arXiv subjects

Yinan Li

Publications and source records attributed to Yinan Li.

68 records · Page 4Linked to original sources

arXiv2019

PANDA: AdaPtive Noisy Data Augmentation for Regularization of Undirected Graphical Models

We propose an AdaPtive Noise Augmentation (PANDA) technique to regularize the estimation and construction of undirected graphical models. PANDA iteratively optimizes the objective function given the noise augmented data until convergence to achieve regularization on model parameters. The augmented noises can be designed to achieve various regularization effects on graph estimation, such as the bridge (including lasso and ridge), elastic net, adaptive lasso, and SCAD penalization; it also realizes the group lasso and fused ridge. We examine the tail bound of the noise-augmented loss function and establish that the noise-augmented loss function and its minimizer converge almost surely to the expected penalized loss function and its minimizer, respectively. We derive the asymptotic distributions for the regularized parameters through PANDA in generalized linear models, based on which, inferences for the parameters can be obtained simultaneously with variable selection. We show the non-inferior performance of PANDA in constructing graphs of different types in simulation studies and apply PANDA to an autism spectrum disorder data to construct a mixed-node graph. We also show that the inferences based on the asymptotic distribution of regularized parameter estimates via PANDA achieve nominal or near-nominal coverage and are far more efficient, compared to some existing post-selection procedures. Computationally, PANDA can be easily programmed in software that implements (GLMs) without resorting to complicated optimization techniques.

stat.ML↗

arXiv2019

AdaPtive Noisy Data Augmentation (PANDA) for Simultaneous Construction of Multiple Graph Models

We extend the data augmentation technique PANDA by Li et al. (2018) that regularizes single graph estimation to jointly learning multiple graphical models with various node types in a unified framework. We design two types of noise to augment the observed data: the first type regularizes the estimation of each graph while the second type promotes either the structural similarity, referred as the \joint group lasso regularization, or the numerical similarity, referred as the joint fused ridge regularization, among the edges in the same position across graphs. The computation in PANDA is straightforward and only involves obtaining maximum likelihood estimator in generalized linear models in an iterative manner. The simulation studies demonstrate PANDA is non-inferior to existing joint estimation approaches for Gaussian graphical models, and significantly improves over the naive differencing approach for non-Gaussian graphical models. We apply PANDA to a real-life lung cancer microarray data to simultaneously construct four protein networks.

stat.ME↗

arXiv2019

Incorporating Weisfeiler-Leman into algorithms for group isomorphism

In this paper we combine many of the standard and more recent algebraic techniques for testing isomorphism of finite groups (GpI) with combinatorial techniques that have typically been applied to Graph Isomorphism. In particular, we show how to combine several state-of-the-art GpI algorithms for specific group classes into an algorithm for general GpI, namely: composition series isomorphism (Rosenbaum-Wagner, Theoret. Comp. Sci., 2015; Luks, 2015), recursively-refineable filters (Wilson, J. Group Theory, 2013), and low-genus GpI (Brooksbank-Maglione-Wilson, J. Algebra, 2017). Recursively-refineable filters -- a generalization of subgroup series -- form the skeleton of this framework, and we refine our filter by building a hypergraph encoding low-genus quotients, to which we then apply a hypergraph variant of the k-dimensional Weisfeiler-Leman technique. Our technique is flexible enough to readily incorporate additional hypergraph invariants or additional characteristic subgroups.

cs.CC↗

arXiv2018

Distinguishing Unitary Gates on the IBM Quantum Processor

An unknown unitary gates, which is secretly chosen from several known ones, can always be distinguished perfectly. In this paper, we implement such a task on IBM's quantum processor. More precisely, we experimentally demonstrate the discrimination of two qubit unitary gates, the identity gate and the $\frac{2}{3}π$-phase shift gate, using two discrimination schemes -- the parallel scheme and the sequential scheme. We program these two schemes on the \emph{ibmqx4}, a $5$-qubit superconducting quantum processor via IBM cloud, with the help of the $QSI$ modules [S. Liu et al.,~arXiv:1710.09500, 2017]. We report that both discrimination schemes achieve success probabilities at least 85%.

