A Convolutional Layer
(Published on 14. February 2024, 23:36 by crt)
A Convolutional Layer
My first ever puzzle.
RULES:
Normal sudoku rules apply.
Cells on a green German whisper line must differ by at least 5.
Cells in killer cages must sum to the clue in the top left corner.
Digits along a thermometer must strictly increase from the bulb end.
A grey square is an even digit.
A 2x2 box with a white square in the center performs a max-pool operation (as in convolutional layers) into the cell in the south east corner of the 2x2 box (i.e. R4C4 is the max value of digits around the white square in box 1).
A 2x2 box with a white cirlce in the center performs a convolution with a kernel 0210 (also a 2x2 matrix) and maps the value to the south east corner cell. A convolution is a matgematical operation, where for this particular case of 0210 we calculate the value as: NW box corner × 0 + NE corner × 2 + SW corner × 1 + SE corner × 0. So it is just a weighted average of the positive diagonal elements of the 2×2 box, i.e. if R7C5 is 3 and R8C4 is 2, then R9C6= 2 × 3 + 1 x 2 = 8
Enjoy!
Solution code: Row 3, followed by column 2 (no spaces)
Comments
on 18. February 2024, 00:50 by lmdemasi
The puzzle is interesting, but the rules are much too complicated for what you want in a 2* puzzle.
Overall the actual difficulty of the puzzle is 2* once you understand the rules.
on 15. February 2024, 10:24 by crt
Add some explanation to tge the white circles
on 15. February 2024, 10:18 by crt
Hi big tiger, I agree that they are not explained in the best way. Ill try to make it more clear and later in the day I will also add some images to show how they work. Hope that helps
on 15. February 2024, 08:07 by Big Tiger
I am totally lost on the 2x2 squares and circles ...
