Beskjeder
Dear all, we hope you all are doing well and that you enjoyed the long weekend.
Here's just a quick reminder that this coming Thursday (May 23) and next Thursday (May 30) we will use the time 215pm-4pm for project work and discussions only. These will be our last two sessions of organized activity this semester.
Don't hesitate to contact us and best wishes to you all,
Keran, Morten and Ruben
Dear all, this is the last week of lectures and due to a Machine Learning workshop starting Tuesday (see https://sites.google.com/view/physmlworkshop24/program if you are interested) we decided to move Tuesday's lecture to Monday at 1015am-12pm.
Same place as before and for those who cannot attend in person, zoom is an option. And the last session will also be recorded.
We will also have lab on Thursday and throughout May for those interested. We will keep our labs at Thursdays.
On Monday the plan is to give a summary of what we have covered this semester. This year we have covered
We have covered
- Discriminative methods
a. With a review of neural networks
b. CNNs and RNNs
...
Dear all, welcome back to our second last session this semester. Note the following: the lecture this Tuesday is digital (zoom) only since I (Morten) am in the US May 5-11 for a short research stay.
Furthermore, since our original time 1015am-12pm means 315am-5am in the night for me, I would like to ask if we can move the live zoom session to 1.15pm-3pm Oslo time. I hope this works, if not, you can always watch the recording which I will upload right afterwards.
The plan for this week is to wrap up our discussions of Generative models. The emphasis is on
Finalizing discussion of Generative Adversarial Networks
Mathematics of diffusion models and selected examples.
Reading on diffusion models.
A central paper is the one by Sohl-Dickstein et al, Deep Unsupervised Learning using Non-e...
Dear all, welcome back to a new week with FYS5429/9429.
This week we continue with our discussion of
Deep generative models
Our focus is on
Summary of Variational Autoencoders
Generative Adversarial Networks (GANs)
Start discussion of diffusion models
For the lecture this week we recommend the following background reading with discussion of codes as well
Reading recommendation: Goodfellow et al, for GANs see sections 20.10-20.11
For codes and background, see Raschka et al, Machine with PyTorch and Scikit-Learn, chapter 17, see https://github.com/rasbt/python-machine-learning-book-3rd-edition/tree/master/ch17 for codes
- ...
Dear all, I just wanted to share some links which could be of interest (and thx to Philip and August for the references here).
1) A school in Germany which could be of interest, see https://www.fz-juelich.de/en/ias/jsc/education/master-students/guest-student-programme
2) A hot software, if you love Pandas, PandasAI is magic, see https://www.pandabi.ai/auth/sign-in. The API is free and it is just cool!
3) From Philip Hoel, this video is very very interesting, extracting physical laws from ML, see
https://www.youtube.com/watch?v=fk2r8y5TfNY (Miles Cranmer - The next great scientific theory is hiding inside a neural network)
4) On combinatorial optimization and graph neural networks (t...
Dear all, this week we continue our discussion of variational autoencoders (VAEs), their mathematics and programming implementations. We will discuss again possible paths for project 2, see for example https://github.com/CompPhysics/AdvancedMachineLearning/blob/main/doc/Projects/2024/Project2/pdf/Project2.pdf
Note that we have moved the deadline to June 7.
Two excellent articles summarize VAEs and diffusion models and our lecture this coming Tuesday follows these articles. They are
Kingma and Welling, An Introduction to Variational Autoencoders, see https://arxiv.org/abs/1906.02691.
- Calvin Luo gives an excellent link between VAEs and diffusion models, see ...
Hi all, we hope you have had a great weekend. Hopefully the weather is also getting better!
This week our plans are to continue our discussions of generative models. We plan on
Finalizing the discussion of Boltzmann machines, with implementations using TensorFlow and Pytorch
Discussion of other energy-based models and Langevin sampling
Variational Autoencoders (VAE) and the mathematics thereof
Generative Adversarial Networks (GANs)
Some reading suggestions.
