Master-Seminar - Deep Learning in Physics (IN2107, IN0014)

Prof. Dr. Nils Thuerey

Time, Place: Mondays, 12:00-14:00 in 02.13.010
Kick-OFF: via BBB
https://bbb.rbg.tum.de/nil-djw-hjw
Date: January 31., 2022, time: 15:00
Begin: Monday, April 25., 2022 (online recording)
Details: Takes place online via BBB - Only online recordings until on-campus lectures possible again
Prerequisites: Introduction to Informatics I, Analysis, Linear Algebra, Game Physics and Introduction to Deep Learning recommended
Registration: Please register via the matching system

Content

Using deep learning methods for physical problems is a very quickly developing area of research. The research group of Prof. Thuerey has studied learning-based methods for Navier-Stokes problems and fluid flow applications in recent years, examples of which include learning latent-spaces for physical predictions, generative adversarial networks with temporal coherence, and the inference of Reynolds-averaged Navier-Stokes flows around airfoils. Beyond these physics-based deep learning works of the Thuerey group, this seminar will give an overview of recent developments in the field.

In this course, students will autonomously investigate recent research about machine learning techniques in the physical simulation area. Independent investigation for further reading, critical analysis, and evaluation of the topic are required.

Requirements

Participants are required to first read the assigned paper and start writing a report. This will help you prepare for your presentation.
Attendance
  • It is only allowed to miss a single time-slot. Missing a second one means failing the seminar. If you have to miss any, please let us know in advance.
Advisor
  • An advisor is assigned to each one with the paper.
  • Two weeks before the talk there will be a mandatory meeting with your advisor to review the report and discuss the structure of the presentation.
Report
  • A short report (4 pages max., excluding references in the ACM SIGGRAPH TOG format (acmtog) - you can download the precompiled latex template) should be prepared before the meeting with the advisor.
  • Guideline: You can begin with writing a summary of the work you present as a start point; but, it would be better if you focus more on your own research rather than just finishing with the summary of the paper. We, including you, are not interested in revisiting the work done before; it is more meaningful if you make an effort to put your own reasoning about the work, such as pros and cons, limitation, possible future work, your own ideas for the issues, etc.
Presentation (slides)
  • The participants have to present their topics in a talk (in English), which should last 30 minutes. Don't put too many technical details into the talk, make sure the audience gets the paper's main idea. Be prepared to answer questions regarding the technical details, you could prepare backup slides for that.
  • Afterwards, a short discussion session will follow.
  • Plagiarism is important; please do not simply copy the original authors' slides. You can certainly refer to them.
  • The semi-final slides (PDF) should be sent one week before the talk; otherwise, the talk will be canceled. We strongly encourage you to finalize the semi-final version as far as possible. We will take a look at the version and give feedback. You can revise your slides until your presentation.
  • Be ready in advance. We suggest testing the machines you are going to use before the lecture starts. You can bring your laptop or ask us one (also any converter you need for the projector) in advance. A laser pointer will be provided, so you can use if you want.
  • The final slides and report should be sent after the talk.

Virtual Seminar

BigBlueButton
  • For our seminar we use BigBlueButton: https://bigbluebutton.org/html5/
  • Please remain muted unless you have permission to speak. 
  • If you have a question or comment, please let us know in the chat, we will let you know as soon as you can speak, or post it directly in the chat.
Hardware Setup

A Laptop with built in microphone and speakers is terrible for everyone else in the virtual meeting!

  • Laptop fan makes noise (fan will most probably rev up after a couple of minutes, especially if you share your screen and if there are many users in the virtual meeting).
  • Typing and/or touchpad sounds will be transferred to communication partners.
  • Most likely there will be echo/feedback from the speakers.
  • Use the headset that came with your mobile phone. Even the cheapest headset will perform better than the built in microphone in a notebook.
  • (Cheap) bluetooth headsets are okay as well, but they perform not as good as wired ones!
  • Use an external microphone in combination with headphones. Some webcams have good quality microphones built in (e.g. Logitech).
  • Tablets like Apple iPads have quite solid built in microphones (and no fan). The same is possibly true for other tablets and even for smart phones. If you do not have a better alternative, they will most probably perform better than a Laptop.
Giving a Talk Online

Giving a talk in a virtual meeting scenario is quite different from giving a talk in a classroom scenario. You do not get any feedback from the audience during the talk (do they look happy/bored/satisfied?) and you cannot interact as easily in a non-verbal way with the audience as in the classroom, e.g. by pointing to things displayed on the projector screen. For this reason, mentally prepare yourself to talking to your screen for 20-30 minutes!

