-
Dagger Behind Smile: Fool LLMs with a Happy Ending Story
Authors:
Xurui Song,
Zhixin Xie,
Shuo Huai,
Jiayi Kong,
Jun Luo
Abstract:
The wide adoption of Large Language Models (LLMs) has attracted significant attention from $\textit{jailbreak}$ attacks, where adversarial prompts crafted through optimization or manual design exploit LLMs to generate malicious contents. However, optimization-based attacks have limited efficiency and transferability, while existing manual designs are either easily detectable or demand intricate in…
▽ More
The wide adoption of Large Language Models (LLMs) has attracted significant attention from $\textit{jailbreak}$ attacks, where adversarial prompts crafted through optimization or manual design exploit LLMs to generate malicious contents. However, optimization-based attacks have limited efficiency and transferability, while existing manual designs are either easily detectable or demand intricate interactions with LLMs. In this paper, we first point out a novel perspective for jailbreak attacks: LLMs are more responsive to $\textit{positive}$ prompts. Based on this, we deploy Happy Ending Attack (HEA) to wrap up a malicious request in a scenario template involving a positive prompt formed mainly via a $\textit{happy ending}$, it thus fools LLMs into jailbreaking either immediately or at a follow-up malicious request.This has made HEA both efficient and effective, as it requires only up to two turns to fully jailbreak LLMs. Extensive experiments show that our HEA can successfully jailbreak on state-of-the-art LLMs, including GPT-4o, Llama3-70b, Gemini-pro, and achieves 88.79\% attack success rate on average. We also provide quantitative explanations for the success of HEA.
△ Less
Submitted 16 February, 2025; v1 submitted 19 January, 2025;
originally announced January 2025.
-
Efficient Deep Learning Infrastructures for Embedded Computing Systems: A Comprehensive Survey and Future Envision
Authors:
Xiangzhong Luo,
Di Liu,
Hao Kong,
Shuo Huai,
Hui Chen,
Guochu Xiong,
Weichen Liu
Abstract:
Deep neural networks (DNNs) have recently achieved impressive success across a wide range of real-world vision and language processing tasks, spanning from image classification to many other downstream vision tasks, such as object detection, tracking, and segmentation. However, previous well-established DNNs, despite being able to maintain superior accuracy, have also been evolving to be deeper an…
▽ More
Deep neural networks (DNNs) have recently achieved impressive success across a wide range of real-world vision and language processing tasks, spanning from image classification to many other downstream vision tasks, such as object detection, tracking, and segmentation. However, previous well-established DNNs, despite being able to maintain superior accuracy, have also been evolving to be deeper and wider and thus inevitably necessitate prohibitive computational resources for both training and inference. This trend further enlarges the computational gap between computation-intensive DNNs and resource-constrained embedded computing systems, making it challenging to deploy powerful DNNs upon real-world embedded computing systems towards ubiquitous embedded intelligence. To alleviate the above computational gap and enable ubiquitous embedded intelligence, we, in this survey, focus on discussing recent efficient deep learning infrastructures for embedded computing systems, spanning from training to inference, from manual to automated, from convolutional neural networks to transformers, from transformers to vision transformers, from vision models to large language models, from software to hardware, and from algorithms to applications. Specifically, we discuss recent efficient deep learning infrastructures for embedded computing systems from the lens of (1) efficient manual network design for embedded computing systems, (2) efficient automated network design for embedded computing systems, (3) efficient network compression for embedded computing systems, (4) efficient on-device learning for embedded computing systems, (5) efficient large language models for embedded computing systems, (6) efficient deep learning software and hardware for embedded computing systems, and (7) efficient intelligent applications for embedded computing systems.
△ Less
Submitted 2 November, 2024;
originally announced November 2024.
