Machine Learning के मूल सिद्धान्त

Namaste friends,

Is hindi me machine learning ke blog me aapka svaagat hai. Main IIT me PhD student hu aur meri blogs likhne me atyant ruchi hai. Maine recently ek hafta ek paper ko samajhne me lagaaya. Aur tab bhi usme maths ki 2 equations nahi samajh pa rha tha. Tab maine apne ek mitra se vo paper discuss kia aur hindi me use samajhne ka prayaas kia. Main paaya ke mera paper samajhne ka romanch atyant badh gaya aur vo equations jinhe mai ek hafte me nahi samajh pa rha tha unhe maine 2 minute me samajh lia. Maine apne andar introspect kia to jaana ke apni matra bhasha me maine sochna aarambh kia to kuch aisa hua ke main saralta se concepts ko samajh saka evam apne mitra ko bhi samjha saka. Usi saralta ne meri pure research paper ki understanding ko aur deep kar dia aur main un do equations ko asaani se samajh gya.

To mitron yadi apko english, yani ki angreji, aati bhi hai to bhi aap mera blog padhein. Kyunkin hum apni bhasha me machine learning ko bhot saral evam romachak banaane vaale hain. Kyunki vishaya bhaasha se aarambh hua hai, jise angreji me language kehte hain, to aaiye hum log machine learning ke funde sikhne "language modelling" se aarambh karein. Jaisa ki aap jaante hi honge ke machine learning ka aarth hai machine ko anubhav de dekar sikhaana evam dheere dheere behtar banaana. Isme anubhav dene ke anek tareeke hain:-

1. Khud se seekhna (unsupervised learning): jaise ek bacha paani gaaram hai ke thanda, use alag alag taap ke paani ko haath lagaane se seekhta hai. Vaise hi machine ko hum paani ka taap de sakte hain thermometer ke dvara measure karke, kyunki machine bache ki tarah mehsoos nahi kar saakti. Ise angreji me unsupervised learning kehte hain kyunki isme koi bhi supervisor (seekhane vaala) nahi hai.

2. Udaharanon se seekhna (supervised learning): Jaise humne gola evam trikon kaise hote hain aur unme bhed kya hai ise bachpan me seekha, vaise hi hum machine ko alag alaag tarah ke goole evam trikonon ki photos ke udaharan (examples) de de ke sikha sakte hain ke gola kaisa hota hai aur machine kaisi? Ise english me supervised learning kehte hain kyunki isme examples supervisors hain.

Isme "language modelling" jo hai vo dusri shreni me aata hai jise kehte hain supervised learning. Language modelling ka arth saral bhasha me yeh hai ke hum machine ko bhot saara bhaasha ka text dekar ye sikhaayein ke ek text ke sequence ke baad me kya sequence aaega? Jaise maine kha "mera naam rohi" iske baad kya aana chahiye. Ab machine  ko bhot saare text se ye paata lag sakta hai ke "mera naam" ke baad "roha" aae to "n" aana chaiye aur uske baad " hai" aana chahiye. To "language model" ek aisa "machine learning model" hai jisse hum is tarah ke savaal pooch sakte hain:-

Rikt sthan bharein:-

Q) Mera naam Roha_ _____ ?
A) n hai

Sambhavnaayein bataayein:-

QMera naam Rohan _____?
A hai/nahi/tha