arXiv subjects

Pascal Michel

Publications and source records attributed to Pascal Michel.

5 recordsLinked to original sources

arXiv2025

Turing machines deciders, part I

The Busy Beaver Challenge (or bbchallenge) aims at collaboratively solving the following conjecture: "$S(5) = 47{,}176{,}870$" [Rad\'o, 1962], [Marxen and Buntrock, 1990], [Aaronson, 2020]. This conjecture says that if a 5-state Turing machine runs for more than 47,176,870 steps without halting, then it will never halt -- starting from the all-0 tape. Proving this conjecture amounts to deciding whether 181,385,789 Turing machines with 5 states halt or not -- starting from the all-0 tape [bbchallenge, 2025]. To do so, we write $\textit{deciders}$: programs that take as input a Turing machine and output either HALT, NONHALT, or UNKNOWN. Each decider is specialised in recognising a particular type of non-halting behavior. After two years of work, the Busy Beaver Challenge achieved its goal in July 2024 by delivering a proof of "$S(5) = 47{,}176{,}870$" formalised in Coq [bbchallenge, 2025]. In this document, we present deciders that were developed before the Coq proof and which were mainly not used in the proof; nonetheless, they are relevant techniques for analysing Turing machines. Part II of this work is the decider section of our paper showing "$S(5) = 47{,}176{,}870$" [bbchallenge, 2025], presenting the deciders that were used in the Coq proof.

cs.LO↗

arXiv2020

Differentially Private Generation of Small Images

We explore the training of generative adversarial networks with differential privacy to anonymize image data sets. On MNIST, we numerically measure the privacy-utility trade-off using parameters from $\epsilon$-$\delta$ differential privacy and the inception score. Our experiments uncover a saturated training regime where an increasing privacy budget adds little to the quality of generated images. We also explain analytically why differentially private Adam optimization is independent of the gradient clipping parameter. Furthermore, we highlight common errors in previous works on differentially private deep learning, which we uncovered in recent literature. Throughout the treatment of the subject, we hope to prevent erroneous estimates of anonymity in the future.

cs.LG↗

arXiv2014

Simulation of the Collatz 3x+1 function by Turing machines

We give new Turing machines that simulate the iteration of the Collatz 3x+1 function. First, a never halting Turing machine with 3 states and 4 symbols, improving the known 3x5 and 4x4 Turing machines. Second, Turing machines that halt on the final loop, in the classes 3x10, 4x6, 5x4, and 13x2.

math.LO↗

arXiv2013

Problems in number theory from busy beaver competition

By introducing the busy beaver competition of Turing machines, in 1962, Rado defined noncomputable functions on positive integers. The study of these functions and variants leads to many mathematical challenges. This article takes up the following one: How can a small Turing machine manage to produce very big numbers? It provides the following answer: mostly by simulating Collatz-like functions, that are generalizations of the famous 3x+1 function. These functions, like the 3x+1 function, lead to new unsolved problems in number theory.

math.LO↗

arXiv2009

The Busy Beaver Competition: a historical survey

Tibor Rado defined the Busy Beaver Competition in 1962. He used Turing machines to give explicit definitions for some functions that are not computable and grow faster than any computable function. He put forward the problem of computing the values of these functions on numbers 1, 2, 3, ... More and more powerful computers have made possible the computation of lower bounds for these values. In 1988, Brady extended the definitions to functions on two variables. We give a historical survey of these works. The successive record holders in the Busy Beaver Competition are displayed, with their discoverers, the date they were found, and, for some of them, an analysis of their behavior.

math.LO↗