-
Tantalum airbridges for scalable superconducting quantum processors
Authors:
Kunliang Bu,
Sainan Huai,
Zhenxing Zhang,
Dengfeng Li,
Yuan Li,
Jingjing Hu,
Xiaopei Yang,
Maochun Dai,
Tianqi Cai,
Yi-Cong Zheng,
Shengyu Zhang
Abstract:
The unique property of tantalum (Ta), particularly its long coherent lifetime in superconducting qubits and its exceptional resistance to both acid and alkali, makes it promising for superconducting quantum processors. It is a notable advantage to achieve high-performance quantum processors with neat and unified fabrication of all circuit elements, including coplanar waveguides (CPW), qubits, and…
▽ More
The unique property of tantalum (Ta), particularly its long coherent lifetime in superconducting qubits and its exceptional resistance to both acid and alkali, makes it promising for superconducting quantum processors. It is a notable advantage to achieve high-performance quantum processors with neat and unified fabrication of all circuit elements, including coplanar waveguides (CPW), qubits, and airbridges, on the tantalum film-based platform. Here, we propose a reliable tantalum airbridges with separate or fully-capped structure fabricated via a novel lift-off method, where a barrier layer with aluminium (Al) film is first introduced to separate two layers of photoresist and then etched away before the deposition of tantalum film, followed by cleaning with piranha solution to remove the residual photoresist on the chip. We characterize such tantalum airbridges as the control line jumpers, the ground plane crossovers and even coupling elements. They exhibit excellent connectivity, minimal capacitive loss, effectively suppress microwave and flux crosstalk and offer high freedom of coupling. Besides, by presenting a surface-13 tunable coupling superconducting quantum processor with median $T_1$ reaching above 100 $μ$s, the overall adaptability of tantalum airbridges is verified. The median single-qubit gate fidelity shows a tiny decrease from about 99.95% for the isolated Randomized Benchmarking to 99.94% for the simultaneous one. This fabrication method, compatible with all known superconducting materials, requires mild conditions of film deposition compared with the commonly used etching and grayscale lithography. Meanwhile, the experimental achievement of non-local coupling with controlled-Z (CZ) gate fidelity exceeding 99.2% may further facilitate qLDPC codes, laying a foundation for scalable quantum computation and quantum error correction with entirely tantalum elements.
△ Less
Submitted 7 January, 2024;
originally announced January 2024.
-
Do Not DeepFake Me: Privacy-Preserving Neural 3D Head Reconstruction Without Sensitive Images
Authors:
Jiayi Kong,
Xurui Song,
Shuo Huai,
Baixin Xu,
Jun Luo,
Ying He
Abstract:
While 3D head reconstruction is widely used for modeling, existing neural reconstruction approaches rely on high-resolution multi-view images, posing notable privacy issues. Individuals are particularly sensitive to facial features, and facial image leakage can lead to many malicious activities, such as unauthorized tracking and deepfake. In contrast, geometric data is less susceptible to misuse d…
▽ More
While 3D head reconstruction is widely used for modeling, existing neural reconstruction approaches rely on high-resolution multi-view images, posing notable privacy issues. Individuals are particularly sensitive to facial features, and facial image leakage can lead to many malicious activities, such as unauthorized tracking and deepfake. In contrast, geometric data is less susceptible to misuse due to its complex processing requirements, and absence of facial texture features. In this paper, we propose a novel two-stage 3D facial reconstruction method aimed at avoiding exposure to sensitive facial information while preserving detailed geometric accuracy. Our approach first uses non-sensitive rear-head images for initial geometry and then refines this geometry using processed privacy-removed gradient images. Extensive experiments show that the resulting geometry is comparable to methods using full images, while the process is resistant to DeepFake applications and facial recognition (FR) systems, thereby proving its effectiveness in privacy protection.
△ Less
Submitted 14 December, 2024; v1 submitted 7 December, 2023;
originally announced December 2023.
