Search

One major problem in black-box adversarial attacks is the high query complexity in the hard-label attack setting, where only the top-1 predicted label is available. In this paper, we propose a novel geometric-based approach called Tangent Attack (TA), which identifies an optimal tangent point of a virtual hemisphere located on the decision boundary to reduce the distortion of the attack. Assuming the decision boundary is locally flat, we theoretically prove that the minimum $\ell_2$ distortion can be obtained by reaching the decision boundary along the tangent line passing through such tangent point in each iteration. To improve the robustness of our method, we further propose a generalized method which replaces the hemisphere with a semi-ellipsoid to adapt to curved decision boundaries. Our approach is free of pre-training. Extensive experiments conducted on the ImageNet and CIFAR-10 datasets demonstrate that our approach can consume only a small number of queries to achieve the low-magnitude distortion. The implementation source code is released online at https://github.com/machanic/TangentAttack., Comment: Accepted at NeurIPS 2021. The missing square term in Eqn.(13), as well as many other mistakes of the previous version, have been fixed in the current version

We study a common delivery problem encountered in nowadays online food-ordering platforms: Customers order dishes online, and the restaurant delivers the food after receiving the order. Specifically, we study a problem where $k$ vehicles of capacity $c$ are serving a set of requests ordering food from one restaurant. After a request arrives, it can be served by a vehicle moving from the restaurant to its delivery location. We are interested in serving all requests while minimizing the maximum flow-time, i.e., the maximum time length a customer waits to receive his/her food after submitting the order. We show that the problem is hard in both offline and online settings: There is a hardness of approximation of $\Omega(n)$ for the offline problem, and a lower bound of $\Omega(n)$ on the competitive ratio of any online algorithm, where $n$ is number of points in the metric. Our main result is an $O(1)$-competitive online algorithm for the uncapaciated (i.e, $c = \infty$) food delivery problem on tree metrics. Then we consider the speed-augmentation model. We develop an exponential time $(1+\epsilon)$-speeding $O(1/\epsilon)$-competitive algorithm for any $\epsilon > 0$. A polynomial time algorithm can be obtained with a speeding factor of $\alpha_{TSP}+ \epsilon$ or $\alpha_{CVRP}+ \epsilon$, depending on whether the problem is uncapacitated. Here $\alpha_{TSP}$ and $\alpha_{CVRP}$ are the best approximation factors for the traveling salesman (TSP) and capacitated vehicle routing (CVRP) problems respectively. We complement the results with some negative ones.

This paper presents the SPIIR pipeline used for public alerts during the third advanced LIGO and Virgo observation run (O3 run). The SPIIR pipeline uses infinite impulse response (IIR) filters to perform extremely low-latency matched filtering and this process is further accelerated with graphics processing units (GPUs). It is the first online pipeline to select candidates from multiple detectors using a coherent statistic based on the maximum network likelihood ratio statistic principle. Here we simplify the derivation of this statistic using the singular-value-decomposition (SVD) technique and show that single-detector signal-to-noise ratios from matched filtering can be directly used to construct the statistic for each sky direction. Coherent searches are in general more computationally challenging than coincidence searches due to extra search over sky direction parameters. The search over sky directions follows an embarrassing parallelization paradigm and has been accelerated using GPUs. The detection performance is reported using a segment of public data from LIGO-Virgo's second observation run. We demonstrate that the median latency of the SPIIR pipeline is less than 9 seconds, and present an achievable roadmap to reduce the latency to less than 5 seconds. During the O3 online run, SPIIR registered triggers associated with 38 of the 56 non-retracted public alerts. The extreme low-latency nature makes it a competitive choice for joint time-domain observations, and offers the tantalizing possibility of making public alerts prior to the merger phase of binary coalescence systems involving at least one neutron star., Comment: 17 pages, 15 figures

In this paper, we study the problem of estimating latent variable models with arbitrarily corrupted samples in high dimensional space ({\em i.e.,} $d\gg n$) where the underlying parameter is assumed to be sparse. Specifically, we propose a method called Trimmed (Gradient) Expectation Maximization which adds a trimming gradients step and a hard thresholding step to the Expectation step (E-step) and the Maximization step (M-step), respectively. We show that under some mild assumptions and with an appropriate initialization, the algorithm is corruption-proofing and converges to the (near) optimal statistical rate geometrically when the fraction of the corrupted samples $\epsilon$ is bounded by $ \tilde{O}(\frac{1}{\sqrt{n}})$. Moreover, we apply our general framework to three canonical models: mixture of Gaussians, mixture of regressions and linear regression with missing covariates. Our theory is supported by thorough numerical results., Comment: Accepted at Machine Learning

Recently, many machine learning and statistical models such as non-linear regressions, the Single Index, Multi-index, Varying Coefficient Index Models and Two-layer Neural Networks can be reduced to or be seen as a special case of a new model which is called the \textit{Stochastic Linear Combination of Non-linear Regressions} model. However, due to the high non-convexity of the problem, there is no previous work study how to estimate the model. In this paper, we provide the first study on how to estimate the model efficiently and scalably. Specifically, we first show that with some mild assumptions, if the variate vector $x$ is multivariate Gaussian, then there is an algorithm whose output vectors have $\ell_2$-norm estimation errors of $O(\sqrt{\frac{p}{n}})$ with high probability, where $p$ is the dimension of $x$ and $n$ is the number of samples. The key idea of the proof is based on an observation motived by the Stein's lemma. Then we extend our result to the case where $x$ is bounded and sub-Gaussian using the zero-bias transformation, which could be seen as a generalization of the classic Stein's lemma. We also show that with some additional assumptions there is an algorithm whose output vectors have $\ell_\infty$-norm estimation errors of $O(\frac{1}{\sqrt{p}}+\sqrt{\frac{p}{n}})$ with high probability. We also provide a concrete example to show that there exists some link function which satisfies the previous assumptions. Finally, for both Gaussian and sub-Gaussian cases we propose a faster sub-sampling based algorithm and show that when the sub-sample sizes are large enough then the estimation errors will not be sacrificed by too much. Experiments for both cases support our theoretical results. To the best of our knowledge, this is the first work that studies and provides theoretical guarantees for the stochastic linear combination of non-linear regressions model., Comment: This paper is a substantially extended version of our previous work appeared in AAAI'20