วีดีโอ

Je suis confiant quant à votre projet, c'est pourquoi j'ai investi mon argent dans votre start up dès son IPO.

How to train it on GPU?

Great video!

Passionnant ! and great. I shared this video on LinkedIn.

AI462 neural networks has some examples of RP associative memory.

A LSH with +1,-1 outputs dot product with weight vector gives an associative memory. To train, get the recall error, divide by the dimension of the weight vector, then add or subtract that value to each weight according to the corresponding LSH +1,-1 output value, to make the error zero. LSH in your case would be a Random Projection followed by +1,-1 binarization. You need to know the variance equation for linear combinations of random variables applies to dot products. Also the Central Limit Theorm applies. And some other things. Eg. If you want the value 1 out of a dot product( which has equal noise across all its inputs) you can make one input 1 and one weight in a weight vector 1. That cuts out most of the noise. You can make all the inputs 1 and all the weights 1/d (d=dimension of the dot product). That averages out the noise. Averaging is better. In both cases the angle between the input vector and the weight vector is zero. As you increase the angle toward 90 degrees the length of the weight vector must increase to keep the output of the dot product at 1. And the output gets more and more noisy.