>>22703004The Code/Math descriptor of Evolution:
Observing Evolution of Biota in mammals is what ultimately started the ball rolling on artificial intelligence, machine learning classifiers, and software on a Turing-Tape computer that is able to improve it's own performance on a set of tasks after choosing success criteria and spending energy that allows it to continue to exist and improve over time.
Down close to the metal there isn't much help in understanding AGI/ML or Evolution. Maximally close to the metal you're staring at impulses through a Transistor. A bit higher up and you're looking at addition and multiplication of floating point numbers in 01 binary. A bit higher still and you see matrix multiplication, vectors of squared errors, probability field equations, partial derivatives with respect to directions, hashmaps, data plumbing, tournament brackets. A bit higher you see A-star search, the genetic algorithm, guess and check, Hill-climbing, Mutation autotuning to relative error. A bit higher still and now you're reading notes from eggheads speaking of multiple multi-dimensional surfaces on a shaker table sorted by a cost function. Above that it's a variation on the 16 lines of code:
data = data_load(csv_data_source)
theta = make_a_model(data);
until(convergence criteria){
theta = buff_model(theta);
gradient = accuracy_evaluation(theta);
theta = gradient_ascent(gradient, theta, data);
}
return theta;
def gradient_ascent(g,t,d){
Use g to modify t and test against d
if (improvement){
return t;
}
}
