Category: Uncategorized

VAST DATA. "The factors very quickly become too numerous for humans to keep track of. Not a breakdown of logic per se, but the effects of sloppy shortcuts, bad handwriting, and immense volumes of code.” https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

LIMITED BY VISUALIZATION. "Sometimes with neural networks on images you can display the weights in the form of an image and see it select certain parts of the image but beyond that we don't have good ways of finding out what the weights mean." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

NEURAL NETWORKS. "These weights get refined by training algorithms. The classic being back propagation. You hand the network an input chunk of data along with what the expected output is. Then it tweaks all the weights in the network. Little by little the network begins to approximate whatever it is you're training it for." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

NODES AND WEIGHT. "Neural networks work as a bunch of nodes (neurons) hooked together by weighted connections. Weighted just means that the output of one node gets multiplied by that weight before input to the node on the other side of the connection. These weights are what makes the network learn things." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

"NEURAL NETWORKS are called black boxes because the training algorithm finds the optimal node values to solve a problem, but looking at the solution it is impossible to tell why that solution works without decomposing every element of the network. In other words, the node values are extremely sensitive to the context (nodes they connect to), so you have to map out the entire thing to understand it." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

BLACK BOX. -- "The neural network actually beats the level instead of pausing the game."" "-- "Yes but neural network heuristics are black magic that I will never understand." " "-- "Funny you say that, because the values of the nodes are generally considers to be a black box. Humans cannot understand the reason behind the node values. Just that (for a well-trained network) they work." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

CODER BACKCHAT. -- "The evolution mechanism decided that the best way to win was to not play/fight. Evolution. Nice."" "-- "A strange game. The only winning move is to not play."" "-- "Goddamnit, I'd piss on a spark plug if I thought it'd do any good." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

MACHINE SMARTS. "Someone programmed a software that could play super mario [edit:] Tetris, and the goal was to stay alive as long as possible. sometime along the road the program figured out that pressing pause had the best results and stuck with it. goddamn thing figured out best way to win the game was not to play it." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

AI TEMPER TANTRUM. "Oh man. We've done it. We've finally forced an AI to ragequit." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn

GENETIC PROGRAMMING. "My 'ant colony' produced ants that got progressively better at finishing the first 80% of the maze, but the maze got more difficult towards the end, and as things got more difficult, they got stuck - so they would get faster and faster at getting 80% of the way there and then, unable to figure out the next bit, just hide to maximize the 'points' my system would grant them, and their chances of survival - how awesome is that! - my own (extremely basic) computer system outsmarted me." https://www.reddit.com/r/todayilearned/comments/3d3vct View in LinkedIn