A Linear Frequency Principle Model to Understand the Absence of Overfitting in Neural Networks

Why heavily parameterized neural networks(NNs) do not overfit the data is an important long standing open question. We propose a phenomenological model of the NN training to explain this non-overfitting puzzle. Our linear frequency principle(LFP) model accounts for a key dynamical feature of NNs: they learn low frequencies first, irrespective of microscopic details. Theory based on our LFP model shows that low frequency dominance of target functions is the key condition for the non-overfitting of NNs and is verified by experiments. Furthermore,through an ideal two-layer NN, we unravel how detailed microscopic NN training dynamics statistically gives rise to an LFP model with quantitative prediction power.     

时间序列的变化点检测和趋势分析

趋势分析和变化点检测是时间序列分析中常用的工具. 变化点检测是识别过程行为的自然或人为的突然的变化,而趋势可以定义为对逐渐偏离过去的规范的估计. 本文使用了Cox-Stuart方法和变化点算法分析时间序列数据趋势的存在,并以澳大利亚的近地表风速时间序列为例. 澳大利亚的近地表风速趋势是根据研究出的新开发的风速数据集,通过使用局部表面粗糙度信息,以及不同高度收集的混合观测数据构建. 10 m处的风的速度趋势通常会增加,而2 m处则趋于减小. 假设检验测试,变化点分析和人工检查记录表明有几个因素可能是导致差异的原因,例如伴随仪器变化的系统性偏差,随机数据错误(例如累积日错误)和数据采样问题. 均质化以及基于变化点检测的技术和多期趋势分析阐明了风速趋势不一致的根源.     

Phase Shift Modulations for Stable,Oscillatory, Traveling,Strongly Nonlinear Waves

Oscillatory dispersive waves propagating in a slowly varying medium are analyzed for Klein-Gordon equations with perturbations. The method of multiple scales is extended to include two fast scales, the usual traveling-wave phase and time, in order to allow initial conditions not usually permitted. An exact wave-action equation is introduced if the traveling wave is stable, involving averages over the periodic wave as well as time. This is equivalent to an extended averaged Lagrangian principle. The equation for the slow modulations of the phase shift of the traveling wave is derived from the higher order terms in the exact action equation and is shown to be the same as in earlier more restrictive studies.     

OH+C2H3F的反应机理及产物支化比反应的理论研究

本文对HOC₂H₃F可能解离通道的势能面进行从头算CCSD(T)/CBS//B3LYP/6-311G(d,p)计算,同时对速率常数进行Rice-Ramsperger-Kassel-Marcus计算. 生成主要产物CH₂CHO+HF最有利的反应途径是OHC₂H₃F→i2→TS14→i6→TS9→i3→TS3→CH₂CHO+HF,其中速率决定步骤是HF通过TS11从CO桥接位置解离,能量比反应物高3.8 kcal/mol. 借助中间态TS14,F原子从C_α迁移到C_β位置生成CH₂O+CH₂F,然后通过中间态TS16,H从O迁移到C_α位置;通过中间态TS5,C-C键断裂生成产物,其能量比反应物低1.8 kcal/mol,比TS11低4.0 kcal/mol.     

Regularity of flat free boundaries in two-phase problems for the p-Laplace operator

In this paper we continue the study in Lewis and Nyström (2010) [19], concerning the regularity of the free boundary in a general two-phase free boundary problem for the p-Laplace operator, by proving regularity of the free boundary assuming that the free boundary is close to a Lipschitz graph.     

Integer linear programming models for grid-based light post location problem

Selecting optimal location is a key decision problem in business and engineering. This research focuses to develop mathematical models for a special type of location problems called grid-based location problems. It uses a real-world problem of placing lights in a park to minimize the amount of darkness and excess supply. The non-linear nature of the supply function (arising from the light physics) and heterogeneous demand distribution make this decision problem truly intractable to solve. We develop ILP models that are designed to provide the optimal solution for the light post problem: the total number of light posts, the location of each light post, and their capacities (i.e., brightness). Finally, the ILP models are implemented within a standard modeling language and solved with the CPLEX solver. Results show that the ILP models are quite efficient in solving moderately sized problems with a very small optimality gap.     

