No final workshop
 

Dear All,

with too many of you being stressed with exams and so forth, we ended up with dropping the workshop. The unfortunate thing this year is that many exams extended (from Monday Nov 26) into a regular teaching week. Having only less than half of you being able to attend the workshop is also not ideal.

However, with your consent, may be we could share a common link with the final projects (but only if you agree to do so). 

So, sorry for the late note, I was frantically hoping we could run the workshop, but seen the boundary conditions, we thought it is better to skip it. 

Anyway, best wishes to everybody and thx again for a fantastic semester.

Best wishes with your exams and project 3 (deadline Dec 17)

Bendik, Kristine, Morten and ?yvind

Weekly update FYS-STK3155/4155

Dear All, this is sadly our last week of lectures. It has been incredibly fun. And the amount of labor you all have put in your projects is amazing. Thx a million for all these heroic efforts. we hope to be able to come with feedback to project 2 by the end of this week. The deadline for project 3 is now December 17. Else, this week the lab on Wednesday runs as normal and the last lecture with the final discussion of Boltzmann machines as well as a summary of the course plus final discussion of project 3 is on Thursday the 29th. There is no lecture on November 30. 

Furthermore, as you can see from the poll here, I am still trying to organize the final workshop. Based on your inputs it seems that it is only December the 11 which fits. 

For the workshop, the idea is that every group makes a 5 min presentation. There are so many exciting proposals for project 3...

Plans for week 47

Dear all, this week we will start with our last topic, a gentle introduction to unsupervised learning. Our focus will be on Boltzmann machines, in particular so-called restricted Boltzmann machines. These have received quite some interest lately, from imaging to the solution of Schroedinger's equation in quantum mechanics, see for example Carleo's article in Science last year, http://science.sciencemag.org/content/355/6325/602

It will also allow us to continue partly with the probabilistic themes discussed last week and bring up what is called MCMC simulations, Markov Chain Monte Carlo simulations. So, in addition to discussing the theory behind Boltzmann machines  (with links to the Hopfield and Ising models as well), we will also discuss the Metropolis and Gibbs sampling approach, all highly relevant for Bayesian theory as well.

This will end the topics covered this semester.  We e...

Dear All, 

a brief update on the plans for this week and project 3. First thx to everybody for proposing so many great variants of project 3. We really look forward to see your results.

Project 3 is now set up, see for example https://compphysics.github.io/MachineLearning/doc/Projects/2018/Project3/html/Project3-bs.html (or do a git pull or scroll down https://compphysics.github.io/MachineLearning/doc/web/course.html

It has three variants, one where you define the data, two where you study the credit card data and three to study the solution of the diffusion equation using neural networks. 

We have set the deadline to December 14 and that is because I am still hoping we can run the...

Good morning everybody!  I hope all is well!

This week we continue our discussion of decision trees and random forests, with a repetition of support vector machines well. 

Decision trees are discussed in Hastie et al in chapters 9 and 10 (the discussion is somewhat scattered) and Geron chapters 6 and 7. 

This will more or less conclude our part on supervised learning from a frequentist's approach.

Next week Nils Hjort (not a minimization of my last name) will teach about Bayesian statistics.

Nils is a great teacher and I hope you'll enjoy his lectures. He is one of the leading experts on Bayesian statistics.  The department of mathematics offers an advanced course on Bayesian statistics for those of you who might be interested, see /studier/emner/matnat/math/STK4021/index-eng.htm...

Weekly update for FYS-STK3155/4155

Good afternoon to everybody.

Here follows our weekly digest and plans for machine learning.

A note on classroom tomorrow Thursday November 1. Due to the  high-school teachers' conference, we have to change auditorium on Thursday. We will be in auditorium 2 in the chemistry department, see /studier/emner/matnat/fys/FYS-STK3155/h18/timeplan/index.html#FOR

On Friday we are back to our normal auditorium.

Last week we discussed how to reduce the dimensionality of our data set using fex the principal component analysis. We also started discussing support vector machines. We will conclude this part tomorrow, with code examples etc. Geron's text and its chapter 5 gives a good survey. C...

Weekly plans for FYS-STK3155/4155

Dear All,

here comes the weekly update with plans etc. 

Last week we finalized neural part for now, ending with a more superficial discussion of Convolutional NNs. This is an important field, in particular in imaging. For those of you interested, there is a dedicated course (spring, IN5400) on imaging and machine learning taught by Anne Solberg. It is highly recommended. We limited ourselves to an overarching view and ow to use 

CNNs with tensorflow and keras. 

This week we will discuss other methods for classification in particular, but we will also discuss some famous methods for dimensionality reduction. The plan for the rest of the semester looks like as follows

1) This week (week 43):

     Thursday: Dimensionality red...

Weekly update FYS-STK3155/4155

Hi all, again, we hope this week has started the best possible way for everybody.

A longer mail this time, sorry for that.

1) There is a new version of project 2, now with the accuracy function only for the logistic regression part (part c) and for the neural network classification part (part e). 

