2017-10-02
3.
Perceptron update rule
- Correct: Do Nothing
- False Positive: w = w-x, theta = theta+1
- False Negative: w = w+x, theta = theta-1
given:
- w = [1,2,3] theta = 1;
- x [1,1,1] false positive example
update weights
(Edited: 2017-10-02) - w = [1,2,3] - [1,1,1] = [0,1,2]
- theta = 1+1 = 2
3.
<u>Perceptron update rule</u>
* Correct: Do Nothing
* False Positive: w = w-x, theta = theta+1
* False Negative: w = w+x, theta = theta-1
'''given:'''
* w = [1,2,3] theta = 1;
* x [1,1,1] false positive example
'''update weights'''
* w = [1,2,3] - [1,1,1] = [0,1,2]
* theta = 1+1 = 2
Winnow update rule
The input [1, 1, 1] is a false positive.
Winnow will update w like this -> for each wi in w = alpha^xi * wi
w = [2^-1 * 1, 2^-1 * 2, 2^-1 * 3] = [0.5, 1, 1.5]
(Edited: 2017-10-02) <u>Winnow update rule</u>
The input [1, 1, 1] is a false positive.
Winnow will update w like this -> for each wi in w = alpha^xi * wi
w = [2^-1 * 1, 2^-1 * 2, 2^-1 * 3] = [0.5, 1, 1.5]
7 solution
7 solution
((resource:WhatsApp Image 2017-10-02 at 5.03.44 PM.jpeg|Resource Description for WhatsApp Image 2017-10-02 at 5.03.44 PM.jpeg))
Solution for Q. 7
Solution for Q. 7((resource:WhatsApp Image 2017-10-02 at 5.03.06 PM.jpeg|Resource Description for WhatsApp Image 2017-10-02 at 5.03.06 PM.jpeg))
Solution for Problem 10. (Anish & Ishan)
S-K Update Rule (Non-Kernel Version):
If `vec{x}'_t in X^+'`,
set `vec{w}_2 := vec{w}_2` and set `vec{w}_1 := (1-q) vec{w}_1 + q vec{x}'_t` where `q = min(1, ((vec{x}'_t -vec{w}_2)\cdot(vec{w}_1 - vec{w}_2))/(||vec{w}_1 - vec{w}_2||^2))`
else:
set `vec{w}_1 := vec{w}_1` and set `vec{w}_2 := (1-q) vec{w}_2 + q vec{x}'_t` where `q = min(1, ((vec{x}'_t -vec{w}_1)\cdot(vec{w}_2 - vec{w}_1))/(||vec{w}_1 - vec{w}_2||^2))`
(Edited: 2017-10-02) Solution for Problem 10. (Anish & Ishan)
S-K Update Rule (Non-Kernel Version):
If @BT@vec{x}'_t in X^+'@BT@,
set @BT@vec{w}_2 := vec{w}_2@BT@ and set @BT@vec{w}_1 := (1-q) vec{w}_1 + q vec{x}'_t@BT@ where @BT@q = min(1, ((vec{x}'_t -vec{w}_2)\cdot(vec{w}_1 - vec{w}_2))/(||vec{w}_1 - vec{w}_2||^2))@BT@
else:
set @BT@vec{w}_1 := vec{w}_1@BT@ and set @BT@vec{w}_2 := (1-q) vec{w}_2 + q vec{x}'_t@BT@ where @BT@q = min(1, ((vec{x}'_t -vec{w}_1)\cdot(vec{w}_2 - vec{w}_1))/(||vec{w}_1 - vec{w}_2||^2))@BT@
Vasudha, Avinash
Vasudha, Avinash
((resource:WhatsApp Image 2017-10-02 at 17.19.28.jpeg|Resource Description for WhatsApp Image 2017-10-02 at 17.19.28.jpeg))
Aarish, Rahul
(Edited: 2017-10-02)
Aarish, Rahul((resource:20171002_171233.jpg|Resource Description for 20171002_171233.jpg))
Solution to Problem 8:
Solution to Problem 8:
((resource:Solution-8.jpeg|Resource Description for Solution-8.jpeg))
Solution to problem 8
Function should return 1 if not all bits are same
Three gates are needed
Gate 1 seperates (0,0,...0) and Gate 2 seperates (1,1,1,...1)
Gate 3 is an AND of Gate 1 and Gate 2
Solution to problem 8
Function should return 1 if not all bits are same
Three gates are needed
Gate 1 seperates (0,0,...0) and Gate 2 seperates (1,1,1,...1)
Gate 3 is an AND of Gate 1 and Gate 2
((resource:Solution-8.jpeg|Resource Description for Solution-8.jpeg))
(c) 2024 Yioop - PHP Search Engine