{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/6c3953bdf60a4a9fa93e24ca4b044db5\" frameborder=\"0\" width=\"1110\" height=\"832\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":832,"width":1110,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":832,"thumbnail_width":1110,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/6c3953bdf60a4a9fa93e24ca4b044db5-2163e51d10c9a55d.gif","duration":249.215,"title":"VideoDB Online Hackathon Demo","description":"This Loom explains what gradient descent is and how it trains a neural network. It describes setting up a machine learning task as an optimization problem, using audio inputs mapped to a three-entry output vector for three spoken commands. For each training example, the network output is compared to the ideal output by taking the squared difference and summing to form a cost function, then summing across examples for the overall cost. Gradient descent then repeatedly uses the negative gradient, computed via backpropagation, to take small steps that reduce the cost, with variants like stochastic, adaptive, and momentum gradient descent discussed."}