Normally, your slides should not contain too much written text. In the case of virtual meetings, some more information than usual is not wrong. Added info can help the audience to bridge gaps in case there was a glitch in the network connection, etc. Lastly, it is a good idea to open the content of the slides step by step to avoid that the audience can read faster than you talk.

Normally, when you give a talk, you are standing. This creates some tension in you, you will sound more energetic compared to talking while sitting, and, lastly, it helps you to concentrate. So it is a good idea to create a setup for giving the talk where you can stand in front of your screen.

You are supposed to give a talk. So do not write a script that you read to the audience. Reading out a pre-made script would be boring in a classroom scenario and it will be as boring in the virtual meeting!

Also important: switch off any type of messengers, notifications, etc. which can distract you. As you are pretty much isolated from the audience, it is a good idea to set a timer to keep track of elapsed time. Lastly, prepare for technical problems. Have some kind of second channel open to your advisor that can be used to contact you in case something unexpected happens.

Especially if you are using a quite “wide” (omnidirectional) microphone that sits on your desk, please avoid loud sounds from typing on your keyboard when you flip through slides while talking.

Papers

No Date Presenter Paper Advisor
01 16.05 G. Prastalo Accelerating Eulerian Fluid Simulation With Convolutional Networks Lukas
02 16.05 R. Koprinkov Learning to Simulate Complex Physics with Graph Networks Benjamin
16 16.05 A. Sleimi Purely data-driven medium-range weather forecasting achieves comparable skill to physical models at similar resolution Nilam
         
18 23.05 A. Xavier Deep Learning Methods for Reynolds-Averaged Navier-Stokes Simulations of Airfoil Flows Lukas
19 23.05 B. Nederkorn Data-driven nonlinear aeroelastic models of morphing wings for control Nilam
12 23.05 J. Tandberg Deep learning methods for super-resolution reconstruction of turbulent flows Benjamin
         
03 30.05 N. Wallat Physics Informed Deep Learning: Data-driven Solutions of Nonlinear Partial Differential Equations Lukas
04 30.05 L. Q. Nguyen Learning data-driven discretizations for partial differential equations Nilam
06 30.05 O. N. M. Al-Dabooni Neural Ordinary Differential Equations Benjamin
         
13 13.06 B. Börcsök Learning to control PDEs with differentiable physics Lukas
09 13.06 G. Ekaterina Combining Differentiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction Benjamin
05 13.06 T. B. P. Nguyen Discovering physical concepts with neural networks Nilam
         
20 20.06 P. Ding tempoGAN: A Temporally Coherent, Volumetric GAN for Super-resolution Fluid Flow Lukas
28 20.06 L. Gökalpay CFDNet: a deep learning-based accelerator for fluid simulations Nilam
10 20.06 I. Alvian Machine learning accelerated computational fluid dynamics Nilam
         
33 27.06 S. Ben Mohamed Learning to Assimilate in Chaotic Dynamical Systems Benjamin
32 27.06 Y. A. Tunali Multiwavelet-based Operator Learning for Differential Equations Benjamin
27 27.06 R. Ayari Shape As Points: A Differentiable Poisson Solver Lukas
         
15 04.07 P. Ferreira dos Santos Saraiva Deep learning and the Schrödinger equation Nilam
07 04.07 Y. Chen Hamiltonian Neural Networks Benjamin
26 04.07 K. Mio Learned Coarse Models for Efficient Turbulence Simulation Benjamin
         
08 - - Lagrangian Fluid Simulation with Continuous Convolutions -
11 - - Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solvers -
14 - - Assessment of unsteady flow predictions using hybrid deep learning based reduced-order models -
17 - - Model identification of reduced order fluid dynamics systems using deep learning -
21 - - Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations -
22 - - Solving high-dimensional partial differential equations using deep learning -
23 - - Transfer learning for nonlinear dynamics and its application to fluid turbulence -
24 - - SPNets: Differentiable Fluid Dynamics for Deep Neural Networks -
25 - - E(n) Equivariant Graph Neural Networks -
29 - - Towards Physics-informed Deep Learning for Turbulent Flow Prediction -
30 - - LSH-SMILE: Locality Sensitive Hashing Accelerated Simulation and Learning -
31 - - Compositional Modeling of Nonlinear Dynamical Systems with ODE-based Random Features -

You can access the papers through TUM library's eAccess.