基于smoothL1改进的边框回归损失函数

为了进一步提高目标检测任务中的边框回归精度,提出一种基于smoothL1改进的边框回归损失函数.通过自适应地增大smoothL1中非离散点的梯度,缓解了模型反向传播中对离散点和非离散点梯度分布不平衡问题,提高了模型的精度.实验结果表明,在PASCAL VOC2007测试集上,基于改进的smoothL1的目标检测模型Faster R-CNN,平均精度均值(mAP)达到了70.8％,相较smoothL1,模型精度有所提高.     

界定Stackelberg博弈下的混合平衡交通网络效率损失

考虑一个受控制的交通网络,一类用户属于领导者,按照系统最优原则选择出行路径;另一类用户属于跟随者且具有不完全信息,按照Logit型随机用户平衡原则选择出行路径.建立了描述这种Stackelberg博弈下的混合平衡出行行为的变分不等式模型,给出了满足此种混合平衡的交通网络的效率损失上界,结果表明,效率损失上界与被研究的交通网络拓扑结构,交通需求及控制系数有关.     

Bounding the piercing number

It is shown that for everyk and everyp≥q≥d+1 there is ac=c(k,p,q,d)<∞ such that the following holds. For every familyℋ whose members are unions of at mostk compact convex sets inR ^d in which any set ofp members of the family contains a subset of cardinalityq with a nonempty intersection there is a set of at mostc points inR ^d that intersects each member ofℋ. It is also shown that for everyp≥q≥d+1 there is aC=C(p,q,d)<∞ such that, for every family of compact, convex sets inR ^d so that among andp of them someq have a common hyperplane transversal, there is a set of at mostC hyperplanes that together meet all the members of . This research was supported in part by a United States-Israel BSF Grant and by the Fund for Basic Research administered by the Israel Academy of Sciences.     

Bounding the Reliability of Networks

Various applied problems require the analysis of systems with randomly failing elements. A typical such instance occurs in assessing the reliability of a transportation or communication network, in which the individual links fail independently with known probabilities. One is then interested in the probability that goods or information will be successfully transported from some source location to some terminal location. Exact calculation of this probability is known to be difficult, so this paper develops a method for obtaining lower and upper bounds on the required value. In fact, our procedure yields a sequence of matched lower and upper bounds, which can be monitored as the algorithm progresses. Rather than simply producing a numerical value, the method produces a functional (symbolic) form for the answer, especially useful for subsequent sensitivity analyses.     

Bounding the sup-norm of automorphic forms

No Abstract.  .     

Improved Alignment-Based Algorithm for Multilingual Text Compression

Multilingual text compression exploits the existence of the same text in several languages to compress the second and subsequent copies by reference to the first. This is done based on bilingual text alignment, a mapping of words and phrases in one text to their semantic equivalents in the translation. A new multilingual text compression scheme is suggested, which improves over an immediate generalization of bilingual algorithms. The idea is to store the necessary markup data within the source language text; the incurred compression loss due to this overhead is smaller than the savings in the compressed target language texts, for a large enough number of the latter. Experimental results are presented for a parallel corpus in six languages extracted from the EUR-Lex website of the European Union. These results show the superiority of the new algorithm as a function of the number of languages.     

An Adaptive Compression Algorithm in Besov Spaces

Given a function f on [0,1] and a wavelet-type expansion of f , we introduce a new algorithm providing an approximation $\tilde f of f with a prescribed number D of nonzero coefficients in its expansion. This algorithm depends only on the number of coefficients to be kept and not on any smoothness assumption on f . Nevertheless it provides the optimal rate D ^-α of approximation with respect to the L _q -norm when f belongs to some Besov space B ^α _p,∈fty whenever α>(1/p-1/q) ⁺ . These results extend to more general expansions including splines and piecewise polynomials and to multivariate functions. Moreover, this construction allows us to compute easily the metric entropy of Besov balls. June 21, 1996. Dates revised: April 9, 1998; October 14, 1998. Date accepted: October 20, 1998.     

Bounding the diameter of distance-regular graphs

LetG be a connected distance-regular graph with valencyk>2 and diameterd, but not a complete multipartite graph. Suppose that is an eigenvalue ofG with multiplicitym and that±k. We prove that bothd andk are bounded by functions ofm. This implies that, ifm>1 is given, there are only finitely many connected, co-connected distance-regular graphs with an eigenvalue of multiplicitym.This work was supported by NSERC grant A5367.     

Bounding one-way differences

Letf(n, k) denote the maximum length of a sequence (F ₁,F ₂,...) of distinct subsets of ann-element set with the property that|F _i F _j | < k for alli < j. We determine the exact values off(n, 2) and characterize all the extremal sequences. Fork 3 we prove that . Some related problems are also considered.     

界定用户平衡及Logit型随机用户平衡下的混合平衡交通网络效率损失

考虑一个具有两类用户的交通网络,一类用户按照用户平衡原则选择出行路径,另一类用户按照Logit型随机用户平衡原则选择出行路径.建立了描述这种混合平衡出行行为的变分不等式模型,给出了满足此种混合平衡的交通网络效率损失上界,结果表明,效率损失上界与被研究的交通网络拓扑结构,交通需求及两类用户的划分比例系数有关.     

