UDL with Julia/Pluto
Contact
Prof. Dr. Claus Möbus
Room: A02 2-226
claus.moebus@uol.de
-------------------------------------------
Secretary
Manuela Wüstefeld
Room: A02 2-228
Tel: +49 441 / 798-4520
manuela.wuestefeld@uol.de
-------------------------------------------
UDL with Julia/Pluto
Understanding Deep Learning (UDL) with Julia/Pluto.jl
These Julia/Pluto scripts are my personal learning diary when working through Simon J.D. Prince's book Understanding Deep Learning (UDL). The book is accompanied by skeleton Python scripts to be completed by students to test their acquired knowledge. In contrast, I am developing full Julia/Pluto scripts in a style called literate programming to implement the book's neural network’s’ concepts. Though the scripts are carefully tested they are not production-ready.
I have two main goals. First, I try to improve my personal competence in Julia and second I am testing the quality und utility of Julia’s ecosystem for building neural net models.
I use the following libraries: ADTypes, Combinatorics, ComponentArrays, CUDA, cuDNN, DataFrames, Distributions, Flux, GLM, Latexify, LaTeXStrings, LogExpFunctions, Logistics, Lux, LuxCUDA, NNlib, Optim, Optimisers, Plots, Pluto, PlutoUI, Printf, ProgressMeter, Random, SpecialFunctions, Statistics, Symbolics, Zygote.
Here I put special emphasis on LUX.jl and LuxCUDA.jl.
Why Julia and no other languge? Having some background in Fortran, Lisp, Scheme, Prolog, and WebPPL Julia is more appealing to me than other languages. As others say it is a modern, elegant, and powerful language particularly suitabable for scientists (e.g. cognitive scientists) who deal with AI.
Claus Möbus
- Intro
- Supervised Learning
- Shallow Neural Networks
- Shallow Neural Networks I
- Shallow Neural Networks II
- Shallow Network Regions: Julia/Pluto-UDL-Notebook 3.3
- Activation functionst3://page?uid=100426
- Deep Neural Networks
- Composing Networks
- Clipping Functions
- Deep Networks
- Loss Functions
- Identification of a Latent Generator
- Maximum Likelihood Estimation (MLE)
- MLE & Least Squares Loss
- Binary Cross-Entropy Loss
- Multiclass Cross-Entropy Loss
- Fitting Models
- (Gradient-Free) Line Search
- Gradient Descent
- Stochastic Gradient Descent
- Momentum
- Adam
-------------------------------------------------------------------------------------------------------------
This is a draft under the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license. Comments, improvement and issue reports are welcome: claus.moebus(@)uol.de
-------------------------------------------------------------------------------------------------------------