-
Determination of Molecular Energies via Quantum Imaginary Time Evolution in a Superconducting Qubit System
Authors:
Zhiwen Zong,
Sainan Huai,
Tianqi Cai,
Wenyan Jin,
Ze Zhan,
Zhenxing Zhang,
Kunliang Bu,
Liyang Sui,
Ying Fei,
Yicong Zheng,
Shengyu Zhang,
Jianlan Wu,
Yi Yin
Abstract:
As a valid tool for solving ground state problems, imaginary time evolution (ITE) is widely used in physical and chemical simulations. Different ITE-based algorithms in their quantum counterpart have recently been proposed and applied to some real systems. We experimentally realize the variational-based quantum imaginary time evolution (QITE) algorithm to simulate the ground state energy of hydrog…
▽ More
As a valid tool for solving ground state problems, imaginary time evolution (ITE) is widely used in physical and chemical simulations. Different ITE-based algorithms in their quantum counterpart have recently been proposed and applied to some real systems. We experimentally realize the variational-based quantum imaginary time evolution (QITE) algorithm to simulate the ground state energy of hydrogen (H2) and lithium hydride (LiH) molecules in a superconducting qubit system. The H2 molecule is directly simulated using the 3-qubit circuit with unitary-coupled clusters (UCC) ansatz. We also combine QITE with the cluster mean-field (CMF) method to obtain an effective Hamiltonian. The LiH molecule is correspondingly simulated using the 3-qubit circuit with hardware-efficient ansatz. For comparison, the LiH molecule is also directly simulated using the 4-qubit circuit with UCC ansatz at the equilibrium point. All the experimental results show a convergence within 4 iterations, with high-fidelity ground state energy obtained. For a more complex system in the future, the CMF may allow further grouping of interactions to obtain an effective Hamiltonian, then the hybrid QITE algorithm can possibly simulate a relatively large-scale system with fewer qubits.
△ Less
Submitted 2 March, 2023;
originally announced March 2023.
-
Efficient Quantum Simulation of Electron-Phonon Systems by Variational Basis State Encoder
Authors:
Weitang Li,
Jiajun Ren,
Sainan Huai,
Tianqi Cai,
Zhigang Shuai,
Shengyu Zhang
Abstract:
Digital quantum simulation of electron-phonon systems requires truncating infinite phonon levels into $N$ basis states and then encoding them with qubit computational basis. Unary encoding and the more compact binary/Gray encoding are the two most representative encoding schemes, which demand $\mathcal{O}(N)$ and $\mathcal{O}(\log{N})$ qubits as well as $\mathcal{O}(N)$ and…
▽ More
Digital quantum simulation of electron-phonon systems requires truncating infinite phonon levels into $N$ basis states and then encoding them with qubit computational basis. Unary encoding and the more compact binary/Gray encoding are the two most representative encoding schemes, which demand $\mathcal{O}(N)$ and $\mathcal{O}(\log{N})$ qubits as well as $\mathcal{O}(N)$ and $\mathcal{O}(N\log{N})$ quantum gates respectively. In this work, we propose a variational basis state encoding algorithm that reduces the scaling of the number of qubits and quantum gates to both $\mathcal{O}(1)$ for systems obeying the area law of entanglement entropy. The cost for the scaling reduction is a constant amount of additional measurement. The accuracy and efficiency of the approach are verified by both numerical simulation and realistic quantum hardware experiments. In particular, we find using one or two qubits for each phonon mode is sufficient to produce quantitatively correct results across weak and strong coupling regimes. Our approach paves the way for practical quantum simulation of electron-phonon systems on both near-term hardware and error-corrected quantum computers.
△ Less
Submitted 3 May, 2023; v1 submitted 3 January, 2023;
originally announced January 2023.
-
You Only Search Once: On Lightweight Differentiable Architecture Search for Resource-Constrained Embedded Platforms
Authors:
Xiangzhong Luo,
Di Liu,
Hao Kong,
Shuo Huai,
Hui Chen,
Weichen Liu
Abstract:
Benefiting from the search efficiency, differentiable neural architecture search (NAS) has evolved as the most dominant alternative to automatically design competitive deep neural networks (DNNs). We note that DNNs must be executed under strictly hard performance constraints in real-world scenarios, for example, the runtime latency on autonomous vehicles. However, to obtain the architecture that m…
▽ More
Benefiting from the search efficiency, differentiable neural architecture search (NAS) has evolved as the most dominant alternative to automatically design competitive deep neural networks (DNNs). We note that DNNs must be executed under strictly hard performance constraints in real-world scenarios, for example, the runtime latency on autonomous vehicles. However, to obtain the architecture that meets the given performance constraint, previous hardware-aware differentiable NAS methods have to repeat a plethora of search runs to manually tune the hyper-parameters by trial and error, and thus the total design cost increases proportionally. To resolve this, we introduce a lightweight hardware-aware differentiable NAS framework dubbed LightNAS, striving to find the required architecture that satisfies various performance constraints through a one-time search (i.e., \underline{\textit{you only search once}}). Extensive experiments are conducted to show the superiority of LightNAS over previous state-of-the-art methods.
△ Less
Submitted 29 August, 2022;
originally announced August 2022.
