arXiv subjects

Changhong Li

Publications and source records attributed to Changhong Li.

At least 19 recordsLinked to original sources

arXiv2026

RQP: Resource-Oriented Quantiser Pruning for Neural Networks on FPGAs

High granularity quantisation (HGQ) exploits weight-level quantisation and pruning to design resource-efficient neural network accelerators, achieving an attractive trade-off between accuracy and hardware utilisation. HGQ is particularly well suited to FPGA-based edge neural network applications. Standard HGQ workflow starts from a high-precision model and progressively reduces bit width, guided by gradient-based optimisation to outline the Pareto frontier. This monotonic and irreversible pruning process is computationally intensive and can overlook the optimal subnetwork for a given resource level. We propose a resource-oriented one-shot quantiser pruning method that brings the network directly close to the target search space, and then use bidirectional beta scheduling for fine-tuning to enable a more refined scan of the Pareto frontier. Validated on the jet substructure classification, JSC, task, our method reduces the search cost by up to 20.58x compared with monotonic resource reduction in standard HGQ workflows, while achieving a competitive Pareto frontier and final network configuration.

cs.AR↗

arXiv2026

Model Predictive Current Control with Harmonic Correction for Single-Phase AC-DC EV Charging

The increasing integration of Electric Vehicles (EVs) has imposed a growing harmonic challenge on the power grid. For AC/DC Power Factor Correction (PFC) in single-phase On-Board Chargers (OBCs), Model Predictive Current Control (MPCC) improves the current quality by predicting and tracking the inductor current. However, finite control set MPCC selects switching states, resulting in discrete control actions and a limited optimisation space. Moreover, the MPCC cost function based on instantaneous current tracking error has limited capability to compensate for low-order harmonic disturbances induced by dead time, control delay, and model parameter mismatch. This paper proposes a duty cycle predictive MPCC incorporating a real-time harmonic estimation reference. The proposed method dynamically estimates the low-order harmonic components of the input current and corrects the MPCC reference current, enabling continuous duty cycle control and targeted suppression of dominant low-order harmonics. Simulation results on a single-phase OBC demonstrate that the proposed duty cycle predictive MPCC reduces the steady-state current THD_i from 11.47% to 6.10% compared with the switching state predictive MPCC. With the harmonic reference, the THD_i is further reduced to 2.85%.

eess.SY↗

arXiv2026

Stability of Anomalous Dissipation for the Forced 3D Navier--Stokes Equations under Geometric Perturbations

The energy dissipation in the inviscid limit is a central problem in turbulence theory. Kolmogorov's K41 theory predicts a positive dissipation rate independent of viscosity -- a phenomenon known as anomalous dissipation. Bru\'e and De Lellis gave the first rigorous construction, but it relies on extremely precise geometric conditions. Based on quasi-self-similar mixing, we prove structural stability under pure normal perturbations of the central curves. We establish C^2 stability of the maps and C^1 stability of the local fields, and obtain H\"older estimates and high-frequency energy concentration. A contradiction gives a positive dissipation lower bound independent of the perturbation, and embedding into the (2+1/2)-dimensional framework shows C^6 structural stability. The main novelty is that the Bru\'e--De Lellis construction remains stable under such perturbations, so anomalous dissipation occurs in an open neighbourhood of function spaces, providing a rigorous foundation for K41 theory.

math.AP↗

arXiv2026

SSC-UNet: UNet with Self-Supervised Contrastive Learning for Phonocardiography Noise Reduction

Congenital Heart Disease (CHD) remains a significant global health concern affecting approximately 1\% of births worldwide. Phonocardiography has emerged as a supplementary tool to diagnose CHD cost-effectively. However, the performance of these diagnostic models highly depends on the quality of the phonocardiography, thus, noise reduction is particularly critical. Supervised UNet effectively improves noise reduction capabilities, but limited clean data hinders its application. The complex time-frequency characteristics of phonocardiography further complicate finding the balance between effectively removing noise and preserving pathological features. In this study, we proposed a self-supervised phonocardiography noise reduction model based on Noise2Noise to enable training without clean data. Augmentation and contrastive learning are applied to enhance its performance. We obtained an average SNR of 12.98 dB after filtering under 10~dB of hospital noise. Classification sensitivity after filtering was improved from 27\% to 88\%, indicating its promising pathological feature retention capabilities in practical noisy environments.

