Lecture 19

• Lecture 19
  • Video
  • Entropy
  • Local Beam Search
  • Stochastic Beam Search
  • Gradient Descent Algorithm
  • Linear Regression

Video

link

Entropy

• measure of uncertainity

Local Beam Search

• sequential and independent trials = random restarts
• start with k initial seeds x01, x02, .., x0k
• look at neighbors/successors of all these
• if someone is goal, algo is stopped. else select k best successor and repeat

• how diff from k random restarts?
  • independent nahi

Stochastic Beam Search

• local beam search me no randomness
• instead of picking k best from kp, pick k succ at random, with prob of choosing a given succ being increasing fn of its values

Gradient Descent Algorithm

• batch GD
• mini-batch GD
• stochastic GD
• other variants

Linear Regression

• h theta is heuristic fn whihch is approximating fn
• cost is sum of squared erros
• find thetas such that cost is minimized
• updte thetas st you move to minima/maxima
  • the updation is done using following eqns

• these are simultaneous updates
• if m is very large, I need to make many updates to change value, vs when m is smaller