Bounding multiplicative energy by the sumset

We prove that the sumset or the productset of any finite set of real numbers, A, is at least |A|^4/3−ε, improving earlier bounds. Our main tool is a new upper bound on the multiplicative energy, E(A,A).     

Bounding the Elliptic Mahler Measure II

The theory of heights for rational points on arithmetic ellipticcurves is becoming well known. An important fact in the basictheory is the relationship between the naïve and the canonicalheight of a rational point; in fact, they differ by a uniformlybounded amount. The paper provides a generalisation of thisfact from the point of view of heights of polynomials, ratherthan heights of points. This concept is extended to polynomialsin several variables.     

Bounding the spectrum of large Hermitian matrices

Estimating upper bounds of the spectrum of large Hermitian matrices has long been a problem with both theoretical and practical significance. Algorithms that can compute tight upper bounds with minimum computational cost will have applications in a variety of areas. We present a practical algorithm that exploits k-step Lanczos iteration with a safeguard step. The k is generally very small, say 5-8, regardless of the large dimension of the matrices. This makes the Lanczos iteration economical. The safeguard step can be realized with marginal cost by utilizing the theoretical bounds developed in this paper. The bounds establish the theoretical validity of a previous bound estimator that has been successfully used in various applications. Moreover, we improve the bound estimator which can now provide tighter upper bounds with negligible additional cost.     

Bounding the diameter of a distance-transitive graph

A graph Γ is distance-transitive if for all vertices u, v, x, y such that d(u, v) = d(x, y) there is an automorphism h of Γ such that uh = x, vh = y. We show how to find a bound for the diameter of a bipartite distance-transitive graph given a bound for the order |G_α| of the stabilizer of a vertex.     

Bounding the mim-width of hereditary graph classes

A large number of $\mathsf{NP}$-hard graph problems are solvable in $\mathsf{XP}$ time when parameterized by some width parameter. Hence, when solving problems on special graph classes, it is helpful to know if the graph class under consideration has bounded width. In this paper we consider maximum-induced matching width (mim-width), a particularly general width parameter that has a number of algorithmic applications whenever a decomposition is “quickly computable” for the graph class under consideration. We start by extending the toolkit for proving (un)boundedness of mim-width of graph classes. By combining our new techniques with known ones we then initiate a systematic study into bounding mim-width from the perspective of hereditary graph classes, and make a comparison with clique-width, a more restrictive width parameter that has been well studied. We prove that for a given graph $H$, the class of $H$-free graphs has bounded mim-width if and only if it has bounded clique-width. We show that the same is not true for $(H_1,H_2)$-free graphs. We identify several general classes of $(H_1,H_2)$-free graphs having unbounded clique-width, but bounded mim-width; moreover, we show that a branch decomposition of constant mim-width can be found in polynomial time for these classes. Hence, these results have algorithmic implications: when the input is restricted to such a class of $(H_1,H_2)$-free graphs, many problems become polynomial-time solvable, including classical problems, such as $k$- Colouring and Independent Set , domination-type problems known as Locally Checkable Vertex Subset and Vertex Partitioning (LC-VSVP) problems, and distance versions of LC-VSVP problems, to name just a few. We also prove a number of new results showing that, for certain $H_1$ and $H_2$, the class of $(H_1,H_2)$-free graphs has unbounded mim-width. Boundedness of clique-width implies boundedness of mim-width. By combining our results with the known bounded cases for clique-width, we present summary theorems of the current state of the art for the boundedness of mim-width for $(H_1,H_2)$-free graphs. In particular, we classify the mim-width of $(H_1,H_2)$-free graphs for all pairs $(H_1,H_2)$ with $\mid V(H_1)\mid +\mid V(H_2)\mid \le 8$. When $H_1$ and $H_2$ are connected graphs, we classify all pairs $(H_1,H_2)$ except for one remaining infinite family and a few isolated cases.     

Bounding the exponent of a verbal subgroup

We deal with the following conjecture. If \(w\) is a group word and \(G\) is a finite group in which any nilpotent subgroup generated by \(w\) -values has exponent dividing \(e\) , then the exponent of the verbal subgroup \(w(G)\) is bounded in terms of \(e\) and \(w\) only. We show that this is true in the case where \(w\) is either the \(n\text{ th }\) Engel word or the word \([x^n,y_1,y_2,\ldots ,y_k]\) (Theorem A). Further, we show that for any positive integer \(e\) there exists a number \(k=k(e)\) such that if \(w\) is a word and \(G\) is a finite group in which any nilpotent subgroup generated by products of \(k\) values of the word \(w\) has exponent dividing \(e\) , then the exponent of the verbal subgroup \(w(G)\) is bounded in terms of \(e\) and \(w\) only (Theorem B).     

Bounding the zeros of an interval equation

In this paper, we consider the problem of finding reliably and with certainty all zeros of an interval equation within a given search interval for continuously differentiable functions over real numbers. We propose a new formality of interval arithmetic which is treated in a theoretical manner to develop and prove a new method, lying on the context of interval Newton methods. Some important theoretical aspects of the new method are stated and proved. Finally, an algorithmic realization of our method is proposed to be applied on a set of test functions, where the promising theoretical results are verified.