Reading recommendation: Goodfellow et al, for VAEs and GANs see sections 20.10-20.11
To create Boltzmann machine using Keras, see Babcock and Bali chapter 4, see https://github...
Dear all, welcome back to FYS5429. This week we will try to wrap up our discussions of Boltzmann machines (our first encounter with generative models). Our plans are to discuss
Generative methods: energy models and Boltzmann machines.
Restricted Boltzmann machines, reminder from last week
Reminder on Markov Chain Monte Carlo and Gibbs sampling
Discussions of various Boltzmann machines, RBMs, deep belief networks and deep Boltzmann machines
Implementation of Boltzmann machines using TensorFlow and Pytorch, discussion of codes
Reading recommendation: Goodfellow et al chapters 18.1-18.2, 20.1-20-7; To create Boltzmann machine using Keras, see Babcock and Bali chapter 4, see ...
Dear all, we hope you had an enjoyable Easter break and have been able to properly recharge your batteries.
This week we will start with generative models and the aim is to give you an overview of
1) Generative models in general and mathematics of generative models
2) Our first generative model, the restricted Boltzmann machine (a so-called energy model).
Most of the material we will cover is described by chapter 16 of Goodfellow et al, seehttps://www.deeplearningbook.org/contents/graphical_models.html. If we get time
we will also discuss chapters 18.1 and 182. If you are not familiar with Markov chains and Monte Carlo methods, we recommend reading chapter 17 of the same authors. The lecture notes this week will be upgraded and wil...
Dear all, we hope you all had an enjoyable weekend. And welcome back to FYS5429. This is our last session before the Easter break. From April 2 and till the end of the semester our focus will be on generative models only.
The plans for this week are
- Finalize our discussion on Autoencoders (AEs) and Principal Component Analysis (PCA)
- Implement Autoencoders with TensorFlow/Keras and PyTorch
- Start discussion of generative models and possible paths for project 2 if we get time
The Reading recommendations are
- Goodfellow et al chapter 14 on AEs and chapter 16 for start generative models
- Rashcka et al. Their chapter 17 contains a brief introduction only.
In addition you may find the following links of of interest
- Deep Learning Tutor...
Dear all, welcome back to a new week! We obviously hope that you've passed a great weekend.
This week we will discuss autoencoders (AEs) and this will be our final topic before the Easter break. The motivation behind the discussion of AEs is that they will allow us to catch at least two birds with one stone;
1) link with the principal component analysis and dimensionality reduction methods in general
2) prepare the ground for generative models, which will be the main topic after the Easter break.
We will dedicate the lecture on this coming Tuesday (March 12) and March 19 to the analysis of AEs.
The more detailed plans for this week are
- Discussion of Autoencoders (AEs)
- Links between Principal Component Analysis (PCA) and AE. The lecture material has a larger focus on the PCA, which is an important unsupervised method as well as an important dimensionali...
Dear all, the plans for this week are
RNNs and discussion of Long-Short-Term memory, see slides this week as jupyter-notebook https://github.com/CompPhysics/AdvancedMachineLearning/blob/main/doc/pub/week8/ipynb/week8.ipynb
Discussion of specific examples relevant for project 1, see project from last year by Daniel and Keran. Link here https://github.com/CompPhysics/AdvancedMachineLearning/blob/main/doc/Projects/2023/ProjectExamples/RNNs.pdfLinks to an external site.
Start discussion of Autoencoders (AEs) if we get time.
Reading recommendations
For RNNs see Goodfellow et al chapter 10.
- ...
Dear all, we hope you've gotten time to enjoy the last weekend and look forward to a new week wit machine learning.
The plans this week are
Finalizing discussion of Convolutional Neural Networks (CNNs), with an emphasis on how write your own code. Here, Eric Reber, who developed the code discussed in the lectures from last week, will discuss how he and Greg Kajda chose to develop the code. Eric is presently doing his master thesis at EPFL in Lausanne.