Single Sample Path-Based Optimization of Markov Chains

Motivated by the needs of on-line optimization of real-world engineering systems, we studied single sample path-based algorithms for Markov decision problems (MDP). The sample path used in the algorithms can be obtained by observing the operation of a real system. We give a simple example to explain the advantages of the sample path-based approach over the traditional computation-based approach: matrix inversion is not required; some transition probabilities do not have to be known; it may save storage space; and it gives the flexibility of iterating the actions for a subset of the state space in each iteration. The effect of the estimation errors and the convergence property of the sample path-based approach are studied. Finally, we propose a fast algorithm, which updates the policy whenever the system reaches a particular set of states and prove that the algorithm converges to the true optimal policy with probability one under some conditions. The sample path-based approach may have important applications to the design and management of engineering systems, such as high speed communication networks.This work was supported in part by     

Middle‐School Mathematics Teachers' Beliefs in NCTM's Vision

This study examined the extent to which seventh‐ and eighth‐grade mathematics teachers are aware of National Council of Teachers of Mathematics (NCTM) standards documents, Curriculum and Evaluation Standards for School Mathematics and Principles and Standards for School Mathematics and agree with NCTM's vision of school mathematics as expressed in these documents. Quantitative data were collected through the Mathematics Standards Belief Survey (MSBS), a survey specifically designed to measure teachers' overall belief in NCTM's vision as well as in certain philosophical tenets of NCTM. Of the 82 seventh‐ and eighth‐grade mathematics teachers in the identified school district of Nevada, 73 (89.0%) participated in this study. The data revealed that, among seventh‐ and eighth‐grade mathematics teachers, secondary‐certified teachers had significantly higher MSBS scores than elementary‐certified teachers. A number of other findings were made, including significant differences among mean belief scores in the philosophical tenets of NCTM.     

The combinatorics and extreme value statistics of protein threading

In protein threading, one is given a protein sequence, together with a database of protein core structures that may contain the natural structure of the sequence. The object of protein threading is to correctly identify the structure(s) corresponding to the sequence. Since the core structures are already associated with specific biological functions, threading has the potential to provide biologists with useful insights about the function of a newly discovered protein sequence. Statistical tests for threading results based on the theory of extreme values suggest several combinatorial problems. For example, what is the number of waysm′=# _t {L _i >x _i } _i ⁼⁰ⁿ of choosing a sequence {X _i } _i ⁼¹ⁿ from the set {1, 2, ...,t}, subject to the difference constraints {L _i =X _i+1?X _i >x _i } _i ⁼⁰ⁿ , whereX ₀=0,X _n+1=t+1, and {x _i } _i ⁼⁰ⁿ is an arbitrary sequence of integers? The quantitym′ has many attractive combinatorial interpretations and reduces in special continuous limits to a probabilistic formula discovered by the Finetti. Just as many important probabilities can be derived from de Finetti's formula, many interesting combinatorial quantities can be derived fromm′. Empirical results presented here show that the combinatorial approach to threading statistics appears promising, but that structural periodicities in proteins and energetically unimportant structure elements probably introduce statistical correlations that must be better understood.     

Intriguing sets in partial quadrangles

The point‐line geometry known as a partial quadrangle (introduced by Cameron in 1975) has the property that for every point/line non‐incident pair (P, ?), there is at most one line through P concurrent with ?. So in particular, the well‐studied objects known as generalized quadrangles are each partial quadrangles. An intriguing set of a generalized quadrangle is a set of points which induces an equitable partition of size two of the underlying strongly regular graph. We extend the theory of intriguing sets of generalized quadrangles by Bamberg, Law and Penttila to partial quadrangles, which gives insight into the structure of hemisystems and other intriguing sets of generalized quadrangles. © 2010 Wiley Periodicals, Inc. J Combin Designs 19:217‐245, 2011