDIT ~ NEURAL NET VIZ ~ HANDCODE BACKPROP ~ TO BE CONTINUED

Conceptual Overview of Backpropagation Algorithm without Calculus

Conceptual Overview of Backpropagation Algorithm without Calculus

Just enough linear algebra to achieve the eigendecomposition (Pt 1/4)

Just enough linear algebra to achieve the eigendecomposition (Pt 1/4)

Euclid's algorithm for greatest common divisor (GCD)

Euclid's algorithm for greatest common divisor (GCD)

What is the probability of AND?: Boolean functions and elementary combinatorics

What is the probability of AND?: Boolean functions and elementary combinatorics

Principal Components Analysis (PCA) Tutorial Part 3/3: Symmetric Matrices

Principal Components Analysis (PCA) Tutorial Part 3/3: Symmetric Matrices

Principal Components Analysis (PCA) Tutorial Part 2/3: Projection and Variance

Principal Components Analysis (PCA) Tutorial Part 2/3: Projection and Variance

Principal Components Analysis (PCA) Tutorial Part 1/3: Basis

Principal Components Analysis (PCA) Tutorial Part 1/3: Basis

[Introductory] Using a Chernoff bound to bound number of samples to achieve a good approximation

[Introductory] Using a Chernoff bound to bound number of samples to achieve a good approximation

Randomized Algorithms - Finding median in linear time

Randomized Algorithms - Finding median in linear time

Two Approaches to the Birthday Paradox

Two Approaches to the Birthday Paradox

The Marriage Problem

The Marriage Problem

Randomized versus Deterministic Algorithm for a Simple Algebra Problem

Randomized versus Deterministic Algorithm for a Simple Algebra Problem