CS 152 NN-23: Generating Sequences: Beam Search

[@TomHutchinson5]

Hi, Neil. I love the clear board setup. Thank you for putting these lectures online!

[@codingblaze4611]

Crystal clear explanation.

[@maqboolurrahimkhan]

very nice explanation such a balance of theory and implementation Thanks

[@NeilRhodesHMC]

Glad you liked it!

[@piyushkumar-wg8cv]

Making small things sound like very big things

[@denizturkmen3521]

i am confused with something: dont we force beam search to select shorter sequences(sentences) in this way since score of a path is calculated by multiplication of probabilities.

[@NeilRhodesHMC]

That is an issue: beam search tends to favor shorter sequences. Solutions include: 1) normalize by dividing by the length of the sequence, 2) Predict the output length of the sequence and then add a penalty based on the difference in predicted vs. actual length, 3) do more complicated normalization.