eess.SP↗

arXiv2025

LogicSparse: Enabling Engine-Free Unstructured Sparsity for Quantised Deep-learning Accelerators

FPGAs have been shown to be a promising platform for deploying Quantised Neural Networks (QNNs) with high-speed, low-latency, and energy-efficient inference. However, the complexity of modern deep-learning models limits the performance on resource-constrained edge devices. While quantisation and pruning alleviate these challenges, unstructured sparsity remains underexploited due to irregular memory access. This work introduces a framework that embeds unstructured sparsity into dataflow accelerators, eliminating the need for dedicated sparse engines and preserving parallelism. A hardware-aware pruning strategy is introduced to improve efficiency and design flow further. On LeNet-5, the framework attains 51.6 x compression and 1.23 x throughput improvement using only 5.12% of LUTs, effectively exploiting unstructured sparsity for QNN acceleration.

cs.AR↗

arXiv2025

ReTiDe: Real-Time Denoising for Energy-Efficient Motion Picture Processing with FPGAs

Denoising is a core operation in modern video pipelines. In codecs, in-loop filters suppress sensor noise and quantisation artefacts to improve rate-distortion performance; in cinema post-production, denoisers are used for restoration, grain management, and plate clean-up. However, state-of-the-art deep denoisers are computationally intensive and, at scale, are typically deployed on GPUs, incurring high power and cost for real-time, high-resolution streams. This paper presents Real-Time Denoise (ReTiDe), a hardware-accelerated denoising system that serves inference on data-centre Field Programmable Gate Arrays (FPGAs). A compact convolutional model is quantised (post-training quantisation plus quantisation-aware fine-tuning) to INT8 and compiled for AMD Deep Learning Processor Unit (DPU)-based FPGAs. A client-server integration offloads computation from the host CPU/GPU to a networked FPGA service, while remaining callable from existing workflows, e.g., NUKE, without disrupting artist tooling. On representative benchmarks, ReTiDe delivers 37.71$\times$ Giga Operations Per Second (GOPS) throughput and 5.29$\times$ higher energy efficiency than prior FPGA denoising accelerators, with negligible degradation in Peak Signal-to-Noise Ratio (PSNR)/Structural Similarity Index (SSIM). These results indicate that specialised accelerators can provide practical, scalable denoising for both encoding pipelines and post-production, reducing energy per frame without sacrificing quality or workflow compatibility. Code is available at https://github.com/RCSL-TCD/ReTiDe.

eess.IV↗

arXiv2025

Analytic Gravitational Wave Spectrum in Next-to-Minimal Bouncing Cosmology

Bouncing cosmology offers a singularity-free alternative to inflation, but its minimal realization-comprising only four cosmic phases-predicts a simple power-law stochastic gravitational-wave background (SGWB) with a narrow observational window. We introduce the next-to-minimal bouncing cosmology (NMBC), which adds an extra early contraction phase that imprints a broken power-law feature in the SGWB spectrum, enhancing detectability. Using our matrix-representation method grounded in an inequality algebra, we derive a closed-form expression for the NMBC SGWB spectrum. From this analytical result, we show that all NMBC models satisfying the current $\Delta N_{\rm eff}$ bound $\Omega_{\rm GW}h^2(f)<1.7\times10^{-6}$ automatically avoid the trans-Planckian problem, $\rho_{s\downarrow}^{1/4}<0.79\,m_{\rm pl}$. These findings establish the NMBC as a self-consistent, self-contained framework capable of generating a potentially detectable SGWB in both astrophysical and laboratory searches, and demonstrate the broad utility of our matrix-representation method for future SGWB analyses in multi-phase cosmologies.

astro-ph.CO↗

arXiv2025

Forecast Constraints on Bouncing Cosmology from High Frequency Gravitational Waves Using Superconducting LC Circuits and Resonant Cavities