quant-ph↗

arXiv2018

Whiteout: Gaussian Adaptive Noise Regularization in Deep Neural Networks

Noise injection (NI) is an efficient technique to mitigate over-fitting in neural networks (NNs). The Bernoulli NI procedure as implemented in dropout and shakeout has connections with $l_1$ and $l_2$ regularization for the NN model parameters. We propose whiteout, a family NI regularization techniques (NIRT) through injecting adaptive Gaussian noises during the training of NNs. Whiteout is the first NIRT than imposes a broad range of the $l_γ$ sparsity regularization $(γ\in(0,2))$ without having to involving the $l_2$ regularization. Whiteout can also be extended to offer regularizations similar to the adaptive lasso and group lasso. We establish the regularization effect of whiteout in the framework of generalized linear models with closed-form penalty terms and show that whiteout stabilizes the training of NNs with decreased sensitivity to small perturbations in the input. We establish that the noise-perturbed empirical loss function (pelf) with whiteout converges almost surely to the ideal loss function (ilf), and the minimizer of the pelf is consistent for the minimizer of the ilf. We derive the tail bound on the pelf to establish the practical feasibility in its minimization. The superiority of whiteout over the Bernoulli NIRTs, dropout and shakeout, in learning NNs with relatively small-sized training sets and non-inferiority in large-sized training sets is demonstrated in both simulated and real-life data sets. This work represents the first in-depth theoretical, methodological, and practical examination of the regularization effects of both additive and multiplicative Gaussian NI in deep NNs.

stat.ML↗

arXiv2018

Quantum Divide-and-Conquer Anchoring for Separable Non-negative Matrix Factorization

It is NP-complete to find non-negative factors $W$ and $H$ with fixed rank $r$ from a non-negative matrix $X$ by minimizing $\|X-WH^\top\|_F^2$. Although the separability assumption (all data points are in the conical hull of the extreme rows) enables polynomial-time algorithms, the computational cost is not affordable for big data. This paper investigates how the power of quantum computation can be capitalized to solve the non-negative matrix factorization with the separability assumption (SNMF) by devising a quantum algorithm based on the divide-and-conquer anchoring (DCA) scheme. The design of quantum DCA (QDCA) is challenging. In the divide step, the random projections in DCA is completed by a quantum algorithm for linear operations, which achieves the exponential speedup. We then devise a heuristic post-selection procedure which extracts the information of anchors stored in the quantum states efficiently. Under a plausible assumption, QDCA performs efficiently, achieves the quantum speedup, and is beneficial for high dimensional problems.

quant-ph↗

arXiv2017

Tripartite-to-bipartite Entanglement Transformation by Stochastic Local Operations and Classical Communication and the Structure of Matrix Spaces

We study the problem of transforming a tripartite pure state to a bipartite one using stochastic local operations and classical communication (SLOCC). It is known that the tripartite-to-bipartite SLOCC convertibility is characterized by the maximal Schmidt rank of the given tripartite state, i.e. the largest Schmidt rank over those bipartite states lying in the support of the reduced density operator. In this paper, we further study this problem and exhibit novel results in both multi-copy and asymptotic settings. In the multi-copy regime, we observe that the maximal Schmidt rank is strictly super-multiplicative, i.e. the maximal Schmidt rank of the tensor product of two tripartite pure states can be strictly larger than the product of their maximal Schmidt ranks. We then provide a full characterization of those tripartite states whose maximal Schmidt rank is strictly super-multiplicative when taking tensor product with itself. In the asymptotic setting, we focus on determining the tripartite-to-bipartite SLOCC entanglement transformation rate, which turns out to be equivalent to computing the asymptotic maximal Schmidt rank of the tripartite state, defined as the regularization of its maximal Schmidt rank. Despite the difficulty caused by the super-multiplicative property, we provide explicit formulas for evaluating the asymptotic maximal Schmidt ranks of two important families of tripartite pure states, by resorting to certain results of the structure of matrix spaces, including the study of matrix semi-invariants. These formulas give a sufficient and necessary condition to determine whether a given tripartite pure state can be transformed to the bipartite maximally entangled state under SLOCC, in the asymptotic setting. Applying the recent progress on the non-commutative rank problem, we can verify this condition in deterministic polynomial time.

