-
Finite-Time Frequentist Regret Bounds of Multi-Agent Thompson Sampling on Sparse Hypergraphs
Authors:
Tianyuan Jin,
Hao-Lun Hsu,
William Chang,
Pan Xu
Abstract:
We study the multi-agent multi-armed bandit (MAMAB) problem, where $m$ agents are factored into $ρ$ overlapping groups. Each group represents a hyperedge, forming a hypergraph over the agents. At each round of interaction, the learner pulls a joint arm (composed of individual arms for each agent) and receives a reward according to the hypergraph structure. Specifically, we assume there is a local…
▽ More
We study the multi-agent multi-armed bandit (MAMAB) problem, where $m$ agents are factored into $ρ$ overlapping groups. Each group represents a hyperedge, forming a hypergraph over the agents. At each round of interaction, the learner pulls a joint arm (composed of individual arms for each agent) and receives a reward according to the hypergraph structure. Specifically, we assume there is a local reward for each hyperedge, and the reward of the joint arm is the sum of these local rewards. Previous work introduced the multi-agent Thompson sampling (MATS) algorithm \citep{verstraeten2020multiagent} and derived a Bayesian regret bound. However, it remains an open problem how to derive a frequentist regret bound for Thompson sampling in this multi-agent setting. To address these issues, we propose an efficient variant of MATS, the $ε$-exploring Multi-Agent Thompson Sampling ($ε$-MATS) algorithm, which performs MATS exploration with probability $ε$ while adopts a greedy policy otherwise. We prove that $ε$-MATS achieves a worst-case frequentist regret bound that is sublinear in both the time horizon and the local arm size. We also derive a lower bound for this setting, which implies our frequentist regret upper bound is optimal up to constant and logarithm terms, when the hypergraph is sufficiently sparse. Thorough experiments on standard MAMAB problems demonstrate the superior performance and the improved computational efficiency of $ε$-MATS compared with existing algorithms in the same setting.
△ Less
Submitted 24 December, 2023;
originally announced December 2023.
-
Graph Partitioning with Fujitsu Digital Annealer
Authors:
Yu-Ting Kao,
Hsiu-Chuan Hsu
Abstract:
Graph partitioning, or community detection, is the cornerstone of many fields, such as logistics, transportation and smart power grids. Efficient computation and efficacious evaluation of communities are both essential, especially in commercial and industrial settings. However, the solution space of graph partitioning increases drastically with the number of vertices and subgroups. With an eye to…
▽ More
Graph partitioning, or community detection, is the cornerstone of many fields, such as logistics, transportation and smart power grids. Efficient computation and efficacious evaluation of communities are both essential, especially in commercial and industrial settings. However, the solution space of graph partitioning increases drastically with the number of vertices and subgroups. With an eye to solving large scale graph partitioning and other optimization problems within a short period of time, the Digital Annealer (DA), a specialized CMOS hardware also featuring improved algorithms, has been devised by Fujitsu Ltd. This study gauges Fujitsu DA's performance and running times. The modularity was implemented as both the objective function and metric for the solutions. The graph partitioning problems were formatted into Quadratic Unconstrained Binary Optimization (QUBO) structures so that they could be adequately imported into the DA. The DA yielded the highest modularity among other studies when partitioning Karate Club, Les Miserables, American Football, and Dolphin. Moreover, the DA was able to partition the Case 1354pegase power grid network into 45 subgroups, calling for 60,930 binary variables, whilst delivering optimal modularity results within a solving time of roughly 80 seconds. Our results suggest that the Fujitsu DA can be applied for rapid and efficient optimization for graph partitioning.
△ Less
Submitted 28 November, 2023;
originally announced November 2023.