We exploit forecast sensitivities to high frequency gravitational waves (HFGWs) from superconducting LC circuits, traditional resonant cavity and superconducting radio frequency (SRF) cavities with electromagnetic and mechanical modes to derive the first projections of the bounce energy scale within the generic bouncing cosmology framework over the frequency window $1\,$kHz $\lesssim f\lesssim10\,$GHz. In comparison with existing astrophysical limits (spanning $10^{-17}\,$Hz $\lesssim f\lesssim1\,$kHz and based on Planck/BICEP, PTA, and aLIGO/LISA) our HFGW forecasts yield substantially tighter constraints across a broad region of parameter space. This work unifies constraints from cosmological observations and quantum measurement experiments, providing comprehensive coverage of the early Universe gravitational wave spectrum from $10^{-17}\,\mathrm{Hz}$ to $10\,\mathrm{GHz}$ and thereby probing the cosmic initial non singularity at ultra high energy scales.

astro-ph.CO↗

arXiv2025

FAV-NSS: An HIL Framework for Accelerating Validation of Automotive Network Security Strategies

Complex electronic control unit (ECU) architectures, software models and in-vehicle networks are consistently improving safety and comfort functions in modern vehicles. However, the extended functionality and increased connectivity introduce new security risks and vulnerabilities that can be exploited on legacy automotive networks such as the controller area network (CAN). With the rising complexity of vehicular systems and attack vectors, the need for a flexible hardware-in-the-loop (HIL) test fixture that can inject attacks and validate the performance of countermeasures in near-real-world conditions in real time is vital. This paper presents an FPGA-based HIL framework tailored towards validating network security approaches (IDS, IPS) and smart integration strategies of such capabilities for an automotive CAN bus. FAV-NSS replicates an actual vehicular system environment with functional ECUs and network infrastructure on an FPGA, allowing functional validation of IDS/IPS algorithms, accelerator designs and integration schemes (software task on ECU, dedicated accelerator). To show the efficacy of FAV-NSS, we evaluate an IDS accelerator integration problem, both as a traditional coupled accelerator (to the ECU), and secondly close to the CAN controller (mimicking an extended CAN controller). We show that the latter strategy can be fully validated by our framework, which would otherwise require integration of specialised CAN modules into otherwise standard HIL fixtures with ability to instrument internal signals for characterising timing performance. The tests demonstrate a promising latency reduction of 6.3x when compared to the traditional coupled accelerator. Our case study demonstrates the potential of FAV-NSS for accelerating the optimisation, integration and verification of smart ECUs and communication controllers in current and future vehicular systems.

cs.AR↗

arXiv2025

Probing the Bounce Energy Scale in Bouncing Cosmologies with Pulsar Timing Arrays

In this work we constrain the bounce energy scale $\rho_{s\downarrow}^{1/4}$ in a generic framework of bouncing cosmologies using the nanohertz stochastic gravitational-wave background recently detected by pulsar timing arrays (NANOGrav 15-yr, EPTA DR2, PPTA DR3, IPTA DR2). A full Bayesian fit of the analytic SGWB spectrum for this bounce scenario reveals, for the first time, two distinct posterior branches in $(\rho_{s\downarrow}^{1/4},w_1)$: one near $w_1\approx0.3$ and one at $w_1\gg1$, where $w_1$ is the contraction phase equation of state. We find that the bouncing model attains larger Bayes factors against each of six conventional SGWB sources (SMBHBs, inflationary GWs, cosmic strings, domain walls, first order phase transitions, scalar induced GWs), demonstrating strong preference of current PTA data for the bounce hypothesis. Compared to the more generic dual inflation bounce scenario, the concrete bounce realization yields smaller Bayes factors, indicating that PTA measurements impose tighter constraints when the bounce scale is explicit. Moreover, the two posterior branches illuminate distinct theoretical frontiers. The right branch ($w_1\gg1$) violates the dominant energy condition (DEC), thereby providing direct empirical impetus for models with novel early Universe physics, e.g. ghost condensates, higher-derivative or modified gravity operators, and extra dimensional effects. Independently, both branches infer $\rho_{s\downarrow}^{1/4}$ above the Planck scale $M_\mathrm{pl}$, demonstrating that current PTAs already probe trans-Planckian regimes. Together, these findings offer a rare observational window into UV completions of cosmology. We further describe how normalizing flow based machine learning can accelerate such Bayesian analyses as PTA data volumes increase.

astro-ph.CO↗

arXiv2025

Accelerated Bayesian Inference for Pulsar Timing Arrays: Normalizing Flows for Rapid Model Comparison Across Stochastic Gravitational-Wave Background Sources