2) I have updated typos in equations in the neural network part and added expressions for the derivatives of the cost function for the cross-entropy version. Please let me know if you spot typos! I appreciate all feedback here. Don't hesitate to mail me or use piazza if you find topics which are not clearly spelled out.

3) For project 2, you may wish to try other activation functions than...

Change in deadline and weekly update for FYS-ST3155/4155

Dear All, first, thx a million for heroic efforts with project 1. We really appreciate this and plan to come with feedback by the end of next week or at latest the beginning of the week thereafter. 

Since some of you experienced problems yesterday evening with handing in, we extended the deadline to today at 23.59 (lack of registration plus minor technical problems). If you have already handed in and wish to perfect your delivery, please feel free to do so. So, the deadline is today at 23.59.

Else, project 2 will be available by the end of the day and deals with neural networks (building a code for an MLP) and compari...

Weeekly FYS-STK3155/4155 update

Dear All, here comes (delayed) the weekly update with plans for the week.

As you may have seen, the project deadline has been extended by a week, with deadline October 8. This means that today's lab session will focus only on project 1. Project 2 will be available next Tuesday and will focus on combining classification and regression using what you have developed in project 1 as well as logistic regression and neural networks.

This leads us to this week's lecture plan:

1) Thursday we will derive the back propagation algorithm and discuss how to build our code for a multilayer perceptron model for deep learning. We will also discuss coding of the algorithm. This is a central algorithm in neural networks. You can find all the...

Hi all,

here follows our weekly update fr FYS-STK3155/4155. Last week

we went through logistic regression (chapter 4.4-4.5 of Hastie et al.)

and since our cost function is defined by a log function that depends

on the parameters of the model ¦Â, we need to find the minima

numerically. We started thus discussing various gradient methods and

will continue this week with stochastic gradient methods.  Much of

this material is not well covered by the book of Hastie et al and thus

most of the material will be covered by the lectures slides at

https://compphysics.github.io/MachineLearning/doc/pub/Splines/html/Splines-bs.html

Also, the text of Murphy at

https://github.com/CompPhysics/MachineLearning/blob/master/doc/Textbooks/MachineLearningMurphy.pdf

contains a better discussion in chapter 8.3...

FYS-STK3155/4155 weekly update

Hi all,

here follows the weekly digest for FYS-STK3155/4155.

Last week we ended (for now) our discussion of linear regression and resampling techniques. Most of this material is covered by chapters 3 and 7 of the text of Hastie et al and the lecture slides on regression. Project 1 addresses many of these aspects, in particular also resampling techniques like bootstrap and cross-validation. An analysis of the bias and the variance as discussed in chapter 7.3 of Hastie et al, https://www.springer.com/gp/book/9780387848570

This week we will continue our discussion of Logistic regression for classification probl...

Project 1 is now ready

Dear all, the first iteration of project 1 is now available. A git pull should give you everything new.

You can find the project at the link https://compphysics.github.io/MachineLearning/doc/web/course.html and scroll down to project 1. You will find a pdf version, two html versions and a latex file for the project . The deadline is October 1.

Feel free to make groups and collaborate. You can also find the files at 

https://github.com/CompPhysics/MachineLearning/tree/master/doc/Projects/2018 and in the Project1 folder you wi...

Weekly update FYS-STK3155/4155

Good morning to everybody, we hope you had a great weekend! Welcome to a new week. 

Here comes the weekly update with ditto plans for the week.

Last week we ended with a discussion of resampling techniques, in particular applied to the training of a model and its pertinent testing/validation. Most of this material is covered by chapter 7 of the text of Hastie et al.

In particular we discussed so-called cross-validation methods and started discussing the popular bootstrap method, as well as mentioning the Jackknife and blocking methods. We will finalize the discussion of these methods on Thursday and start thereafter with a discussion of classification methods. This will lead us to linear methods for classification and what is called logistic regression. The material is covered by chapter 4 of Hastie et al, and we will essentially discuss chapters 4.1-4.4....

Dear all, the following article, see https://arxiv.org/abs/1509.09169,

contains a set of excellent lectures about the Ridge and the Lasso regression methods, including the calculation of various expectation values and much more. Recommended.

Many of the details omitted in the text of Hastie et al (and other texts as well) can be found here. 

Best wishes to you all,

Morten

Dear All, the first lecture is Thursday 23, 1215pm in Store Fysiske Auditorium, Department of Physics.

Lectures are 

Thursdays 1215pm-2pm

Fridays 1215pm-2pm

Place is Store Fysiske Auditorium

The first lecture is Thursday August 23. The last lecture is Friday November 30. The lab sessions (we will split in groups afterwards). The first lab session is Wednesday August 29.  The lab is room FV329, Department of Physics.

Depending on how many register for the course, we may divide the lab session into four groups as follows:

Group 1: Wednesdays 10am-12pm

Group 2: Wednesdays 12pm-2pm

Group 3: Wednesdays 2pm-4pm

Group 4: Wednesdays 4pm-6pm

All course material is available at the github address https://github.com/CompPhysics/MachineLearning

More details about practicalities will be discussed during the lectures and the lab sessions. Welcome to everybody!