Course development

-Have a Quiz B session 1 hour week 5 -Move tutorial 4 seminar to week 5 -Unformally schedule oral exams week 8 -Make the practice Quizs more clearly practice (not points or required but gone over in a seminar or lecture) So complete/incomplete out of 0 points in Canvas Tutorials: • Tutoria l 2: ◦ Make sure yml is up to date or do not give it. ◦ ask them to print out actual tables for questions ▪ 1.1, 1.2, 1.3, 2.1, 3.5, ◦ Tasks 1-3, add to questions specific requests to screenshot the tables they use to answer. ◦ 4.5 be more clear that you want the table filled in. ◦ Make all of task 5 optional. ◦ 5.1 ask them to print out the scores for each model and say which is better. ◦ Or even switch from K2 to KS measure Replace K2 paper with: ▪ https://en.wikipedia.org/wiki/Kolmogorov%E2%80%93Smirnov_test • Tutorial 3: ◦ pdf: LBP 2(a) the psi are defined as the log of psi and not as in the code. ◦ 1.2 b ask for all cliques and which are Maximal cliques ◦ 2.2 Ask them to show some denoised images including with the best parameters. ◦ Ask them to evaluate the 10x 10 grid with MSE. • Tutorial 7 ◦ Number the questions 1-8 rather than having the numbering restart. ◦ Have a more precise way to specify the output for the coding parts. ▪ SeqMC • Fix the figure shown. (It is likely wrong) • Have them all start at x=0 and generate a sequence to estimate first for t=0 to 250 by rolling out the x sampling y and the noise each time. • Have them try 100; 1,000; and 10,000 samples • Then estimate that sequence by using the seq.MC • plot the whole sequence as a, for example, 2D 200x250 image with color map for t=50 to250 on the y and 0.2 bins in the x from -25 to 25. • Have them demonstrate that for enough samples (10,000) the two peaks become sinusoidal ridges (color stripes) centered on +-7 • Have them plot the actual path as a black curve on the image • Have them reduce the variance of the noise in the y to 0.5, generate a new sequence and see what happens when they run the seq.MC using the new measurement model with and without resampling. • Have them after 250 time steps stand still and just run resampling 1,000 times using w=1/N. Plot the histogram over x before and after doing that repeated resampling. ▪ MCMC • Have them try different kernel sizes from sigma 1 10, 20 and report the average Acceptance rates for each, plot the histograms. Run each sequence for 100 iterations per sample starting from a random point near 0. • Move the second mode to 30 and repeat.

See the other DD2420 course instance VT25.