1317
Comment:
|
2881
|
Deletions are marked like this. | Additions are marked like this. |
Line 6: | Line 6: |
|| '''Organisation:''' || Seulki Yeom: seulki.yeom@tu-berlin.de, Philipp Seegerer: philipp.seegerer@tu-berlin.de, David Lassner: lassner@tu-berlin.de || | || '''Organisation:''' || Seulki Yeom: yeom@tu-berlin.de, Philipp Seegerer: philipp.seegerer@tu-berlin.de, David Lassner: lassner@tu-berlin.de || |
Line 23: | Line 23: |
(temporary) Workshop Lecture topics are: 1. Clustering, mixtures, density estimation * Density estimation: kernel density estimation, Parzen windows, parametric density/MaxLikelihood * K means clustering * Gaussian mixture models, EM algorithm * Curse of dimensionality 2. Manifold learning * LLE * Embeddings (RBF?)? * Multidimensional scaling * tSNE? 3. Bayesian Methods * What is learning? * Frequentist vs Bayes * Bayes rule * Naive Bayes * Bayesian linear regression * Bayesian/Akaike information criterion, Occam's razor? 4. Classical and linear methods * Matrix factorization * Logistic regression * Regularization, Lasso, Ridge regression * Fisher's Linear discriminant * Gradient descent? Where should it go? NNs? * Decision boundaries 5. Support Vector Machine * Linear SVM * Linear separability, margins * Duality in optimization, KKT conditions * SVM for regression * Multi-class SVM? * Applications 6. Kernels * Feature transformations * Kernel trick * Cross references to previous methods: ridge regression, PCA, SVM * Nadaraya-Watson kernel regression 7. Neural Networks * Rosenblatt's Perceptron * Multi layer perceptron * Motivation with Logistic regression * Backpropagation, (Stochastic) (Minibatch) gradient descent * Convolutional NNs * Famous Conv nets (imagenet winners): AlexNet, GoogleNet, ResNet * Recurrent NNs? * Applications * Practical recommendations for Training of DNNs (following e.g. Bengio's 2012 paper), hyperparameters 8. Recap |
Beginners Workshop Machine Learning
From:
2018-09-03
To:
2018-09-14
Exam:
2018-09-24
Organisation:
Seulki Yeom: yeom@tu-berlin.de, Philipp Seegerer: philipp.seegerer@tu-berlin.de, David Lassner: lassner@tu-berlin.de
Language
English
Enrollment / Limited number of participants
If you intend to participate, please send an e-mail to lassner@tu-berlin.de with title "Beginners Workshop Enrollment" and this text:
Name: Your name Matr.Nr: Your student ID (Matrikelnummer) Degree: The degree you are enrolled in and want to use this course for. TU student: Yes/No (Are you a enrolled as a regular student at TU Berlin?) Other student: If you are not a regular student, please write your status. ML1: Yes/No (Did you take the course Machine Learning 1 at TU Berlin?) Other ML course: If you did not take ML1 at TU Berlin, please write if you took any equivalent course.
Participation spots are mostly assigned on a random basis. Please keep in mind that auditing students and Nebenhörer can only participate if less than the maximum number of regular TU students register for the course (http://www.studsek.tu-berlin.de/menue/studierendenverwaltung/gast_und_nebenhoererschaft/parameter/en/).
(temporary) Workshop Lecture topics are:
1. Clustering, mixtures, density estimation
- Density estimation: kernel density estimation, Parzen windows, parametric density/MaxLikelihood
- K means clustering
- Gaussian mixture models, EM algorithm
- Curse of dimensionality
2. Manifold learning
- LLE
- Embeddings (RBF?)?
- Multidimensional scaling
- tSNE?
3. Bayesian Methods
- What is learning?
- Frequentist vs Bayes
- Bayes rule
- Naive Bayes
- Bayesian linear regression
- Bayesian/Akaike information criterion, Occam's razor?
4. Classical and linear methods
- Matrix factorization
- Logistic regression
- Regularization, Lasso, Ridge regression
- Fisher's Linear discriminant
- Gradient descent? Where should it go? NNs?
- Decision boundaries
5. Support Vector Machine
- Linear SVM
- Linear separability, margins
- Duality in optimization, KKT conditions
- SVM for regression
- Multi-class SVM?
- Applications
6. Kernels
- Feature transformations
- Kernel trick
- Cross references to previous methods: ridge regression, PCA, SVM
- Nadaraya-Watson kernel regression
7. Neural Networks
- Rosenblatt's Perceptron
- Multi layer perceptron
- Motivation with Logistic regression
- Backpropagation, (Stochastic) (Minibatch) gradient descent
- Convolutional NNs
Famous Conv nets (imagenet winners): AlexNet, GoogleNet, ResNet
- Recurrent NNs?
- Applications
- Practical recommendations for Training of DNNs (following e.g. Bengio's 2012 paper), hyperparameters
8. Recap