arXiv subjects

Huan Zhang

Publications and source records attributed to Huan Zhang.

At least 181 records · Page 10Linked to original sources

arXiv2023

Global quantum thermometry based on the optimal biased bound

Thermometry is a fundamental parameter estimation problem which is crucial in the development process of natural sciences. One way to solve this problem is to the extensive used local thermometry theory, which makes use of the classical and quantum Cramér-Rao bound as benchmarks of thermometry precision. However, such a thermometry theory can only be used for decreasing temperature fluctuations around a known temperature value and hardly tackle the precision thermometry problem over a wide temperature range. For this reason, we derive two basic bounds on thermometry precision in the global setting and further show their thermometry performance by two specific applications, i.e., noninteracting spin-1/2 gas and a general N-level thermal equilibrium quantum probe.

quant-ph↗

arXiv2023

Can Agents Run Relay Race with Strangers? Generalization of RL to Out-of-Distribution Trajectories

In this paper, we define, evaluate, and improve the ``relay-generalization'' performance of reinforcement learning (RL) agents on the out-of-distribution ``controllable'' states. Ideally, an RL agent that generally masters a task should reach its goal starting from any controllable state of the environment instead of memorizing a small set of trajectories. For example, a self-driving system should be able to take over the control from humans in the middle of driving and must continue to drive the car safely. To practically evaluate this type of generalization, we start the test agent from the middle of other independently well-trained \emph{stranger} agents' trajectories. With extensive experimental evaluation, we show the prevalence of \emph{generalization failure} on controllable states from stranger agents. For example, in the Humanoid environment, we observed that a well-trained Proximal Policy Optimization (PPO) agent, with only 3.9\% failure rate during regular testing, failed on 81.6\% of the states generated by well-trained stranger PPO agents. To improve "relay generalization," we propose a novel method called Self-Trajectory Augmentation (STA), which will reset the environment to the agent's old states according to the Q function during training. After applying STA to the Soft Actor Critic's (SAC) training procedure, we reduced the failure rate of SAC under relay-evaluation by more than three times in most settings without impacting agent performance and increasing the needed number of environment interactions. Our code is available at https://github.com/lan-lc/STA.

cs.LG↗

arXiv2023

On the Robustness of Safe Reinforcement Learning under Observational Perturbations

Safe reinforcement learning (RL) trains a policy to maximize the task reward while satisfying safety constraints. While prior works focus on the performance optimality, we find that the optimal solutions of many safe RL problems are not robust and safe against carefully designed observational perturbations. We formally analyze the unique properties of designing effective observational adversarial attackers in the safe RL setting. We show that baseline adversarial attack techniques for standard RL tasks are not always effective for safe RL and propose two new approaches - one maximizes the cost and the other maximizes the reward. One interesting and counter-intuitive finding is that the maximum reward attack is strong, as it can both induce unsafe behaviors and make the attack stealthy by maintaining the reward. We further propose a robust training framework for safe RL and evaluate it via comprehensive experiments. This paper provides a pioneer work to investigate the safety and robustness of RL under observational attacks for future safe RL studies. Code is available at: \url{https://github.com/liuzuxin/safe-rl-robustness}

cs.LG↗

arXiv2022

Sharpness-Aware Minimization with Dynamic Reweighting

Deep neural networks are often overparameterized and may not easily achieve model generalization. Adversarial training has shown effectiveness in improving generalization by regularizing the change of loss on top of adversarially chosen perturbations. The recently proposed sharpness-aware minimization (SAM) algorithm conducts adversarial weight perturbation, encouraging the model to converge to a flat minima. SAM finds a common adversarial weight perturbation per-batch. Although per-instance adversarial weight perturbations are stronger adversaries and can potentially lead to better generalization performance, their computational cost is very high and thus it is impossible to use per-instance perturbations efficiently in SAM. In this paper, we tackle this efficiency bottleneck and propose sharpness-aware minimization with dynamic reweighting (delta-SAM). Our theoretical analysis motivates that it is possible to approach the stronger, per-instance adversarial weight perturbations using reweighted per-batch weight perturbations. delta-SAM dynamically reweights perturbation within each batch according to the theoretically principled weighting factors, serving as a good approximation to per-instance perturbation. Experiments on various natural language understanding tasks demonstrate the effectiveness of delta-SAM.