quant-ph↗

arXiv2017

$Q|SI\rangle$: A Quantum Programming Environment

This paper describes a quantum programming environment, named $Q|SI\rangle$. It is a platform embedded in the .Net language that supports quantum programming using a quantum extension of the $\mathbf{while}$-language. The framework of the platform includes a compiler of the quantum $\mathbf{while}$-language and a suite of tools for simulating quantum computation, optimizing quantum circuits, and analyzing and verifying quantum programs. Throughout the paper, using $Q|SI\rangle$ to simulate quantum behaviors on classical platforms with a combination of components is demonstrated. The scalable framework allows the user to program customized functions on the platform. The compiler works as the core of $Q|SI\rangle$ bridging the gap from quantum hardware to quantum software. The built-in decomposition algorithms enable the universal quantum computation on the present quantum hardware.

quant-ph↗

arXiv2017

Linear algebraic analogues of the graph isomorphism problem and the Erdős-Rényi model

A classical difficult isomorphism testing problem is to test isomorphism of p-groups of class 2 and exponent p in time polynomial in the group order. It is known that this problem can be reduced to solving the alternating matrix space isometry problem over a finite field in time polynomial in the underlying vector space size. We propose a venue of attack for the latter problem by viewing it as a linear algebraic analogue of the graph isomorphism problem. This viewpoint leads us to explore the possibility of transferring techniques for graph isomorphism to this long-believed bottleneck case of group isomorphism. In 1970's, Babai, Erdős, and Selkow presented the first average-case efficient graph isomorphism testing algorithm (SIAM J Computing, 1980). Inspired by that algorithm, we devise an average-case efficient algorithm for the alternating matrix space isometry problem over a key range of parameters, in a random model of alternating matrix spaces in vein of the Erdős-Rényi model of random graphs. For this, we develop a linear algebraic analogue of the classical individualisation technique, a technique belonging to a set of combinatorial techniques that has been critical for the progress on the worst-case time complexity for graph isomorphism, but was missing in the group isomorphism context. As a consequence of the main algorithm, we establish a weaker linear algebraic analogue of Erdős and Rényi's classical result that most graphs have the trivial automorphism group. We finally show that Luks' dynamic programming technique for graph isomorphism (STOC 1999) can be adapted to slightly improve the worst-case time complexity of the alternating matrix space isometry problem in a certain range of parameters.

cs.DS↗

arXiv2017

Indistinguishability of bipartite states by positive-partial-transpose operations in the many-copy scenario

A bipartite subspace $S$ is called strongly positive-partial-transpose-unextendible (PPT-unextendible) if for every positive integer $k$, there is no PPT operator supporting on the orthogonal complement of $S^{\otimes k}$. We show that a subspace is strongly PPT-unextendible if it contains a PPT-definite operator (a positive semidefinite operator whose partial transpose is positive definite). Based on these, we are able to propose a simple criterion for verifying whether a set of bipartite orthogonal quantum states is indistinguishable by PPT operations in the many copy scenario. Utilizing this criterion, we further point out that any entangled pure state and its orthogonal complement cannot be distinguished by PPT operations in the many copy scenario. On the other hand, we investigate that the minimum dimension of strongly PPT-unextendible subspaces in an $m\otimes n$ system is $m+n-1$, which involves a generalization of the result that non-positive-partial-transpose (NPT) subspaces can be as large as any entangled subspace [N. Johnston, Phys. Rev. A 87: 064302 (2013)].