The recent detection of nanohertz stochastic gravitational-wave backgrounds (SGWBs) by pulsar timing arrays (PTAs) promises unique insights into astrophysical and cosmological origins. However, traditional Markov Chain Monte Carlo (MCMC) approaches become prohibitively expensive for large datasets. We employ a normalizing flow (NF)-based machine learning framework to accelerate Bayesian inference in PTA analyses. For the first time, we perform Bayesian model comparison across SGWB source models in the framework of machine learning by training NF architectures on the PTA dataset (NANOGrav 15-year) and enabling direct evidence estimation via learned harmonic mean estimators. Our examples include 10 conventional SGWB source models such as supermassive black hole binaries, power-law spectrum, cosmic strings, domain walls, scalar-induced GWs, first-order phase transitions, and dual scenario/inflationary gravitational wave. Our approach jointly infers 20 red noise parameters and 2 SGWB parameters per model in $\sim 20$\,hours (including training), compared to $\sim 10$\,days with MCMC. Critically, the NF method preserves rigorous model selection accuracy, with small Hellinger distances ($\lesssim 0.3$) relative to MCMC posteriors, and reproduces MCMC-based Bayes factors across all tested scenarios. This scalable technique for SGWB source comparison will be essential for future PTA expansions and next-generation arrays such as the SKA, offering orders-of-magnitude efficiency gains without sacrificing physical interpretability.

astro-ph.CO↗

arXiv2025

Gravitational Wave Constraints on the Bouncing Energy Scale of Big Bounce Cosmology

Big bounce cosmology provides a solution to the Universe's initial singularity, and stochastic gravitational wave background (SGWB) searches offer a promising avenue for testing this paradigm. In this work, we establish an analytical relation between the bouncing energy scale, $\rho_{s\downarrow}^{1/4}$, and SGWB spectrum, $\Omega_\mathrm{GW}(f)h^2$, for big bounce cosmology. By combining sensitivities from major GW detectors (e.g., Planck/BICEP, PTA, and LIGO/Virgo across low, medium, and high frequencies, respectively), we provide the first systematic GW constraint on $\rho_{s\downarrow}^{1/4}$. Our results show that the region $-\tfrac{1}{3} < w_1 < -0.17$ is excluded by current SGWB searches, given the constraint $\rho_{s\downarrow}^{1/4} > 1~\mathrm{TeV}$, where $w_1$ is the contraction-phase equation of state parameter. Additionally, no detectable SGWB can be generated for $0.038 < w_1 < \infty$ with $\rho_{s\downarrow}^{1/4} < 10^{16}~\mathrm{TeV}$. We identify a window, $-0.17 < w_1 < 0.038$, in which a detectable SGWB can be produced, disfavoring nearly all big bounce models except for the matter-dominated contraction model ($w_1 \simeq 0$).

astro-ph.CO↗

arXiv2024

Multimodal Difference Learning for Sequential Recommendation

Sequential recommendations have drawn significant attention in modeling the user's historical behaviors to predict the next item. With the booming development of multimodal data (e.g., image, text) on internet platforms, sequential recommendation also benefits from the incorporation of multimodal data. Most methods introduce modal features of items as side information and simply concatenates them to learn unified user interests. Nevertheless, these methods encounter the limitation in modeling multimodal differences. We argue that user interests and item relationships vary across different modalities. To address this problem, we propose a novel Multimodal Difference Learning framework for Sequential Recommendation, MDSRec for brevity. Specifically, we first explore the differences in item relationships by constructing modal-aware item relation graphs with behavior signal to enhance item representations. Then, to capture the differences in user interests across modalities, we design a interest-centralized attention mechanism to independently model user sequence representations in different modalities. Finally, we fuse the user embeddings from multiple modalities to achieve accurate item recommendation. Experimental results on five real-world datasets demonstrate the superiority of MDSRec over state-of-the-art baselines and the efficacy of multimodal difference learning.

cs.IR↗

arXiv2024

Primordial Gravitational Waves of Big Bounce Cosmology in Light of Stochastic Gravitational Wave Background