cs.LG↗

arXiv2022

General Cutting Planes for Bound-Propagation-Based Neural Network Verification

Bound propagation methods, when combined with branch and bound, are among the most effective methods to formally verify properties of deep neural networks such as correctness, robustness, and safety. However, existing works cannot handle the general form of cutting plane constraints widely accepted in traditional solvers, which are crucial for strengthening verifiers with tightened convex relaxations. In this paper, we generalize the bound propagation procedure to allow the addition of arbitrary cutting plane constraints, including those involving relaxed integer variables that do not appear in existing bound propagation formulations. Our generalized bound propagation method, GCP-CROWN, opens up the opportunity to apply general cutting plane methods for neural network verification while benefiting from the efficiency and GPU acceleration of bound propagation methods. As a case study, we investigate the use of cutting planes generated by off-the-shelf mixed integer programming (MIP) solver. We find that MIP solvers can generate high-quality cutting planes for strengthening bound-propagation-based verifiers using our new formulation. Since the branching-focused bound propagation procedure and the cutting-plane-focused MIP solver can run in parallel utilizing different types of hardware (GPUs and CPUs), their combination can quickly explore a large number of branches with strong cutting planes, leading to strong verification performance. Experiments demonstrate that our method is the first verifier that can completely solve the oval20 benchmark and verify twice as many instances on the oval21 benchmark compared to the best tool in VNN-COMP 2021, and also noticeably outperforms state-of-the-art verifiers on a wide range of benchmarks. GCP-CROWN is part of the $α,\!β$-CROWN verifier, the VNN-COMP 2022 winner. Code is available at http://PaperCode.cc/GCP-CROWN

cs.LG↗

arXiv2022

Understanding and eliminating spurious modes in variational Monte Carlo using collective variables

The use of neural network parametrizations to represent the ground state in variational Monte Carlo (VMC) calculations has generated intense interest in recent years. However, as we demonstrate in the context of the periodic Heisenberg spin chain, this approach can produce unreliable wave function approximations. One of the most obvious signs of failure is the occurrence of random, persistent spikes in the energy estimate during training. These energy spikes are caused by regions of configuration space that are over-represented by the wave function density, which are called ``spurious modes'' in the machine learning literature. After exploring these spurious modes in detail, we demonstrate that a collective-variable-based penalization yields a substantially more robust training procedure, preventing the formation of spurious modes and improving the accuracy of energy estimates. Because the penalization scheme is cheap to implement and is not specific to the particular model studied here, it can be extended to other applications of VMC where a reasonable choice of collective variable is available.

physics.chem-ph↗

arXiv2022

Are AlphaZero-like Agents Robust to Adversarial Perturbations?

The success of AlphaZero (AZ) has demonstrated that neural-network-based Go AIs can surpass human performance by a large margin. Given that the state space of Go is extremely large and a human player can play the game from any legal state, we ask whether adversarial states exist for Go AIs that may lead them to play surprisingly wrong actions. In this paper, we first extend the concept of adversarial examples to the game of Go: we generate perturbed states that are ``semantically'' equivalent to the original state by adding meaningless moves to the game, and an adversarial state is a perturbed state leading to an undoubtedly inferior action that is obvious even for Go beginners. However, searching the adversarial state is challenging due to the large, discrete, and non-differentiable search space. To tackle this challenge, we develop the first adversarial attack on Go AIs that can efficiently search for adversarial states by strategically reducing the search space. This method can also be extended to other board games such as NoGo. Experimentally, we show that the actions taken by both Policy-Value neural network (PV-NN) and Monte Carlo tree search (MCTS) can be misled by adding one or two meaningless stones; for example, on 58\% of the AlphaGo Zero self-play games, our method can make the widely used KataGo agent with 50 simulations of MCTS plays a losing action by adding two meaningless stones. We additionally evaluated the adversarial examples found by our algorithm with amateur human Go players and 90\% of examples indeed lead the Go agent to play an obviously inferior action. Our code is available at \url{https://PaperCode.cc/GoAttack}.

