arXiv subjects

Shuo Liu

Publications and source records attributed to Shuo Liu.

At least 163 records · Page 9Linked to original sources

arXiv2023

MetaML: Automating Customizable Cross-Stage Design-Flow for Deep Learning Acceleration

This paper introduces a novel optimization framework for deep neural network (DNN) hardware accelerators, enabling the rapid development of customized and automated design flows. More specifically, our approach aims to automate the selection and configuration of low-level optimization techniques, encompassing DNN and FPGA low-level optimizations. We introduce novel optimization and transformation tasks for building design-flow architectures, which are highly customizable and flexible, thereby enhancing the performance and efficiency of DNN accelerators. Our results demonstrate considerable reductions of up to 92\% in DSP usage and 89\% in LUT usage for two networks, while maintaining accuracy and eliminating the need for human effort or domain expertise. In comparison to state-of-the-art approaches, our design achieves higher accuracy and utilizes three times fewer DSP resources, underscoring the advantages of our proposed framework.

cs.LG↗

arXiv2023

Towards Selection of Text-to-speech Data to Augment ASR Training

This paper presents a method for selecting appropriate synthetic speech samples from a given large text-to-speech (TTS) dataset as supplementary training data for an automatic speech recognition (ASR) model. We trained a neural network, which can be optimised using cross-entropy loss or Arcface loss, to measure the similarity of a synthetic data to real speech. We found that incorporating synthetic samples with considerable dissimilarity to real speech, owing in part to lexical differences, into ASR training is crucial for boosting recognition performance. Experimental results on Librispeech test sets indicate that, in order to maintain the same speech recognition accuracy as when using all TTS data, our proposed solution can reduce the size of the TTS data down below its $30\,\%$, which is superior to several baseline methods.

eess.AS↗

arXiv2023

Discrete time crystal enabled by Stark many-body localization

Discrete time crystal (DTC) has recently attracted increasing attention, but most DTC models and their properties are only revealed after disorder average. In this Letter, we propose a simple disorder-free periodically driven model that exhibits nontrivial DTC order stabilized by Stark many-body localization (MBL). We demonstrate the existence of DTC phase by analytical analysis from perturbation theory and convincing numerical evidence from observable dynamics. The new DTC model paves a new promising way for further experiments and deepens our understanding of DTC. Since the DTC order doesn't require special quantum state preparation and the strong disorder average, it can be naturally realized on the noisy intermediate-scale quantum (NISQ) hardware with much fewer resources and repetitions. Moreover, besides the robust subharmonic response, there are other novel robust beating oscillations in Stark-MBL DTC phase which are absent in random or quasi-periodic MBL DTC.

cond-mat.str-el↗

arXiv2023

Training variational quantum algorithms with random gate activation

Variational quantum algorithms (VQAs) hold great potentials for near-term applications and are promising to achieve quantum advantage on practical tasks. However, VQAs suffer from severe barren plateau problem as well as have a large probability of being trapped in local minima. In this Letter, we propose a novel training algorithm with random quantum gate activation for VQAs to efficiently address these two issues. This new algorithm processes effectively much fewer training parameters than the conventional plain optimization strategy, which efficiently mitigates barren plateaus with the same expressive capability. Additionally, by randomly adding two-qubit gates to the circuit ansatz, the optimization trajectories can escape from local minima and reach the global minimum more frequently due to more sources of randomness. In real quantum experiments, the new training algorithm can also reduce the quantum computational resources required and be more quantum noise resilient. We apply our training algorithm to solve variational quantum simulation problems for ground states and present convincing results that showcase the advantages of our novel strategy where better performance is achieved by the combination of mitigating barren plateaus, escaping from local minima, and reducing the effect of quantum noises. We further propose that the entanglement phase transition could be one underlying reason why our RA training is so effective.