-
An analysis of $\mathbb{P}$-invariance and dynamical compensation properties from a control perspective
Authors:
Akram Ashyani,
Yu-Heng Wu,
Huan-Wei Hsu,
Torbjörn E. M. Nordling
Abstract:
Dynamical compensation (DC) provides robustness to parameter fluctuations. As an example, DC enable control of the functional mass of endocrine or neuronal tissue essential for controlling blood glucose by insulin through a nonlinear feedback loop. Researchers have shown that DC is related to structural unidentifiability and $\mathbb{P}$-invariance property, and $\mathbb{P}$-invariance property is…
▽ More
Dynamical compensation (DC) provides robustness to parameter fluctuations. As an example, DC enable control of the functional mass of endocrine or neuronal tissue essential for controlling blood glucose by insulin through a nonlinear feedback loop. Researchers have shown that DC is related to structural unidentifiability and $\mathbb{P}$-invariance property, and $\mathbb{P}$-invariance property is a sufficient and necessary condition for the DC property. In this article, we discuss DC and $\mathbb{P}$-invariancy from an adaptive control perspective. An adaptive controller is a self-tuning controller used to compensate for changes in a dynamical system. To design an adaptive controller with the DC property, it is easier to start with a two-dimensional dynamical model. We introduce a simplified system of ordinary differential equations (ODEs) with the DC property and extend it to a general form. The value of the ideal adaptive control lies in developing methods to synthesize DC to variations in multiple parameters. Then we investigate the stability of the system with time-varying input and disturbance signals, with a focus on the system's $\mathbb{P}$-invariance properties. This study provides phase portraits and step-like response graphs to visualize the system's behavior and stability properties.
△ Less
Submitted 9 November, 2023; v1 submitted 20 March, 2023;
originally announced March 2023.
-
Statistics of Partial Permutations via Catalan matrices
Authors:
Yen-Jen Cheng,
Sen-Peng Eu,
Hsiang-Chun Hsu
Abstract:
A generalized Catalan matrix $(a_{n,k})_{n,k\ge 0}$ is generated by two seed sequences $\mathbf{s}=(s_0,s_1,\ldots)$ and $\mathbf{t}=(t_1,t_2,\ldots)$ together with a recurrence relation. By taking $s_\ell=2\ell+1$ and $t_\ell=\ell^2$ we can interpret $a_{n,k}$ as the number of partial permutations, which are $n\times n$ $0,1$-matrices of $k$ zero rows with at most one $1$ in each row or column. I…
▽ More
A generalized Catalan matrix $(a_{n,k})_{n,k\ge 0}$ is generated by two seed sequences $\mathbf{s}=(s_0,s_1,\ldots)$ and $\mathbf{t}=(t_1,t_2,\ldots)$ together with a recurrence relation. By taking $s_\ell=2\ell+1$ and $t_\ell=\ell^2$ we can interpret $a_{n,k}$ as the number of partial permutations, which are $n\times n$ $0,1$-matrices of $k$ zero rows with at most one $1$ in each row or column. In this paper we prove that most of fundamental statistics and some set-valued statistics on permutations can also be defined on partial permutations and be encoded in the seed sequences. Results on two interesting permutation families, namely the connected permutations and cycle-up-down permutations, are also given.
△ Less
Submitted 20 July, 2022;
originally announced July 2022.