cs.AI↗

arXiv2022

Simultaneous multiple angular displacement estimation precision enhanced by the intramode correlation

The angular displacement estimation is one of significant branches of quantum parameter estimation. However, most of the studies have focused on the single-angular displacement estimation, while the multiple angular displacement estimation in ideal and noisy scenarios is still elusive. In this paper, we investigate the simultaneous multiple angular displacement estimation based on an orbital angular momentum (OAM), together with inputting (d + 1)-mode NOON-like states as the probe state. By revealing the role of the intramode correlation of the probe state, this allows us to give a reasonable explanation for the corresponding quantum Cramer-Rao bound (QCRB) behaviors with and without photon losses. Our analyses suggest that the QCRB for the multiple angular displacement estimation is always positively related to the intramode correlation, especially for the multimode entangled squeezed vacuum state showing the best performance compared to another probe state. More importantly, strengthening the robustness of multiple angular-displacement estimation systems can be achieved by increasing the OAM quantum number.

quant-ph↗

arXiv2022

Quantum multiparameter estimation with multi-mode photon catalysis entangled squeezed state

We propose a method to generate the multi-mode entangled catalysis squeezed vacuum states (MECSVS) by embedding the cross-Kerr nonlinear medium into the Mach-Zehnder interferometer. This method realizes the exchange of quantum states between different modes based on Fredkin gate. In addition, we study the MECSVS as the probe state of multi-arm optical interferometer to realize multi-phase simultaneous estimation. The results show that the quantum Cramer-Rao bound (QCRB) of phase estimation can be improved by increasing the number of catalytic photons or decreasing the transmissivity of the optical beam splitter using for photon catalysis. In addition, we also show that even if there is photon loss, the QCRB of our photon catalysis scheme is lower than that of the ideal entangled squeezed vacuum states (ESVS), which shows that by performing the photon catalytic operation is more robust against photon loss than that without the catalytic operation. The results here can find important applications in quantum metrology for multiparatmeter estimation.

quant-ph↗

arXiv2022

Efficiently Computing Local Lipschitz Constants of Neural Networks via Bound Propagation

Lipschitz constants are connected to many properties of neural networks, such as robustness, fairness, and generalization. Existing methods for computing Lipschitz constants either produce relatively loose upper bounds or are limited to small networks. In this paper, we develop an efficient framework for computing the $\ell_\infty$ local Lipschitz constant of a neural network by tightly upper bounding the norm of Clarke Jacobian via linear bound propagation. We formulate the computation of local Lipschitz constants with a linear bound propagation process on a high-order backward graph induced by the chain rule of Clarke Jacobian. To enable linear bound propagation, we derive tight linear relaxations for specific nonlinearities in Clarke Jacobian. This formulate unifies existing ad-hoc approaches such as RecurJac, which can be seen as a special case of ours with weaker relaxations. The bound propagation framework also allows us to easily borrow the popular Branch-and-Bound (BaB) approach from neural network verification to further tighten Lipschitz constants. Experiments show that on tiny models, our method produces comparable bounds compared to exact methods that cannot scale to slightly larger models; on larger models, our method efficiently produces tighter results than existing relaxed or naive methods, and our method scales to much larger practical models that previous works could not handle. We also demonstrate an application on provable monotonicity analysis. Code is available at https://github.com/shizhouxing/Local-Lipschitz-Constants.