quant-ph↗

arXiv2023

The Solar Upper Transition Region Imager (SUTRI) onboard the SATech-01 satellite

The Solar Upper Transition Region Imager (SUTRI) onboard the Space Advanced Technology demonstration satellite (SATech-01), which was launched to a sun-synchronous orbit at a height of 500 km in July 2022, aims to test the on-orbit performance of our newly developed Sc-Si multi-layer reflecting mirror and the 2kx2k EUV CMOS imaging camera and to take full-disk solar images at the Ne VII 46.5 nm spectral line with a filter width of 3 nm. SUTRI employs a Ritchey-Chretien optical system with an aperture of 18 cm. The on-orbit observations show that SUTRI images have a field of view of 41.6'x41.6' and a moderate spatial resolution of 8" without an image stabilization system. The normal cadence of SUTRI images is 30 s and the solar observation time is about 16 hours each day because the earth eclipse time accounts for about 1/3 of SATech-01's orbit period. Approximately 15 GB data is acquired each day and made available online after processing. SUTRI images are valuable as the Ne VII 46.5 nm line is formed at a temperature regime of 0.5 MK in the solar atmosphere, which has rarely been sampled by existing solar imagers. SUTRI observations will establish connections between structures in the lower solar atmosphere and corona, and advance our understanding of various types of solar activity such as flares, filament eruptions, coronal jets and coronal mass ejections.

astro-ph.SR↗

arXiv2023

Analysis of high-energy drop impact onto deep liquid pool

The present work is devoted to the analysis of drop impact onto a deep liquid pool. It is focused on the effects of high-energy splash regimes, caused by the impact of large raindrops at high velocities. Such cases are characterized by short time scales and complex mechanisms, and they have thus received little attention until now. The BASILISK open-source solver is used to perform three-dimensional Direct Numerical Simulations (DNS). The capabilities of the octree adaptive mesh refinement techniques enable to capture the small-scale features of the flow, while the Volume of Fluid (VOF) approach combined with a balanced force surface tension calculation is applied to advect the volume fraction of the liquids and reconstruct the interfaces. The numerical results compare well with experimental visualizations: both the evolution of crown and cavity, the emanation of ligaments, the formation of bubble canopy, and the growth of the downward spiral jet that pierces through the cavity bottom, are correctly reproduced. Reliable quantitative agreements are also obtained regarding the time evolution of rim positions, cavity dimensions and droplet distributions through an observation window. Furthermore, simulation gives access to various aspects of the internal flows (velocity, vorticity, pressure, energy budget), which allows to better explain the corresponding physical phenomena. Details of the early dynamics of bubble ring entrapment and splashing performance, the formation/collapse of bubble canopy, and the spreading of drop liquid, are discussed. The statistics of droplet size show the bimodal distribution in time, corroborating distinct primary mechanisms of droplet production at different stages.

physics.flu-dyn↗

arXiv2023

Probing many-body localization by excited-state VQE

Non-equilibrium physics including many-body localization (MBL) has attracted increasing attentions, but theoretical approaches of reliably studying non-equilibrium properties remain quite limited. In this Letter, we propose a systematic approach to probe MBL phases via the excited-state variational quantum eigensolver (VQE) and demonstrate convincing results of MBL on a quantum hardware, which we believe paves a promising way for future simulations of non-equilibrium systems beyond the reach of classical computations in the noisy intermediate-scale quantum (NISQ) era. Moreover, the MBL probing protocol based on excited-state VQE is NISQ-friendly, as it can successfully differentiate the MBL phase from thermal phases with relatively-shallow quantum circuits, and it is also robust against the effect of quantum noises.

quant-ph↗

arXiv2023

Wave Statistics and Energy Dissipation of Breaking Waves Generated by a Wave Plate

