Source author record

You Lu

Explore this author’s publications, collaborators and research topics. This catalog record has not been claimed by its author.

ResearcherUnclaimed source record

math.CO Machine Learning physics.chem-ph

Catalog footprint

Research overview

9works

3topics

4close collaborators

Actions

Is this your work?

Open graph Browse papers

preprint2021arXiv

Woodbury Transformations for Deep Generative Flows

Normalizing flows are deep generative models that allow efficient likelihood calculation and sampling. The core requirement for this advantage is that they are constructed using functions that can be efficiently inverted and for which the determinant of the function's Jacobian can be efficiently computed. Researchers have introduced various such flow operations, but few of these allow rich interactions among variables without incurring significant computational costs. In this paper, we introduce Woodbury transformations, which achieve efficient invertibility via the Woodbury matrix identity and efficient determinant calculation via Sylvester's determinant identity. In contrast with other operations used in state-of-the-art normalizing flows, Woodbury transformations enable (1) high-dimensional interactions, (2) efficient sampling, and (3) efficient likelihood evaluation. Other similar operations, such as 1x1 convolutions, emerging convolutions, or periodic convolutions allow at most two of these three advantages. In our experiments on multiple image datasets, we find that Woodbury transformations allow learning of higher-likelihood models than other flow architectures while still enjoying their efficiency advantages.

preprint2020arXiv

Structured Output Learning with Conditional Generative Flows

Traditional structured prediction models try to learn the conditional likelihood, i.e., p(y|x), to capture the relationship between the structured output y and the input features x. For many models, computing the likelihood is intractable. These models are therefore hard to train, requiring the use of surrogate objectives or variational inference to approximate likelihood. In this paper, we propose conditional Glow (c-Glow), a conditional generative flow for structured output learning. C-Glow benefits from the ability of flow-based models to compute p(y|x) exactly and efficiently. Learning with c-Glow does not require a surrogate objective or performing inference during training. Once trained, we can directly and efficiently generate conditional samples. We develop a sample-based prediction method, which can use this advantage to do efficient and effective inference. In our experiments, we test c-Glow on five different tasks. C-Glow outperforms the state-of-the-art baselines in some tasks and predicts comparable outputs in the other tasks. The results show that c-Glow is versatile and is applicable to many different structured prediction problems.

preprint2015arXiv

Shortest circuit covers of signed graphs

A shortest circuit cover ${\cal F}$ of a bridgeless graph $G$ is a family of circuits that covers every edge of $G$ and is of minimum total length. The total length of a shortest circuit cover ${\cal F}$ of $G$ is denoted by $SCC(G)$ . For ordinary graphs (graphs without sign), the subject of shortest circuit cover is closely related to some mainstream areas, such as, Tutte's integer flow theory, circuit double cover conjecture, Fulkerson conjecture, and others. For signed graphs $G$ , it is proved recently by Máčajová, Raspaud, Rollová and Škoviera that $SCC(G) \leq 11|E|$ if $G$ is s-bridgeless, and $SCC(G) \leq 9|E|$ if $G$ is $2$ -edge-connected. In this paper this result is improved as follows, $SCC(G) ~ \leq ~ |E| + 3|V| +z$ where $z ~=~ \min \{ \frac{2}{3}|E|+\frac{4}{3}ε_N-7,~ |V| + 2ε_N -8\}$ and $ε_N$ is the negativeness of $G$ . The above upper bound can be further reduced if $G$ is $2$ -edge-connected with even negativeness.

preprint2014arXiv

Efficient Calculation of Electronic Absorption Spectra by Means of Intensity-Selected TD-DFTB

During the last two decades density functional based linear response approaches have become the de facto standard for the calculation of optical properties of small and medium-sized molecules. At the heart of these methods is the solution of an eigenvalue equation in the space of single-orbital transitions, whose quickly increasing number makes such calculations costly if not infeasible for larger molecules. This is especially true for time-dependent density functional tight binding (TD-DFTB), where the evaluation of the matrix elements is inexpensive. For the relatively large systems that can be studied the solution of the eigenvalue equation therefore determines the cost of the calculation. We propose to do an oscillator strength based truncation of the single-orbital transition space to reduce the computational effort of TD-DFTB based absorption spectra calculations. We show that even a sizeable truncation does not destroy the principal features of the absorption spectrum, while naturally avoiding the unnecessary calculation of excitations with small oscillator strengths. We argue that the reduced computational cost of intensity-selected TD-DFTB together with its ease of use compared to other methods lowers the barrier of performing optical properties calculations of large molecules, and can serve to make such calculations possible in a wider array of applications.