cs.LG↗

arXiv2022

Outage Performance Analysis of Type-I HARQ Aided V2V Communications

Vehicle-to-vehicle (V2V) communications under dense urban environments usually experience severe keyhole fading effect, especially for multi-input multi-output (MIMO) channels, which degrades the capacity and outage performance due to the rank deficiency. To avoid these, the integration of MIMO and Type-I hybrid automatic repeat request (HARQ) is proposed to assist V2V communications in this paper. By using the Meijer G-function, the outage probability of the proposed V2V communications system is derived in closed form. With the result, meaningful insights are gained by conducting the asymptotic outage analysis. Specifically, it is revealed that full-time diversity order can be achieved, while full spatial diversity order is unreachable as compared to Type-I HARQ aided MIMO systems without keyhole effect. Moreover, we prove that the asymptotic outage probability is a monotonically increasing and convex function of the transmission rate. Finally, the analytical results are validated through extensive numerical experiments.

cs.IT↗

arXiv2022

Outage Performance Analysis of HARQ-Aided Multi-RIS Systems

Reconfigurable intelligent surface (RIS) has recently attracted a spurt of interest due to its innate advantages over Massive MIMO on power consumption. In this paper, we study the outage performance of multi-RIS system with the help of hybrid automatic repeat request (HARQ) to improve the RIS system reliability, where the destination received channels are modeled by Rician fading and the phase shift setting only depends on the line-of-sight (LoS) component. Both the exact and asymptotic outage probabilities under Type-I HARQ and HARQ with chase combining (HARQ-CC) schemes are derived. Particulary, the tractable asymptotic results empower us to derive meaningful insights for HARQ-aided multi-RIS system. On the one hand, we find that both the Type-I and the HARQ-CC schemes can achieve full diversity that is equal to the maximal number of HARQ rounds. On the other hand, the closed-form expression of the optimal phase shift setting with respect to outage probability minimization is obtained. The optimal solution indicates that the reflecting link direction should be consistent with direct link LoS component. Finally, the analytical results are validated by Monte-Carlo simulations.

cs.IT↗

arXiv2022

Performance Analysis of MIMO-HARQ Assisted V2V Communications With Keyhole Effect

Vehicle-to-vehicle (V2V) communications under dense urban environments usually experience severe keyhole fading effect especially for multi-input multi-output (MIMO) channels, which degrades the capacity and outage performance due to the rank deficiency. To avoid these, the integration of MIMO and hybrid automatic repeat request (HARQ) is proposed to assist V2V communications in this paper. By using the methods of integral transforms, the outage probabilities are derived in closed-form for different HARQ-assisted schemes, including Type I-HARQ, HARQ with chase combining (HARQ-CC), and HARQ with incremental redundancy (HARQ-IR). With the results, meaningful insights are gained by conducting the asymptotic outage analysis. Specifically, it is revealed that full time diversity order can be achieved, while full spatial diversity order is unreachable as compared to MIMO-HARQ systems without keyhole effect. Moreover, we prove that the asymptotic outage probability is a monotonically increasing and convex function of the transmission rate. More importantly, although HARQ-IR performs better than HARQ-CC owing to its higher coding complexity, this advantage becomes negligible in the large-scale array regime. Finally, the numerical results are verified by Monte-Carlo simulations along with some in-depth discussions.

cs.IT↗

arXiv2022

On the Performance of Variable-Rate HARQ-IR over Beckmann Fading Channels

This paper thoroughly investigates the performance of variable-rate hybrid automatic repeat request with incremental redundancy (HARQ-IR) over Beckmann fading channels, where Beckmann channel model is used to characterize the impacts of line-of-sight (LOS) path, time correlation between fading channels and unequal means & variances between the in-phase and the quadrature components. The intricate channel model and variable-rate transmission strategy significantly complicate the performance analysis. The complex form of the joint distribution of channel coefficients makes the exact analysis rather difficult. Hence, the asymptotic analysis of HARQ-IR is conducted to obtain tractable results. The outage probability and the long term average throughput (LTAT) are then derived in closed-form with clear insights. Furthermore, the simplicity of the obtained expressions enables the maximization of the LTAT given an outage constraint through choosing appropriate transmission rates. Particularly, the increasing monotonicity and the convexity of outage probability with respect to transmission rates allow us to solve the problem suboptimally by combining together alternately iterating optimization and concave fractional programming, and the proposed suboptimal algorithm has a lower computational complexity than the exhaustive search algorithm.

