Deep Learning Lecture 5: Regularization, model complexity and data complexity (part 2)

Показать описание

Course taught in 2015 at the University of Oxford by Nando de Freitas with great help from Brendan Shillingford.

Рекомендации по теме

Комментарии

Was reading about how Coursera and other portals carry out significant data analysis on how to make lecture videos more engaging - having the right background colors, etc. etc.
They should just look at Professor Nando de Freitas. With his outstanding lecture series, he shows how its done and in the end what really matters - his teaching style/method, and passion for the subject, and striking a really good balance between mathematical concepts and programming. As he is teaching in this lecture - he neither "underfits" math for a machine learning course nor does he "overfit"...
"Simple" slides, extremely knowledgable teacher, a very humble teaching style, accessible, explaining history behind certain concepts, mentioning advanced concepts he will not cover but giving enough pointers to go chase them down on our own!...

Some of the best lecture series like - EE263 by Stephen Boyd, Statistics 110 by Joe Blitzstein and MIT 18.06 by Prof. Gilbert Strang all share similar characteristics.

sbansal

Hello Professor I have followed all your courses CPSC 340, CPSC 540 and now this. I will say that you have helped me a lot in strengthening my understanding of Machine Learning and I really enjoy watching your lectures.

gauravpathak

hello

at 35:29 the matrix xTrain has been accessed by the notation xTrain[{{i}}] . any reason for the extra { { } } ? I did run the code and the notation given in lecture slides does work, but whats the reason?

roar

Hm, I actually chose detla = 1, because both sides were most equal. I expected them to be roughly equal if the errors were only coming from the actual data noise and not from systematic model errors..

GuillermoValleCosmos

Shouldn't the MSE on the training set always decreases when delta decreases (i.e. when the constraint on theta loosens and we get closer to the MLE)?

LionelJourdain

at 43:15 doesn't this depend on the train/test split? If the split is 90/10 that would change your decision from a 50/50 split no?

mrrodbruno

Definitely an excellent series so far! First lecturer I've had to recommend reading the wiki articles and not be so dismissive of easy information. These lectures are definitely easy to follow as they appeal to all backgrounds in Mathematics, Statistics, Computer Science. By the way, does anyone know if it is possible to run torch on OSX 10.10? So far Yosemite has done me no favours.

harryritchie

In the plot of MSE to the size of training set, how did we pick the baseline of the error to be 4 (the black line)? Is that picked arbitrarily?

Raj-zpiw

i have the matlab code look in matlab site file exchange "deep learning lecture"

michaelscheinfeild

Deep Learning Lecture 5: Regularization, model complexity and data complexity (part 2)

Deep Learning Lecture 5: Regularization, model complexity and data complexity (part 2)

Deep Learning | S23 | Lecture 5: Regularization in Deep Learning, Convolutional Neural Network

Deep Learning - Lecture 5.1 (Regularization: Parameter Penalties)

Deep Learning: Regularization - Part 5

[MXDL-5-01] Regularization [1/2] - Weights and Biases Regularization

Lecture 5 - Regularization

Regularization (C2W1L04)

Deep Learning - Lecture 5.3 (Regularization: Ensemble Methods)

ResNet Explained - Vanishing Gradients, Skip Connections, and Code Implementation | Computer Vision

Regularization of Deep Learning | Lecture 7 | Deep Learning

Regularization in a Neural Network | Dealing with overfitting

Lecture 5 - Recurrent Neural Networks | Deep Learning on Computational Accelerators

Deep Learning Lecture 4: Regularization, model complexity and data complexity (part 1)

Lecture 5 - Regularization, Early Stopping, Dropout.

Deep Learning - Keras - Multi Layer Perceptron Part 5 Regularization

Regularization in Machine and Deep Learning (Ridge, Lasso, and Layer weight regularizers)

Lecture 9 - Normalization and Regularization

Lecture 7 - Regularization for deep learning

Regularization

Overfitting and Regularization For Deep Learning | Two Minute Papers #56

Deep Learning: Regularization - Part 5 (WS 20/21)

Regularization: Deep Learning and Arificial Neural Networks 1.6

Lecture 8: Training Neural Networks: Normalization, Regularization, etc

2022 ML100: Lecture 5 - Regularization