quant-ph↗

arXiv2017

On rank-critical matrix spaces

A matrix space of size $m\times n$ is a linear subspace of the linear space of $m\times n$ matrices over a field $\mathbb{F}$. The rank of a matrix space is defined as the maximal rank over matrices in this space. A matrix space $\mathcal{A}$ is called rank-critical, if any matrix space which properly contains it has rank strictly greater than that of $\mathcal{A}$. In this note, we first exhibit a necessary and sufficient condition for a matrix space $\mathcal{A}$ to be rank-critical, when $\mathbb{F}$ is large enough. This immediately implies the sufficient condition for a matrix space to be rank-critical by Draisma (Bull. Lond. Math. Soc. 38(5):764--776, 2006), albeit requiring the field to be slightly larger. We then study rank-critical spaces in the context of compression and primitive matrix spaces. We first show that every rank-critical matrix space can be decomposed into a rank-critical compression matrix space and a rank-critical primitive matrix space. We then prove, using our necessary and sufficient condition, that the block-diagonal direct sum of two rank-critical matrix spaces is rank-critical if and only if both matrix spaces are primitive, when the field is large enough.

math.RA↗

arXiv2016

Invariance Control Synthesis for Switched Systems: An Interval Analysis Approach

This paper focuses on the invariance control problem for discrete-time switched nonlinear systems. The proposed approach computes controlled invariant sets in a finite number of iterations and directly yields a partition-based invariance controller using the information recorded during the computation. In contrast with Lyapunov-based control methods, this method does not require the subsystems to have common equilibrium points. Algorithms are developed for computing both outer and inner approximations of the maximal controlled invariant sets, which are represented as finite unions of intervals. The general convergence results of interval methods allow us to obtain arbitrarily precise approximations without any stability assumptions. In addition, invariant inner approximations can be computed provided that the switched system satisfies a robustly controlled invariance condition. Under the same condition, we also prove the existence of an invariance controller based on partitions of the state space. Our method is illustrated with three examples drawn from different applications and compared with existing work in the literature.

math.OC↗

arXiv2016

Parallel distinguishability of quantum operations

We find that the perfect distinguishability of two quantum operations by a parallel scheme depends only on an operator subspace generated from their Choi-Kraus operators. We further show that any operator subspace can be obtained from two quantum operations in such a way. This connection enables us to study the parallel distinguishability of operator subspaces directly without explicitly referring to the underlining quantum operations. We obtain a necessary and sufficient condition for the parallel distinguishability of an operator subspace that is either one-dimensional or Hermitian. In both cases the condition is equivalent to the non-existence of positive definite operator in the subspace, and an optimal discrimination protocol is obtained. Finally, we provide more examples to show that the non-existence of positive definite operator is sufficient for many other cases, but in general it is only a necessary condition.

quant-ph↗

arXiv2015

Computing finite abstractions with robustness margins via local reachable set over-approximation

This paper proposes a method to compute finite abstractions that can be used for synthesizing robust hybrid control strategies for nonlinear systems. Most existing methods for computing finite abstractions utilize some global, analytical function to provide bounds on the reachable sets of nonlinear systems, which can be conservative and lead to spurious transitions in the abstract systems. This problem is even more pronounced in the presence of imperfect measurements and modelling uncertainties, where control synthesis can easily become infeasible due to added spurious transitions. To mitigate this problem, we propose to compute finite abstractions with robustness margins by over-approximating the local reachable sets of nonlinear systems. We do so by linearizing the nonlinear dynamics into linear affine systems and keeping track of the linearization error. It is shown that this approach provides tighter approximations and several numerical examples are used to illustrate of effectiveness of the proposed methods.

eess.SY↗