Skip to main content
6 events
when toggle format what by license comment
Nov 24, 2024 at 5:08 vote accept user81993
Nov 2, 2024 at 19:14 answer added Robert Long timeline score: 3
Mar 23, 2022 at 14:00 comment added amon NNs generally output a vector of floats that can be interpreted as probabilities. You can go further and round them up/down to 0/1 to make decisions, but that's not always appropriate. E.g. for a labelling problem you'd have one output neuron per label, and then might display the labels with the top three probabilities. You associate the output neurons with a particular meaning via training. If neuron #3 means “cat found” and you train the NN on a cat image and the neuron doesn't output 1.0, you'd back-propagate the error in order to update weights. 10k params is large for a toy model though.
Mar 22, 2022 at 10:08 comment added Stack Exchange Broke The Law What is the output supposed to represent? They aren't just numbers that are there for no reason! What is the network trying to learn?
Mar 22, 2022 at 8:52 comment added IMSoP FYI, there is also an Artificial Intelligence Stack Exchange where you might get more in-depth answers (but would need to make sure your question met their standards). A couple of relevant questions from a quick search: ai.stackexchange.com/q/7911/34038 and ai.stackexchange.com/q/28406/34038
Mar 22, 2022 at 1:38 history asked user81993 CC BY-SA 4.0