Most existing neural architecture search (NAS) algorithms are dedicated to the downstream tasks, e.g., image classification in computer vision. However, extensive experiments have shown that, prominent neural architectures, such as ResNet in computer
vision and LSTM in natural language processing, are generally good at extracting patterns from the input data and perform well on different downstream tasks. These observations inspire us to ask: Is it necessary to use the performance of specific downstream tasks to evaluate and search for good neural architectures? Can we perform NAS effectively and efficiently while being agnostic to the downstream task? In this work, we attempt to affirmatively answer the above two questions and improve the state-of-the-art NAS solution by proposing a novel and generic NAS framework, termed Generic NAS (GenNAS). GenNAS does not use task-specific labels but instead adopts textit{regression} on a set of manually designed synthetic signal bases for architecture evaluation. Such a self-supervised regression task can effectively evaluate the intrinsic power of an architecture to capture and transform the input signal patterns, and allow more sufficient usage of training samples. We then propose an automatic task search to optimize the combination of synthetic signals using limited downstream-task-specific labels, further improving the performance of GenNAS. We also thoroughly evaluate GenNASs generality and end-to-end NAS performance on all search spaces, which outperforms almost all existing works with significant speedup.
E-commerce companies have to face abnormal sellers who sell potentially-risky products. Typically, the risk can be identified by jointly considering product content (e.g., title and image) and seller behavior. This work focuses on behavior feature ex
traction as behavior sequences can provide valuable clues for the risk discovery by reflecting the sellers operation habits. Traditional feature extraction techniques heavily depend on domain experts and adapt poorly to new tasks. In this paper, we propose a self-supervised method InfoBehavior to automatically extract meaningful representations from ultra-long raw behavior sequences instead of the costly feature selection procedure. InfoBehavior utilizes Bidirectional Transformer as feature encoder due to its excellent capability in modeling long-term dependency. However, it is intractable for commodity GPUs because the time and memory required by Transformer grow quadratically with the increase of sequence length. Thus, we propose a hierarchical grouping strategy to aggregate ultra-long raw behavior sequences to length-processable high-level embedding sequences. Moreover, we introduce two types of pretext tasks. Sequence-related pretext task defines a contrastive-based training objective to correctly select the masked-out coarse-grained/fine-grained behavior sequences against other distractor behavior sequences; Domain-related pretext task designs a classification training objective to correctly predict the domain-specific statistical results of anomalous behavior. We show that behavior representations from the pre-trained InfoBehavior can be directly used or integrated with features from other side information to support a wide range of downstream tasks. Experimental results demonstrate that InfoBehavior significantly improves the performance of Product Risk Management and Intellectual Property Protection.
Adversarial examples are perturbed inputs which can cause a serious threat for machine learning models. Finding these perturbations is such a hard task that we can only use the iterative methods to traverse. For computational efficiency, recent works
use adversarial generative networks to model the distribution of both the universal or image-dependent perturbations directly. However, these methods generate perturbations only rely on input images. In this work, we propose a more general-purpose framework which infers target-conditioned perturbations dependent on both input image and target label. Different from previous single-target attack models, our model can conduct target-conditioned attacks by learning the relations of attack target and the semantics in image. Using extensive experiments on the datasets of MNIST and CIFAR10, we show that our method achieves superior performance with single target attack models and obtains high fooling rates with small perturbation norms.
We analyse the precision limits for simultaneous estimation of a pair of conjugate parameters in a displacement channel using Gaussian probes. Having a set of squeezed states as an initial resource, we compute the Holevo Cramer-Rao bound to investiga
te the best achievable estimation precisions if only passive linear operations are allowed to be performed on the resource prior to probing the channel. The analysis reveals the optimal measurement scheme and allows us to quantify the best precision for one parameter when the precision of the second conjugate parameter is fixed. To estimate the conjugate parameter pair with equal precision, our analysis shows that the optimal probe is obtained by combining two squeezed states with orthogonal squeezing quadratures on a 50:50 beam splitter. If different importance are attached to each parameter, then the optimal mixing ratio is no longer 50:50. Instead it follows a simple function of the available squeezing and the relative importance between the two parameters.
Deep Neural Networks (DNNs) are known to be vulnerable to the maliciously generated adversarial examples. To detect these adversarial examples, previous methods use artificially designed metrics to characterize the properties of textit{adversarial su
bspaces} where adversarial examples lie. However, we find these methods are not working in practical attack detection scenarios. Because the artificially defined features are lack of robustness and show limitation in discriminative power to detect strong attacks. To solve this problem, we propose a novel adversarial detection method which identifies adversaries by adaptively learning reasonable metrics to characterize adversarial subspaces. As auxiliary context information, textit{k} nearest neighbors are used to represent the surrounded subspace of the detected sample. We propose an innovative model called Neighbor Context Encoder (NCE) to learn from textit{k} neighbors context and infer if the detected sample is normal or adversarial. We conduct thorough experiment on CIFAR-10, CIFAR-100 and ImageNet dataset. The results demonstrate that our approach surpasses all existing methods under three settings: textit{attack-aware black-box detection}, textit{attack-unaware black-box detection} and textit{white-box detection}.