The present study focuses on direct numerical simulations of breaking waves generated by a wave plate at constant water depths. The aim is to quantify the dynamics and kinematics in the breaking process, together with the air entrainment and energy dissipation due to breaking. Good agreement is achieved between numerical and experimental results in terms of free-surface profiles, energy budget, and bubble statistics during wave breaking. A parametric study was conducted to examine the effects of wave properties and initial conditions on breaking features. According to research on the Bond number ($Bo$, the ratio of gravitational to surface tension forces), a larger surface tension produces more significant parasitic capillaries at the forward face of the wave profile and a thicker plunging jet, which causes a delayed breaking time and is tightly correlated with the main cavity size. A close relationship between wave statistics and the initial conditions of the wave plate is discovered, showing that breaker types can be classified according to the ratio of wave height to water depth, $H/d$. Moreover, the energy dissipation rate due to breaking can be related to the initial conditions of the wave plate by applying the conventional dissipation scaling of turbulence theory, and further correlated with the local breaking crest geometry using inertial-scaling arguments. The proposed scaling of the dissipation rate during the active breaking period is found to be in good agreement with our numerical results.

physics.flu-dyn↗

arXiv2023

TensorCircuit: a Quantum Software Framework for the NISQ Era

TensorCircuit is an open source quantum circuit simulator based on tensor network contraction, designed for speed, flexibility and code efficiency. Written purely in Python, and built on top of industry-standard machine learning frameworks, TensorCircuit supports automatic differentiation, just-in-time compilation, vectorized parallelism and hardware acceleration. These features allow TensorCircuit to simulate larger and more complex quantum circuits than existing simulators, and are especially suited to variational algorithms based on parameterized quantum circuits. TensorCircuit enables orders of magnitude speedup for various quantum simulation tasks compared to other common quantum software, and can simulate up to 600 qubits with moderate circuit depth and low-dimensional connectivity. With its time and space efficiency, flexible and extensible architecture and compact, user-friendly API, TensorCircuit has been built to facilitate the design, simulation and analysis of quantum algorithms in the Noisy Intermediate-Scale Quantum (NISQ) era.

quant-ph↗

arXiv2023

HEAR4Health: A blueprint for making computer audition a staple of modern healthcare

Recent years have seen a rapid increase in digital medicine research in an attempt to transform traditional healthcare systems to their modern, intelligent, and versatile equivalents that are adequately equipped to tackle contemporary challenges. This has led to a wave of applications that utilise AI technologies; first and foremost in the fields of medical imaging, but also in the use of wearables and other intelligent sensors. In comparison, computer audition can be seen to be lagging behind, at least in terms of commercial interest. Yet, audition has long been a staple assistant for medical practitioners, with the stethoscope being the quintessential sign of doctors around the world. Transforming this traditional technology with the use of AI entails a set of unique challenges. We categorise the advances needed in four key pillars: Hear, corresponding to the cornerstone technologies needed to analyse auditory signals in real-life conditions; Earlier, for the advances needed in computational and data efficiency; Attentively, for accounting to individual differences and handling the longitudinal nature of medical data; and, finally, Responsibly, for ensuring compliance to the ethical standards accorded to the field of medicine.

cs.SD↗

arXiv2022

Absence of logarithmic and algebraic scaling entanglement phases due to skin effect

Measurement-induced phase transition in the presence of competition between projective measurement and random unitary evolution has attracted increasing attention due to the rich phenomenology of entanglement structures. However, in open quantum systems with free fermions, a generalized measurement with conditional feedback can induce skin effect and render the system short-range entangled without any entanglement transition, meaning the system always remains in the ``area law'' entanglement phase. In this work, we demonstrate that the power-law long-range hopping does not alter the absence of entanglement transition brought on by the measurement-induced skin effect for systems with open boundary conditions. In addition, for the finite-size systems, we discover an algebraic scaling $S(L, L/4)\sim L^{3/2-p}$ when the power-law exponent $p$ of long-range hopping is relatively small. For systems with periodic boundary conditions, we find that the measurement-induced skin effect disappears and observe entanglement phase transitions among ``algebraic law'', ``logarithmic law'', and ``area law'' phases.

