0:00:48 Machine Learning Task
0:02:03 Image Classification
0:07:23 Unsupervised & Supervised Learning
0:10:49 How can we learn to perform image classification
0:18:16 Nearest Neighbor
0:28:50 How to choose Hyperparameters
0:33:14 Cross Validation
0:35:03 Decision Boundaries
0:36:46 Linear Regression
0:41:42 Linear Prediction
0:50:53 How to Obtain the Model
0:52:59 Loss Function
0:55:25 Linear Least Squares
1:01:02 Maximum Likelihood Estimate
1:11:23 Regression vs Classification
1:12:41 Logistic Regression
1:23:22 Why Machine Learning so Cool
1:25:10 Examples of Logistic Regression
idllecture
1:21:24: The binary cross-entropy loss formula is missing a minus and parentheses at the beginning and end. When the "jump" from likelihood function to loss function is done, the minus should be added. It then magically reappears in the cost function at 1:22:07 instead.
There are also some inconsistencies in notation at 1:17:05. While the bottom equation is the likelihood that should be maximized (probability of y given X, theta), the top equation is not the likelihood but instead the probability that the model gives for all examples being of class 1 (probability for y=1 given X, theta). These two equations are not equal. The bottom equation is then correctly used as part of the maximum likeliihood estimation.
sebastianbirk
Simpler explanation. Thank you, Prof. MN.
KERENALLISUDARSHANA
The last part of Motivation is just so exciting, WELL DONE Prof ！
hzc
I have attended a few other courses on ML and AI, but this course is beautifully structured and the concepts are explained really well. 10/10 recommend.
ashwanthramesh
PLEASE provide all your classes online! GREAT LECTURE!!!
seyeeet
Nice lecture! I only disagree with the saturation explanation for the Coronavirus in the end 1:25:13. South Korea plateaued at 10k even though they have 51 millionen citizens, even if we assume a high dark figure the point where it plateaued is nowhere close to saturation. This plateau was more likely formed by the political measures. If a country wouldn't employ any measures the saturation could be observed of course, but only at around 60-80% infection where the reinfection rate sinks below 1.
ralfkinkel
It looks like a one and a half hour lecture. But it's a four and a half hour lecture to me...
jazli
Sort by "Newest First", you can see the time stamp.
idllecture
you're saying "ähm" a lot. But besides that great lecture!