-
Rapid and Unconditional Parametric Reset Protocol for Tunable Superconducting Qubits
Authors:
Yu Zhou,
Zhenxing Zhang,
Zelong Yin,
Sainan Huai,
Xiu Gu,
Xiong Xu,
Jonathan Allcock,
Fuming Liu,
Guanglei Xi,
Qiaonian Yu,
Hualiang Zhang,
Mengyu Zhang,
Hekang Li,
Xiaohui Song,
Zhan Wang,
Dongning Zheng,
Shuoming An,
Yarui Zheng,
Shengyu Zhang
Abstract:
Qubit initialization is a critical task in quantum computation and communication. Extensive efforts have been made to achieve this with high speed, efficiency and scalability. However, previous approaches have either been measurement-based and required fast feedback, suffered from crosstalk or required sophisticated calibration. Here, we report a fast and high-fidelity reset scheme, avoiding the i…
▽ More
Qubit initialization is a critical task in quantum computation and communication. Extensive efforts have been made to achieve this with high speed, efficiency and scalability. However, previous approaches have either been measurement-based and required fast feedback, suffered from crosstalk or required sophisticated calibration. Here, we report a fast and high-fidelity reset scheme, avoiding the issues above without any additional chip architecture. By modulating the flux through a transmon qubit, we realize a swap between the qubit and its readout resonator that suppresses the excited state population to 0.08% $\pm$ 0.08% within 34 ns (284 ns if photon depletion of the resonator is required). Furthermore, our approach (i) can achieve effective second excited state depletion, (ii) has negligible effects on neighbouring qubits, and (iii) offers a way to entangle the qubit with an itinerant single photon, useful in quantum communication applications.
△ Less
Submitted 22 November, 2021; v1 submitted 21 March, 2021;
originally announced March 2021.
-
HSCoNAS: Hardware-Software Co-Design of Efficient DNNs via Neural Architecture Search
Authors:
Xiangzhong Luo,
Di Liu,
Shuo Huai,
Weichen Liu
Abstract:
In this paper, we present a novel multi-objective hardware-aware neural architecture search (NAS) framework, namely HSCoNAS, to automate the design of deep neural networks (DNNs) with high accuracy but low latency upon target hardware. To accomplish this goal, we first propose an effective hardware performance modeling method to approximate the runtime latency of DNNs on target hardware, which wil…
▽ More
In this paper, we present a novel multi-objective hardware-aware neural architecture search (NAS) framework, namely HSCoNAS, to automate the design of deep neural networks (DNNs) with high accuracy but low latency upon target hardware. To accomplish this goal, we first propose an effective hardware performance modeling method to approximate the runtime latency of DNNs on target hardware, which will be integrated into HSCoNAS to avoid the tedious on-device measurements. Besides, we propose two novel techniques, i.e., dynamic channel scaling to maximize the accuracy under the specified latency and progressive space shrinking to refine the search space towards target hardware as well as alleviate the search overheads. These two techniques jointly work to allow HSCoNAS to perform fine-grained and efficient explorations. Finally, an evolutionary algorithm (EA) is incorporated to conduct the architecture search. Extensive experiments on ImageNet are conducted upon diverse target hardware, i.e., GPU, CPU, and edge device to demonstrate the superiority of HSCoNAS over recent state-of-the-art approaches.
△ Less
Submitted 11 March, 2021;
originally announced March 2021.
-
Superposition of macroscopically squeezed states with enhanced squeezing in cavity optomechanical system at single-photon level
Authors:
Sai-Nan Huai,
Wei Nie,
Yun-bo Zhang,
Yu-xi Liu
Abstract:
We propose an efficient approach to generate the superposed macroscopically squeezed states with enhanced squeezing in a two-mode optomechanical system. This can be achieved by introducing a sinusoidal modulation to either the cavity frequencies or the coupling strengths between two cavity modes. The squeezement of the oscillator can be significantly enhanced to 12.16 dB with single photon, once t…
▽ More
We propose an efficient approach to generate the superposed macroscopically squeezed states with enhanced squeezing in a two-mode optomechanical system. This can be achieved by introducing a sinusoidal modulation to either the cavity frequencies or the coupling strengths between two cavity modes. The squeezement of the oscillator can be significantly enhanced to 12.16 dB with single photon, once the relative ratio of coupling strength is optimized under proper conditions. Further enhanced squeezing can be obtained by carefully adjusting the system parameters. In terms of the Wigner quasi-probability distribution, we show the squeezed error ellipses and interference fringes of the Yurke-Stoler-type squeezed states, denoting the squeezing and superposition properties. Our state generation scheme show reliable performance and robust resistance to finite environmental fluctuations, which implies applications for both fundamental interest and practical value.