quant-ph↗

arXiv2022

Universal KPZ scaling in noisy hybrid quantum circuits

Measurement-induced phase transitions (MIPT) have attracted increasing attention due to the rich phenomenology of entanglement structures and their relation with quantum information processing. Since physical systems are unavoidably coupled to environment, quantum noise needs be considered in analyzing a system with MIPT, which may qualitatively modify or even destroy certain entanglement structure of the system. In this Letter, we investigate the effect of quantum noise modeled by reset quantum channel acting on each site with probability $q$ on MIPT. Based on the numerical results from the Clifford circuits, we show that the quantum noise can qualitatively change the entanglement properties - the entanglement obeys ``area law'' instead of ``volume law'' with projective measurement rate $p<p_{c}$. In the quantum noise induced ``area law'' phase, the entanglement exhibits a novel $q^{-1/3}$ power-law scaling. Using an analytic mapping of the quantum model to a classical statistical model, we further show that the ``area law'' entanglement is the consequence of the noise-driven symmetry-breaking field and the $q^{-1/3}$ scaling can be understood as the result of Kardar-Parisi-Zhang (KPZ) fluctuations of the directed polymer with an effective length scale $L_{\rm{eff}} \sim q^{-1}$ in a random environment.

quant-ph↗

arXiv2022

TLP: A Deep Learning-based Cost Model for Tensor Program Tuning

Tensor program tuning is a non-convex objective optimization problem, to which search-based approaches have proven to be effective. At the core of the search-based approaches lies the design of the cost model. Though deep learning-based cost models perform significantly better than other methods, they still fall short and suffer from the following problems. First, their feature extraction heavily relies on expert-level domain knowledge in hardware architectures. Even so, the extracted features are often unsatisfactory and require separate considerations for CPUs and GPUs. Second, a cost model trained on one hardware platform usually performs poorly on another, a problem we call cross-hardware unavailability. In order to address these problems, we propose TLP and MTLTLP. TLP is a deep learning-based cost model that facilitates tensor program tuning. Instead of extracting features from the tensor program itself, TLP extracts features from the schedule primitives. We treat schedule primitives as tensor languages. TLP is thus a Tensor Language Processing task. In this way, the task of predicting the tensor program latency through the cost model is transformed into a natural language processing (NLP) regression task. MTL-TLP combines Multi-Task Learning and TLP to cope with the cross-hardware unavailability problem. We incorporate these techniques into the Ansor framework and conduct detailed experiments. Results show that TLP can speed up the average search time by 9.1X and 3.0X on CPU and GPU workloads, respectively, compared to the state-of-the-art implementation. MTL-TLP can achieve a speed-up of 4.7X and 2.9X on CPU and GPU workloads, respectively, using only 7% of the target hardware data.

cs.LG↗

arXiv2022

An Overview of Affective Speech Synthesis and Conversion in the Deep Learning Era

Speech is the fundamental mode of human communication, and its synthesis has long been a core priority in human-computer interaction research. In recent years, machines have managed to master the art of generating speech that is understandable by humans. But the linguistic content of an utterance encompasses only a part of its meaning. Affect, or expressivity, has the capacity to turn speech into a medium capable of conveying intimate thoughts, feelings, and emotions -- aspects that are essential for engaging and naturalistic interpersonal communication. While the goal of imparting expressivity to synthesised utterances has so far remained elusive, following recent advances in text-to-speech synthesis, a paradigm shift is well under way in the fields of affective speech synthesis and conversion as well. Deep learning, as the technology which underlies most of the recent advances in artificial intelligence, is spearheading these efforts. In the present overview, we outline ongoing trends and summarise state-of-the-art approaches in an attempt to provide a comprehensive overview of this exciting field.

