AI ki Kahani, Part 4

October 2024 mein California ke ek saste hotel mein raat ko phone baja. Sweden se call tha: Geoffrey Hinton ko physics ka Nobel Prize mila tha. Unki kahani wahin se shuru hoti hai jahan Part 3 ruka tha.

Part 4 mein: 1986 mein Rumelhart, Hinton aur Williams ne Nature mein ek paper chhapa. Us tareeqe ka naam tha backpropagation, yaani galti ko peechhe bhejna. Chai se samjho. Chahiye thi 2 chammach cheeni, par pad gayi 5. A ne 4 daali thi aur B ne 1. Galti 3 chammach ki hai, toh A apne hisse ke hisaab se 2.4 kam karega aur B sirf 0.6. Jiska hissa bada, uska sudhaar bhi bada. Network mein bhi yahi hota hai. Aakhri neuron apni galti naapta hai, galti peechhe beech wali layer ko jaati hai, aur har weight apne hisse jitna thoda badalta hai. Ye chakra hazaaron baar chalta hai aur beech wali layer khud seekh jaati hai.

Aaj ye karo: is hafte ki ek galti likho, usse pehle ke 3 kadam likho, aur har kadam ke aage ek chhota sudhaar likho.