△ Less
Submitted 14 May, 2019;
originally announced May 2019.
-
Mechanically modulated emission spectra and blockade of polaritons
Authors:
Sai-Nan Huai,
Yu-Long Liu,
Yunbo Zhang,
Yu-xi Liu
Abstract:
We study a hybrid semiconductor-optomechanical system, which consists of a cavity with an oscillating mirror made by semiconducting materials or with a semiconducting membrane inside. The cavity photons and the excitons in the oscillating mirror or semiconducting membrane form into polaritons. And correspondingly, the optomechanical interaction between the cavity photons and the mirror or membrane…
▽ More
We study a hybrid semiconductor-optomechanical system, which consists of a cavity with an oscillating mirror made by semiconducting materials or with a semiconducting membrane inside. The cavity photons and the excitons in the oscillating mirror or semiconducting membrane form into polaritons. And correspondingly, the optomechanical interaction between the cavity photons and the mirror or membrane is changed into the polariton-mechanical interaction. We theoretically study the eigenenergies and eigenfunctions of this tripartite hybrid system with the generalized rotating-wave approximation. We show that the emission spectrum of polariton mode is modulated by the mechanical resonator. We also study the mechanical effect on the statistical properties of the polariton when the cavity is driven by a weak classical field. This work provides a detailed description of the rich nonlinearity owing to the competition between parametric coupling and three-wave mixing interaction concerning the polariton modes and the phonon mode. It also offers a way to operate the photons, phonons and excitons, e.g., detect the properties of mechanical resonator through the fine spectra of the polaritons or control the transmission of light in the integrated semiconducting-optomechanical platform.
△ Less
Submitted 8 March, 2018;
originally announced March 2018.
-
Vacuum induced transparency and photon number resolved Autler-Townes splitting in a three-level system
Authors:
Jiang-Hao Ding,
Sai-nan Huai,
Hou Ian,
Yu-xi Liu
Abstract:
We study the absorption spectrum of a probe field by a Λ-type three-level system, which is coupled to a quantized control field through the two upper energy levels. The probe field is applied to the ground and the second excited states. When the quantized control field is in vacuum, we derive a threshold condition to discern vacuum induced transparency (VIT) and vacuum induced Autler- Townes split…
▽ More
We study the absorption spectrum of a probe field by a Λ-type three-level system, which is coupled to a quantized control field through the two upper energy levels. The probe field is applied to the ground and the second excited states. When the quantized control field is in vacuum, we derive a threshold condition to discern vacuum induced transparency (VIT) and vacuum induced Autler- Townes splitting (ATS). We also find that the parameter change from VIT to vacuum induced ATS is very similar to that from broken PT symmetry to PT symmetry. Moreover, we find the photon number resolved spectrum in the parameter regime of vacuum induced ATS when the mean photon number of the quantized control field is changed from zero (vacuum) to a finite number. However, there is no photon number resolved spectrum in the parameter regime of VIT even that the quantized control field contains the finite number of photons. Finally, we further discuss possible experimental realization.
△ Less
Submitted 10 July, 2017;
originally announced July 2017.
-
Polariton states in circuit QED for electromagnetically induced transparency
Authors:
Xiu Gu,
Sai-nan Huai,
Franco Nori,
Yu-xi Liu
Abstract:
Electromagnetically induced transparency (EIT) has been extensively studied in various systems. However, it is not easy to observe in superconducting quantum circuits (SQCs), because the Rabi frequency of the strong controlling field corresponding to EIT is limited by the decay rates of the SQCs. Here, we show that EIT can be achieved by engineering decay rates in a superconducting circuit QED sys…
▽ More
Electromagnetically induced transparency (EIT) has been extensively studied in various systems. However, it is not easy to observe in superconducting quantum circuits (SQCs), because the Rabi frequency of the strong controlling field corresponding to EIT is limited by the decay rates of the SQCs. Here, we show that EIT can be achieved by engineering decay rates in a superconducting circuit QED system through a classical driving field on the qubit. Without such a driving field, the superconducting qubit and the cavity field are approximately decoupled in the large detuning regime, and thus the eigenstates of the system are approximately product states of the cavity field and qubit states. However, the driving field can strongly mix these product states and so-called polariton states can be formed. The weights of the states for the qubit and cavity field in the polariton states can be tuned by the driving field, and thus the decay rates of the polariton states can be changed. We choose a three-level system with $Λ$-type transitions in such a driven circuit QED system, and demonstrate how EIT and ATS can be realized in this compound system. We believe that this study will be helpful for EIT experiments using SQCs.
