ZDANOWICZ.DEV▮
ENGINE:HOT --:--:--
NN-31NEURAL LAB — TYPED-ARRAY MLP, ZERO DEPENDENCIES

the same engine that models visitors on the front page — nn.js · flat Float32Array storage, fused backprop, Adam · view-source welcome

ENGINE TEST SUITE

  • …running in your browser

LIVE TRAINING

STEPS 0 LOSS — ACC — NET —

GEN-1 · NEURAL IMAGE FIELD

a coordinate→intensity MLP: feed it (x,y) and it paints that pixel. random weights → a CPPN dream; or train it by gradient descent to become a target — an implicit neural representation, sampled live to pixels and ASCII.

PIXELS · 84²
ASCII · 96×40 

                        

                        

                            LATENT z & COMPLEXITY
                            TRAINING
                            

                                
                                
                                
                                
                            

                            

                                
                            

                            

                                CPPN · random weights, no training — pure structure of the net
                                SGD fitting a target field — REAL GRADIENTS
                            

                            

                                NET —
                                STEPS 0
                                LOSS —