cs.IT↗

arXiv2022

Outage Performance Analysis of Full-Correlated Rayleigh MIMO Channels

The outage performance of multiple-input multiple-output (MIMO) technique has received intensive attention to meet the stringent requirement of reliable communications for 5G applications, e.g., mission-critical machine-type communication (cMTC). To account for spatial correlation effects at both transmit and receive sides, the full-correlated Rayleigh MIMO fading channels are modeled according to Kronecker correlation structure in this paper. The outage probability is expressed as a weighted sum of the generalized Fox's H functions. The simple analytical result empowers asymptotic outage analysis at high signal-to-noise ratio (SNR), which not only reveal helpful insights into understanding the behavior of fading effects, but also offer useful design guideline for MIMO configurations. Particularly, the negative impact of the spatial correlation on the outage probability is revealed by using the concept of majorization, and the asymptotic outage probability is proved to be a monotonically increasing and convex function of the transmission rate. In the end, the analytical results are validated through extensive numerical experiments.

cs.IT↗

arXiv2022

Phase estimation of Mach-Zehnder interferometer via Laguerre excitation squeezed state

Quantum metrology has an important role in the fields of quantum optics and quantum information processing. Here we introduce a kind of non-Gaussian state, Laguerre excitation squeezed state as input of traditional Mach-Zehnder interferometer to examine phase estimation in realistic case. We consider the effects of both internal and external losses on phase estimation by using quantum Fisher information and parity detection. It is shown that the external loss presents a bigger effect than the internal one. The phase sensitivity and the quantum Fisher information can be improved by increasing the photon number and even surpass the ideal phase sensitivity by two-mode squeezed vacuum in a certain region of phase shift for realistic case.

quant-ph↗

arXiv2022

Deep Image Destruction: Vulnerability of Deep Image-to-Image Models against Adversarial Attacks

Recently, the vulnerability of deep image classification models to adversarial attacks has been investigated. However, such an issue has not been thoroughly studied for image-to-image tasks that take an input image and generate an output image (e.g., colorization, denoising, deblurring, etc.) This paper presents comprehensive investigations into the vulnerability of deep image-to-image models to adversarial attacks. For five popular image-to-image tasks, 16 deep models are analyzed from various standpoints such as output quality degradation due to attacks, transferability of adversarial examples across different tasks, and characteristics of perturbations. We show that unlike image classification tasks, the performance degradation on image-to-image tasks largely differs depending on various factors, e.g., attack methods and task objectives. In addition, we analyze the effectiveness of conventional defense methods used for classification models in improving the robustness of the image-to-image models.

cs.CV↗

arXiv2022

Faraday patterns in spin-orbit coupled Bose-Einstein condensates

We study the Faraday patterns generated by spin-orbit-coupling induced parametric resonance in a spinor Bose-Einstein condensate with repulsive interaction. The collective elementary excitations of the Bose-Einstein condensate, including density waves and spin waves, are coupled as the result of the Raman-induced spin-orbit coupling and a quench of the relative phase of two Raman lasers without the modulation of any of the system's parameters. We observed several higher parametric resonance tongues at integer multiples of the driving frequency and investigated the interplay between Faraday instabilities and modulation instabilities when we quench the spin-orbit-coupled Bose-Einstein condensate from zero-momentum phase to plane-wave phase. If the detuning is equal to zero, the wave number of combination resonance barely changes as the strength of spin-orbit coupling increases. If the detuning is not equal to zero after a quench, a single combination resonance tongue will split into two parts.

cond-mat.quant-gas↗