Statistické strojové učení



Cheat sheet

A cheat sheet is allowed for the exam, as described in course note from 2020-01-13:

You are allowed to prepare & use one A4 page with handwritten notes (one sided).

Example cheat sheet:

2019/2020 winter semester, prepared by bartefil


Celý test:



Zkoušky z minulých let, které se mi podařilo získat:



Midterm test, který jsme si letos mohli vypracovat nanečisto.





Solutions by bartefil:

  1. Regular Perceptron algorithm. See slide 4 in svm1_ws2019.pdf.
  2. Assignment 2:
    1. a) Evaluate ER for each h. Select h that maximizes ER.
    2. b) 5000 log(4000)
  3. Assignment 3:
    1. a) \alpha(k) = \frac{p(x, k)}{\sum_{k'} p(x, k')}
    2. b) Partial solution: \max_\pi \sum_{l=1}^m \sum_{k=0}^n \alpha_l(k) \log \pi_k
  4. Assignment 4:
    • Algorithm: for i in range(n): for k in K: p(s_i = k) := \sum_{k' \in K} p(s_i = k | s_{i-1} = k') p(s_{i-1} = k')
    • Complexity: O(n |K|^2)
  5. See slide 38 in ensembling-ws2019.pdf. Discussion missing.
  6. Assignment 5:
    • Auxiliary: y_j = \sum_i x_i w_{i,j}
    • Forward: z_k = \max(y_k, a_k y_k)
    • Backward: dz_k / dx_i = ([y_k > a_k y_k] + [z_k \leq a_k z_k] a_k) w_{i,k}
    • Parameter:
      • dz_k / da_p = [p = k] [y_k \leq a_k y_k] y_k
      • dz_k / dw_{l,m} = [k = m] ([y_k > a_k y_k] + [z_k \leq a_k z_k] a_k) x_l


Helpful materials


Majority of SSU subjects understandably explained here:


Lecture on SVM on MIT

MIT notes: + pokracovanie v part II. a III.

Neural nets + convolutional

3Blue1Brown: Neural Networks (YouTube playlist) Nice basic explanation of how neural networks work. Chapters 3 and 4 provide efficient explanations of backpropagation using good visualizations. Whole course on neural nets and convolutional networks. Very comprehensive lectures, explained from the basic concepts plus nice motivation examples.


First what is likelihood?

EM + gaussian mixture

Andrew Ng: Lecture on clustering, mixture of Gaussians, Jensen's inequality, EM algorithm (CS 229, Stanford University): video, lecture notes

Thomas P. Minka: Expectation-Maximization as lower bound maximization (recommended in lecture slides in 2019)

Bayes learning

MIT notes:


1) Note that pages 5 and 6 duplicate pages 1 and 2 respectively.
courses/be4m33ssu.txt · Poslední úprava: 2020/02/03 16:51 autor: rozsatib
Nahoru = chi`s home Valid CSS Driven by DokuWiki do yourself a favour and use a real browser - get firefox!! Recent changes RSS feed Valid XHTML 1.0