Logic puzzle | EPFL Graph Search

A logic puzzle is a puzzle deriving from the mathematical field of deduction. The logic puzzle was first produced by Charles Lutwidge Dodgson, who is better known under his pen name Lewis Carroll, the author of Alice's Adventures in Wonderland. In his book The Game of Logic he introduced a game to solve problems such as confirming the conclusion "Some greyhounds are not fat" from the statements "No fat creatures run well" and "Some greyhounds run well". Puzzles like this, where we are given a list of premises and asked what can be deduced from them, are known as syllogisms. Dodgson goes on to construct much more complex puzzles consisting of up to 8 premises. In the second half of the 20th century mathematician Raymond M. Smullyan continued and expanded the branch of logic puzzles with books such as The Lady or the Tiger?, To Mock a Mockingbird and Alice in Puzzle-Land. He popularized the "knights and knaves" puzzles, which involve knights, who always tell the truth, and knaves, who always lie. There are also logic puzzles that are completely non-verbal in nature. Some popular forms include Sudoku, which involves using deduction to correctly place numbers in a grid; the nonogram, also called "Paint by Numbers", which involves using deduction to correctly fill in a grid with black-and-white squares to produce a picture; and logic mazes, which involve using deduction to figure out the rules of a maze. Another form of logic puzzle, popular among puzzle enthusiasts and available in magazines dedicated to the subject, is a format in which the set-up to a scenario is given, as well as the object (for example, determine who brought what dog to a dog show, and what breed each dog was), certain clues are given ("neither Misty nor Rex is the German Shepherd"), and then the reader fills out a matrix with the clues and attempts to deduce the solution. These are often referred to as "logic grid" puzzles. The data set of a logic grid puzzles can be any number of categories, but are limited by the corresponding increase in complexity, with most having only two, three, or even four categories.

Landscape and training regimes in deep learning

Matthieu Wyart, Mario Geiger, Leonardo Petrini

Deep learning algorithms are responsible for a technological revolution in a variety oftasks including image recognition or Go playing. Yet, why they work is not understood.Ultimately, they manage to classify data lying in high dimension – a feat generical ...

2021

Bretherton's buoyant bubble

John Martin Kolinski, Wassim Dhaouadi

When a buoyant bubble is inserted into a closed capillary that is slightly smaller than the capillary length, it appears stuck; exactly why this is so is a puzzle that has remained unanswered over the past 50 years. Recent calculations suggest that the bub ...

2019