Numerical Physics and Machine Learning
We will cover many algothims used in many-body problems and complex systems: Monte Carlo methods, molecular dynamics and optmization in complex landscapes. We shall also discuss the use of some machine learning algorithms (Boltzmann machines, Auto-encoder, Deep Learning) for physics problems. We focus on algorithms and physics, not on programming and heavy numerics. The theoretical lecture is followed by a tutorial introducing concrete numerical exercises. You will have to hand in 3 homeworks.
Lectures on machine learning will be done remotely. You will be able to access them on https://epfl.zoom.us/my/krzakalaflorent Tutorials will take place as usual on the gotomeeting.
Where and When
- Lectures on Fridays: 14:00-16:00
- Tutorials on Fridays: 16:00-18:00
- ENS, 24 rue Lhomond, room L367 (third floor)
No previous experience in programming is required.
Programming Language: Python
For practical installation, we recommand either to use Anaconda (See Memento Python) or use google colab.
The Colaboratory platform from Google is quite good way to use powerful computer without buying one: It requires no specific hardware or software, and even allows you to use GPU computing for free, all by writting a jupyter notebook that you can then share.
Homeworks (50% of the mark) + 1 MCQ (50% of the final mark)
Project in Markox Chain Monte Carlo Manon Michel's Project
Friday, September 4, 2020
- Lecture 1 Introduction to Monte Carlo
- Tutorial 1 Markov Matrix
Friday, September 11, 2020
- Lecture 2 Basic Sampling
Friday, September 18, 2020
- Lecture 3: Errors and Precision
Friday, September 25, 2020
- Lecture 4: Ising model and phase transitions
Friday, October 2, 2020
Lecture 5: Quantum particle
Homework 2: Download
Friday, October 9, 2020
Lecture 6: Importance sampling
Friday, October 16, 2020
GoToMeeting link  (Room 1 M2 ICFP)
Lecture 7: Optimization & Dijkstra algorithm
Friday, October 23, 2020
Lecture 8: Maximum Likelyhood estimation:
Due: Homework 2 (send it to Marko)
Friday, November 06, 2020
Lecture 9: Restricted Boltzmann machines
Friday, November 13, 2020
Friday, November 27, 2020
Tomorrow Florent cannot really give the talk in direct live ... but never fear:
- he can make a short Q/A tomorrow at, say, 15h or 15H30
- he registered the whole lecture in video, and put it here:
Friday, December 4, 2020
Tutorial 12 Convolutional neural networks and auto-encoders
Due: Homework 3 (send it to me (Marko))
Friday, December 11, 2020
Multiple Choice Questions:
The Solution 
The MCQ is composed of 20 questions. For each question you have 4 choices: 3 wrong and 1 correct.
- If you check the correct one you get a point.
- If you are wrong you loose 1/4 of a point.
- No answer given: zero points.
The Zoom link
Follow the link . I will be there starting from 13h30, we will discuss the rules and I will be there to help you if you face a problem.
- The exam starts at 14h00: you download your file of questions from that dropbox directory that brings your name.
- Name the file with your answers as familyname_name.txt.
- The answers shuld be presented in the following way:
(if some question is missing - as question 3 here - it is not a problem)
- Send the file with your answers at email@example.com before 4 pm.
- You are allowed to use all material you think useful.
- You are not allowed to communicate with other people. Questions will be randomised to make hard life of cheaters, but please do not be one of them!
Here you can find a trial of the MCQ 10 Questions about numerical physics
you can find a short presentation video of the MCQ on the Dropbox by A. Rosso
- SMAC W. Krauth Statistical Mechanics: Algorithms and Computations (Oxford: Oxford University Press) (2006)
- Other references are specified in each lectures