cs.SD↗

arXiv2022

M2HF: Multi-level Multi-modal Hybrid Fusion for Text-Video Retrieval

Videos contain multi-modal content, and exploring multi-level cross-modal interactions with natural language queries can provide great prominence to text-video retrieval task (TVR). However, new trending methods applying large-scale pre-trained model CLIP for TVR do not focus on multi-modal cues in videos. Furthermore, the traditional methods simply concatenating multi-modal features do not exploit fine-grained cross-modal information in videos. In this paper, we propose a multi-level multi-modal hybrid fusion (M2HF) network to explore comprehensive interactions between text queries and each modality content in videos. Specifically, M2HF first utilizes visual features extracted by CLIP to early fuse with audio and motion features extracted from videos, obtaining audio-visual fusion features and motion-visual fusion features respectively. Multi-modal alignment problem is also considered in this process. Then, visual features, audio-visual fusion features, motion-visual fusion features, and texts extracted from videos establish cross-modal relationships with caption queries in a multi-level way. Finally, the retrieval outputs from all levels are late fused to obtain final text-video retrieval results. Our framework provides two kinds of training strategies, including an ensemble manner and an end-to-end manner. Moreover, a novel multi-modal balance loss function is proposed to balance the contributions of each modality for efficient end-to-end training. M2HF allows us to obtain state-of-the-art results on various benchmarks, eg, Rank@1 of 64.9\%, 68.2\%, 33.2\%, 57.1\%, 57.8\% on MSR-VTT, MSVD, LSMDC, DiDeMo, and ActivityNet, respectively.

cs.MM↗

arXiv2022

Multi-Forgery Detection Challenge 2022: Push the Frontier of Unconstrained and Diverse Forgery Detection

In this paper, we present the Multi-Forgery Detection Challenge held concurrently with the IEEE Computer Society Workshop on Biometrics at CVPR 2022. Our Multi-Forgery Detection Challenge aims to detect automatic image manipulations including but not limited to image editing, image synthesis, image generation, image photoshop, etc. Our challenge has attracted 674 teams from all over the world, with about 2000 valid result submission counts. We invited the Top 10 teams to present their solutions to the challenge, from which three teams are awarded prizes in the grand finale. In this paper, we present the solutions from the Top 3 teams, in order to boost the research work in the field of image forgery detection.

cs.CV↗

arXiv2022

Multistage linguistic conditioning of convolutional layers for speech emotion recognition

In this contribution, we investigate the effectiveness of deep fusion of text and audio features for categorical and dimensional speech emotion recognition (SER). We propose a novel, multistage fusion method where the two information streams are integrated in several layers of a deep neural network (DNN), and contrast it with a single-stage one where the streams are merged in a single point. Both methods depend on extracting summary linguistic embeddings from a pre-trained BERT model, and conditioning one or more intermediate representations of a convolutional model operating on log-Mel spectrograms. Experiments on the MSP-Podcast and IEMOCAP datasets demonstrate that the two fusion methods clearly outperform a shallow (late) fusion baseline and their unimodal constituents, both in terms of quantitative performance and qualitative behaviour. Overall, our multistage fusion shows better quantitative performance, surpassing alternatives on most of our evaluations. This illustrates the potential of multistage fusion in better assimilating text and audio information.

cs.LG↗

arXiv2022

Surface critical properties of the three-dimensional clock model

Using Monte Carlo simulations and finite-size scaling analysis, we show that the $q$-state clock model with $q=6$ on the simple cubic lattice with open surfaces has a rich phase diagram; in particular, it has an extraordinary-log phase, besides the ordinary and extraordinary transitions at the bulk critical point. We prove numerically that the presence of the intermediate extraordinary-log phase is due to the emergence of an O(2) symmetry in the surface state before the surface enters the $Z_{q}$ symmetry-breaking region as the surface coupling is increased at the bulk critical point, while O(2) symmetry emerges for the bulk. The critical behaviors of the extraordinary-log transition, as well as the ordinary and the special transition separating the ordinary and the extraordinary-log transition are obtained.

cond-mat.stat-mech↗