preprint2012arXiv

On the p-reinforcement and the complexity

Let $G=(V,E)$ be a graph and $p$ be a positive integer. A subset $S\subseteq V$ is called a $p$ -dominating set if each vertex not in $S$ has at least $p$ neighbors in $S$ . The $p$ -domination number $\g_p(G)$ is the size of a smallest $p$ -dominating set of $G$ . The $p$ -reinforcement number $r_p(G)$ is the smallest number of edges whose addition to $G$ results in a graph $G'$ with $\g_p(G')<\g_p(G)$. In this paper, we give an original study on the $p$ -reinforcement, determine $r_p(G)$ for some graphs such as paths, cycles and complete $t$ -partite graphs, and establish some upper bounds of $r_p(G)$ . In particular, we show that the decision problem on $r_p(G)$ is NP-hard for a general graph $G$ and a fixed integer $p\geq 2$ .

preprint2012arXiv

The 2-Domination and 2-Bondage Numbers of Grid Graphs

Let $p$ be a positive integer and $G=(V,E)$ be a simple graph. A subset $D\subseteq V$ is a $p$ -dominating set if each vertex not in $D$ has at least $p$ neighbors in $D$ . The $p$ -domination number $\g_p(G)$ is the minimum cardinality among all $p$ -dominating sets of $G$ . The $p$ -bondage number $b_p(G)$ is the cardinality of a smallest set of edges whose removal from $G$ results in a graph with a $p$ -domination number greater than the $p$ -domination number of $G$ . In this note we determine the 2-domination number $\g_2$ and 2-bondage number $b_2$ for the grid graphs $G_{m,n}=P_m\times P_n$ for $2\leq m\leq 4$ .

preprint2012arXiv

The p-Domination Number of Complete Multipartite Graphs

Let $G=(V,E)$ be a graph and $p$ a positive integer. A subset $S\subseteq V$ is called a $p$ -dominating set of $G$ if every vertex not in $S$ has at least $p$ neighbors in $S$ . The $p$ -domination number is the minimum cardinality of a $p$ -dominating set in $G$ . In this paper, we establish an exact formula of the $p$ -domination number of all complete multipartite graphs for arbitrary positive integer $p$ .

preprint2012arXiv

Trees with Maximum p-Reinforcement Number

Let $G=(V,E)$ be a graph and $p$ a positive integer. The $p$ -domination number $\g_p(G)$ is the minimum cardinality of a set $D\subseteq V$ with $|N_G(x)\cap D|\geq p$ for all $x\in V\setminus D$ . The $p$ -reinforcement number $r_p(G)$ is the smallest number of edges whose addition to $G$ results in a graph $G'$ with $\g_p(G')<\g_p(G)$. Recently, it was proved by Lu et al. that $r_p(T)\leq p+1$ for a tree $T$ and $p\geq 2$ . In this paper, we characterize all trees attaining this upper bound for $p\geq 3$ .

preprint2011arXiv

The total bondage number of grid graphs

The total domination number of a graph $G$ without isolated vertices is the minimum number of vertices that dominate all vertices in $G$ . The total bondage number $b_t(G)$ of $G$ is the minimum number of edges whose removal enlarges the total domination number. This paper considers grid graphs. An $(n,m)$ -grid graph $G_{n,m}$ is defined as the cartesian product of two paths $P_n$ and $P_m$ . This paper determines the exact values of $b_t(G_{n,2})$ and $b_t(G_{n,3})$ , and establishes some upper bounds of $b_t(G_{n,4})$ .

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

You Lu

Research overview

Is this your work?

9 publications

Woodbury Transformations for Deep Generative Flows

Structured Output Learning with Conditional Generative Flows

Shortest circuit covers of signed graphs

Efficient Calculation of Electronic Absorption Spectra by Means of Intensity-Selected TD-DFTB

On the p-reinforcement and the complexity

The 2-Domination and 2-Bondage Numbers of Grid Graphs

The p-Domination Number of Complete Multipartite Graphs

Trees with Maximum p-Reinforcement Number

The total bondage number of grid graphs

See the researcher in context

Building this map preview