Thereafter we start with our new topic, which is about recurrent neural networks (RNNs)
Reading recommendations:
a. Goodfellow, Bengio and Courville's chapter 10 from Deep Learning, https://www.deeplearningbook.org/
b. ...
Dear all, welcome back to a new week. We hope you all had a great weekend. Here are the plans for this coming week and note that lecture is in person this week (but we provide a zoom option as always):
Plans for the week February 19-23
Mathematics of Convolutional Neural Networks (CNNs), padding, strid, pooling and more
Discussion of codes for CNNs, own code, TensorFlow+Keras and PyTorch examples
The lecture notes (some material will be added about pooling and padding+stride, see also whiteboard notes to come) are at https://github.com/CompPhysics/AdvancedMachineLearning/blob/main/doc/pub/week6/ipynb/week6.ipynb
Reading recommendations
For CNNs, see Goodfellow et al chapter 9. See also chapter 11 and 12 on practicalities and...
Dear all and welcome back to a new week with FYS5429. Before we outline the plans for tomorrow's lecture (Tuesday Feb 13), we would like to
1) remind you that tomorrow's lecture is digital only via zoom, our zoom link is
FYS5429 zoom link https://msu.zoom.us/j/6424997467?pwd=TEhTL0lmTmpGbHlnejZQa1pCdzRKdz09Links to an external site.
Meeting ID: 642 499 7467 Passcode: FYS4411
2) we have found time and an available room for our lab sessions! From this coming Thursday at 2pm-4pm, we have room F?397 available for us. Keran and Ruben will be there. This course being listed as a self-study course was originally not planned to have a lab but seen all the enlisted participant, we have decided to offer a time slot dedicated for lab activities.
...Dear all and welcome back to FYS5429. We hope you've had a great weekend.
Two important messages first. Since I (=Morten) am away in the US for a short research stay Feb 4-14, the lectures tomorrow (Tuesday Feb 6) and next Tuesday (Feb 13) will be via zoom only. The plan this week is to wrap up ( I have been slow here) the discussion on the mathematics of neural networks and links with automatic differentiation. We will discuss code examples as well. The jupyter-notebook at https://github.com/CompPhysics/AdvancedMachineLearning/blob/main/doc/pub/week4/ipynb/week4.ipynb contains most of the information.
There will be some minor code additions during the day today (note all code examples are yet ready).
The hope is to get started with the basics of CNNs as well and use more...
Dear all, we hope you've had a great weekend. Here are our plans for this week
Discussion of possible projects
Review of neural networks and automatic differentiation
Discussion of codes
Please do send to Morten (mhjensen@uio.no) by the end of Monday (January 29) whether you wish to present a tentative project description (1-3 slides, 5-10 mins presentations and discussions) during the first hour.
As of now only two groups have responded that they would like to present their tentative projects. We are planning to discuss various projects and in case you have not yet decided what you want to work on, this may be a good opportunity to get inspired and/or even join a specific group.
Else, much of the material this week is similar to last week and we will discuss and review (for those of you who have written a neural network code) the basics of neural networks, with the back propag...
Dear all, welcome back to FYS5429. This week our plans are as follows:
Mathematics of neural networks (NNs). The slides are not yet finished. There are some missing math. Ready evening of January 22
Writing own code (bring back to life your NN code). If you have written an NN code, please bring it back to life!
Discussion of first data set and paths for project 1
Videos on Neural Networks
Reminder on books with hands-on material and codes
...
Dear all, welcome to a new semester and FYS5429/9429.
Our first session is January 16 at 1015am-12pm, room F?434 at the Department of Physics, UiO.
All lectures will be recorded and the videos will be posted asap online here.
More information will be sent to all of you during the week of January 8-12.
Best wishes to you all and welcome.
Daniel, Keran, Morten, and Ruben