Field-orthogonal temporal mode analysis of optical fields is recently developed for a new framework of quantum information science. But so far, the exact profiles of the temporal modes are not known, which makes it difficult to achieve mode selection
and de-multiplexing. Here, we report a novel method that measures directly the exact form of the temporal modes. This in turn enables us to make mode-orthogonal homodyne detection with mode-matched local oscillators. We apply the method to a pulse-pumped, specially engineered fiber parametric amplifier and demonstrate temporally multiplexed multi-dimensional quantum entanglement of continuous variables in telecom wavelength. The temporal mode characterization technique can be generalized to other pulse-excited systems to find their eigen modes for multiplexing in temporal domain.
The task of Language-Based Image Editing (LBIE) aims at generating a target image by editing the source image based on the given language description. The main challenge of LBIE is to disentangle the semantics in image and text and then combine them
to generate realistic images. Therefore, the editing performance is heavily dependent on the learned representation. In this work, conditional generative adversarial network (cGAN) is utilized for LBIE. We find that existing conditioning methods in cGAN lack of representation power as they cannot learn the second-order correlation between two conditioning vectors. To solve this problem, we propose an improved conditional layer named Bilinear Residual Layer (BRL) to learning more powerful representations for LBIE task. Qualitative and quantitative comparisons demonstrate that our method can generate images with higher quality when compared to previous LBIE techniques.
Quantum entanglement is a resource in quantum metrology that can be distributed to two orthogonal physical quantities for the enhancement of their joint measurement sensitivity, as demonstrated in quantum dense metrology. On the other hand, we can al
so devote all the quantum resource to phase measurement only for optimum measurement sensitivity. Here, we experimentally implement a dual-beam scheme in an SU(1,1) interferometer for the optimum phase measurement sensitivity. We demonstrate a 3.9-dB improvement in signal-to-noise ratio over the optimum classical method and this is 3-dB better than the traditional single-beam scheme. Furthermore, such a scheme also realizes a quantum optical tap of quantum entangled fields and has the full advantages of an SU(1,1) interferometer for practical applications in quantum metrology and quantum information.
We show here that the Hamiltonian for an electronic system may be written exactly in terms of fluctuation operators that transition constituent fragments between internally correlated states, accounting rigorously for inter-fragment electron exchange
and charge transfer. Familiar electronic structure approaches can be applied to the renormalized Hamiltonian. For efficiency, the basis for each fragment can be truncated, removing high-energy local arrangements of electrons from consideration, and effectively defining collective coordinates for the fragments. For a large number of problems (especially for non-covalently interacting fragments), this has the potential to fold the majority of electron correlation into the effective Hamiltonian, and it should provide a robust approach to incorporating difficult electronic structure problems into large systems. The number of terms in the exactly transformed Hamiltonian formally scales quartically with system size, but this can be reduced to quadratic in the mesoscopic regime, to within an arbitrary error tolerance. Finally, all but a linear-scaling number of these terms may be efficiently decomposed in terms of electrostatic interactions between a linear-scaling number of pre-computed transition densities. In a companion article, this formalism is applied to an excitonic variant of coupled-cluster theory.
A variant of coupled-cluster theory is described here, wherein the degrees of freedom are fluctuations of fragments between internally correlated states. The effects of intra-fragment correlation on the inter-fragment interaction are pre-computed and
permanently folded into an effective Hamiltonian, thus avoiding redundant evaluations of local relaxations associated with coupled fluctuations. A companion article shows that a low-scaling step may be used to cast the electronic Hamiltonians of real systems into the form required. Two proof-of-principle demonstrations are presented here for non-covalent interactions. One uses harmonic oscillators, for which accuracy and algorithm structure can be carefully controlled in comparisons. The other uses small electronic systems (Be atoms) to demonstrate compelling accuracy and efficiency, also when inter-fragment electron exchange and charge transfer must be handled. Since the cost of the global calculation does not depend directly on the correlation models used for the fragments, this should provide a way to incorporate difficult electronic structure problems into large systems. This framework opens a promising path for building tunable, systematically improvable methods to capture properties of systems interacting with a large number of other systems. The extension to excited states is also straightforward.
