Artificial Neural Networks

Q1. Visual recognition is inherently challenging for a computer. Enlist at least three

challenges.

a) Viewpoint variation: the same object looks completoly dorent in picel terms when

Viewed from different angles, so a computer must recognize the object regardless of

crontation.

b) Blumination variation: changes in lighting conditions drant cally alter pixel intensites,

making the same object look very different in bright vs. dim conditions.

e) Occlusion: objects are often partially blocked by other objects in a scene, so only part of

the coject may be visible.

d) Scale variation: objeds can appear at very diferent sizes depending on distance from the

camera

e) Background clutter and intra-class variation: cojects of the same class can vary hugely

in appearance (eg. different beseds/colours of dogs), and busy backgrounds make sepanting

the objed from its survoundings difficult

(Any three of the above are sufficient as a valid arswer.)

02. Explain the concept of backpropagation in the context of fraining an ANN.

Backpropagation (backwed propagatice of enors) is the algorthes used to fraie artificial

neural netwolks by effoery computing the gradient of the loss fundicn with respeto every

weight in the network, so those weights can be updated to raduce error.

How it works:

4) Forward pass — an input is passed through the network layer by layer to

foul prodiction: f0 the lass is comouted by comparing this pesdiction to the trus