△ Less
Submitted 3 February, 2016; v1 submitted 19 January, 2016;
originally announced January 2016.
-
Geometric Curvature and Phase of the Rabi model
Authors:
Lijun Mao,
Sainan Huai,
Liping Guo,
Yunbo Zhang
Abstract:
We study the geometric curvature and phase of the Rabi model. Under the rotating-wave approximation (RWA), we apply the gauge independent Berry curvature over a surface integral to calculate the Berry phase of the eigenstates for both single and two-qubit systems, which is found to be identical with the system of spin-1/2 particle in a magnetic field. We extend the idea to define a vacuum-induced…
▽ More
We study the geometric curvature and phase of the Rabi model. Under the rotating-wave approximation (RWA), we apply the gauge independent Berry curvature over a surface integral to calculate the Berry phase of the eigenstates for both single and two-qubit systems, which is found to be identical with the system of spin-1/2 particle in a magnetic field. We extend the idea to define a vacuum-induced geometric curvature when the system starts from an initial state with pure vacuum bosonic field. The induced geometric phase is related to the average photon number in a period which is possible to measure in the qubit-cavity system. We also calculate the geometric phase beyond the RWA and find an anomalous sudden change, which implies the breakdown of the adiabatic theorem and the Berry phases in an adiabatic cyclic evolution are ill-defined near the anti-crossing point in the spectrum.
△ Less
Submitted 20 December, 2015;
originally announced December 2015.
-
Tavis-Cummings model beyond the rotating wave approximation: Inhomogeneous coupling
Authors:
Lijun Mao,
Sainan Huai,
Yunbo Zhang
Abstract:
We present the analytical solution of the Tavis-Cummings (TC) model for more than one qubit inhomogeneously coupled to a single mode radiation field beyond the rotating-wave approximation (RWA). The significant advantage of the displaced oscillator basis enables us to apply the same truncation techniques adopted in the single qubit Jaynes-Cummings (JC) model to the multiple qubits system. The deri…
▽ More
We present the analytical solution of the Tavis-Cummings (TC) model for more than one qubit inhomogeneously coupled to a single mode radiation field beyond the rotating-wave approximation (RWA). The significant advantage of the displaced oscillator basis enables us to apply the same truncation techniques adopted in the single qubit Jaynes-Cummings (JC) model to the multiple qubits system. The derived analytical spectrum match perfectly the exact diagonalization numerical solutions of the inhomogeneous TC model in the parameter regime where the qubits transition frequencies are far off-resonance with the field frequency and the interaction strengths reach the ultra-strong coupling regime. The two-qubit TC model is quasi-exactly solvable because part of the spectra can be determined exactly in the homogeneous coupling case with two identical qubits or with symmetric(asymmetric) detuning. By means of the fidelity of quantum states we identify several nontrivial level crossing points in the same parity subspace, which implies that homogeneous coupled two-qubit TC model with $ω_1=ω_2$ or $ω_1\pmω_2=2ω_c$ is integrable. We further explore the time evolution of the qubit's population inversion and the entanglement behavior taking two qubits as an example. The analytical methods provide unexpectedly accurate results in describing the dynamics of the qubit in the present experimentally accessible coupling regime, showing that the collapse-revival phenomena emerge, survive, and are finally destroyed when the coupling strength increases beyond the ultra-strong coupling regime. The suggested procedure applies readily to the multiple qubits system such as the GHZ state entanglement evolution and quantum entanglement between a single photon and superconducting qubits of particular experiment interest.
△ Less
Submitted 24 March, 2014;
originally announced March 2014.