-
An Easy-to-implement Construction for $(k,n)$-threshold Progressive Visual Secret Sharing Schemes
Authors:
Hong-Bin Chen,
Hsiang-Chun Hsu,
Justie Su-Tzu Juan
Abstract:
Visual cryptography encrypts the secret image into $n$ shares (transparency) so that only stacking a qualified number of shares can recover the secret image by the human visual system while no information can be revealed without a large enough number of shares. This paper investigates the $(k,n)$-threshold Visual Secret Sharing (VSS) model, where one can decrypt the original image by stacking at l…
▽ More
Visual cryptography encrypts the secret image into $n$ shares (transparency) so that only stacking a qualified number of shares can recover the secret image by the human visual system while no information can be revealed without a large enough number of shares. This paper investigates the $(k,n)$-threshold Visual Secret Sharing (VSS) model, where one can decrypt the original image by stacking at least $k$ shares and get nothing with less than $k$ shares. There are two main approaches in the literature: codebook-based schemes and random-grid-based schemes; the former is the case of this paper. In general, given any positive integers $k$ and $n$, it is not easy to design a valid scheme for the $(k,n)$-threshold VSS model. In this paper, we propose a simple strategy to construct an efficient scheme for the $(k,n)$-threshold VSS model for any positive integers $2\leq k\leq n$. The crucial idea is to establish a seemingly unrelated connection between the $(k,n)$-threshold VSS scheme and a mathematical structure -- the generalized Pascal's triangle. This paper improves and extends previous results in four aspects:
Our construction offers a unified viewpoint and covers several known results;
The resulting scheme has a progressive-viewing property that means the more shares being stacked together the clearer the secret image would be revealed.
The proposed scheme can be constructed explicitly and efficiently based on the generalized Pascal's triangle without a computer.
Performance of the proposed scheme is comparable with known results.
△ Less
Submitted 20 February, 2020;
originally announced February 2020.
-
Folding Phenomenon of Major-balance Identities on Restricted Involutions
Authors:
Tung-Shan Fu,
Hsian-Chun Hsu,
Hsin-Chieh Liao
Abstract:
In this paper we prove a refined major-balance identity on the $321$-avoiding involutions of length $n$, respecting the leading element of permutations. The proof is based on a sign-reversing involution on the lattice paths within a $\lfloor\frac{n}{2}\rfloor\times\lceil\frac{n}{2}\rceil$ rectangle. Moreover, we prove affirmatively a question about refined major-balance identity on the $123$-avoid…
▽ More
In this paper we prove a refined major-balance identity on the $321$-avoiding involutions of length $n$, respecting the leading element of permutations. The proof is based on a sign-reversing involution on the lattice paths within a $\lfloor\frac{n}{2}\rfloor\times\lceil\frac{n}{2}\rceil$ rectangle. Moreover, we prove affirmatively a question about refined major-balance identity on the $123$-avoiding involutions, respecting the number of descents.
△ Less
Submitted 13 November, 2017;
originally announced November 2017.
-
Signed Countings of types B and D permutations and $t,q$-Euler Numbers
Authors:
Sen-Peng Eu,
Tung-Shan Fu,
Hsiang-Chun Hsu,
Hsin-Chieh Liao
Abstract:
It is a classical result that the parity-balance of the number of weak excedances of all permutations (derangements, respectively) of length $n$ is the Euler number $E_n$, alternating in sign, if $n$ is odd (even, respectively). Josuat-Vergès obtained a $q$-analog of the results respecting the number of crossings of a permutation. One of the goals in this paper is to extend the results to the perm…
▽ More
It is a classical result that the parity-balance of the number of weak excedances of all permutations (derangements, respectively) of length $n$ is the Euler number $E_n$, alternating in sign, if $n$ is odd (even, respectively). Josuat-Vergès obtained a $q$-analog of the results respecting the number of crossings of a permutation. One of the goals in this paper is to extend the results to the permutations (derangements, respectively) of types B and D, on the basis of the joint distribution in statistics excedances, crossings and the number of negative entries obtained by Corteel, Josuat-Vergès and Kim.
Springer numbers are analogous Euler numbers that count the alternating permutations of type B, called snakes. Josuat-Vergès derived bivariate polynomials $Q_n(t,q)$ and $R_n(t,q)$ as generalized Euler numbers via successive $q$-derivatives and multiplications by $t$ on polynomials in $t$. The other goal in this paper is to give a combinatorial interpretation of $Q_n(t,q)$ and $R_n(t,q)$ as the enumerators of the snakes with restrictions.
△ Less
Submitted 3 February, 2018; v1 submitted 18 August, 2017;
originally announced August 2017.