Primordial gravitational waves from the very early stages of the universe, such as inflation or bounce processes, are an irreducible cosmological source of the stochastic gravitational wave background (SGWB). The recent detection of SGWB signals around the nano-Hertz frequency by pulsar timing arrays (PTAs), including NANOGrav, EPTA, PPTA, IPTA, and CPTA, opens a new window to explore these very early stages of the universe through these primordial gravitational waves. In this work, we investigate the generation and evolution of primordial gravitational waves in a generic big bounce cosmology by parameterizing its background evolution into four phases, where perturbation modes exit and re-enter the horizon twice. By analytically solving the equation of motion for primordial gravitational waves and matching solutions at the boundaries, we obtain the explicit form of the primordial gravitational wave spectrum in a generic big bounce cosmology. We find that, according to the evolution of primordial gravitational waves, a generic scenario of big bounce cosmology can be categorized into four distinct types. We introduce four toy models for these categories, demonstrating that our analytical results can be straightforwardly applied to various bouncing universe models in which the equation of state of the background is constant in each phase. We also prospect future applications of our results in interpreting SGWB signals searched by PTAs and upcoming advanced gravitational wave detectors such as SKA, Taiji, Tianqin, LISA, DECIGO, and aLIGO/Virgo/KAGRA using Bayesian analysis.

astro-ph.CO↗

arXiv2024

Dual Inflation and Bounce Cosmologies Interpretation of Pulsar Timing Array Data

We explore a dual scenario of generalized inflation and bounce cosmologies, producing a scale-invariant curvature perturbation spectrum. Bayesian analysis with pulsar timing array data identifies, for the first time, viable regions from inflation and bounce that simultaneously explain stochastic gravitational wave background (SGWB) signals and CMB anisotropies. Bayes factor calculations strongly favor this dual scenario over conventional sources and provide initial evidence of a duality between inflation and bounce regarding SGWB, offering new insights for early universe model-building and future observations.

astro-ph.CO↗

arXiv2022

Kinetically stabilized inflation

In this work, we propose a string-inspired two fields inflation model to address the fine-tuning problem that the standard inflation model suffers. The fast-rolling tachyon $\mathcal{T}$ originated from the D-brane and anti-D-brane pair annihilation locks the inflaton $φ$ slowly rolling on a Higgs-like potential $V(φ)=-m_φ^2φ^2+λφ^4$ and drives a kinetically stabilized (KS) inflation. Our numerical simulation confirms such a solution is a dynamic attractor. In particular, for $λ< 0.8\times 10^{-3}$, the e-folding number contributed by the KS inflation phase can be larger than $62$ to solve the horizon and flatness problems of Big Bang theory. Notably, this KS inflation generates a nearly scale-invariant primordial curvature perturbations spectrum consistent with current cosmic microwave background (CMB) observations. It predicts a low tensor-to-scalar ratio, which the current primordial gravitational wave background (the B-modes in CMB) searches favor.

astro-ph.CO↗

arXiv2022

Domain Wall Network: A Dual Solution for Gravitational Waves and Hubble Tension?

We explore the possibility that domain wall networks generate the stochastic gravitational wave background (SGWB) observed as a strong common power-law process in the Data Release-2 of Parkes Pulsar Timing Array. We find that a broad range of parameters, specifically wall tension around $\sigma_{\textrm{DW}} \sim (29-414 , \textrm{TeV})^3$ and wall-decay temperature within $T_d \sim 20-257 , \textrm{MeV}$, can explain this phenomenon at a $68\%$ credible level. Meanwhile, the same parameters could ease the Hubble tension if particles from these domain wall networks decay into dark radiation. We establish a direct analytical relationship, $\Omega_{\textrm{GW}}(f_p,T_0) h^2 \sim \Omega_{\textrm{rad}} h^2 ( \Omega_\nu \Delta N_{\textrm{eff}})^2$, to illustrate this coincidence, underlining its importance in the underlying physics and potential applicability to a wider range of models and data. Conversely, if the common power-law process is not attributed to domain wall networks, our findings impose tight limits on the wall tension and decay temperature.

hep-ph↗

arXiv2022

Perturbative Resonance in WIMP paradigm and its Cosmological Implications on Cosmic Reheating and Primordial Gravitational Wave Detection

We investigate the co-evolution of dark matter (DM) density perturbation and metric perturbation in the WIMP paradigm. Instead of adopting the conventional assumption that DM starts out in thermal equilibrium, we propose a simple phase of DM production for the WIMP paradigm and extend our analysis to this phase. Being free from the envelop of thermal equilibrium, an amplified perturbative resonance between DM density perturbation and scalar modes of metric perturbation takes place during the DM production phase, and consequently results in a suppression of the tensor-to-scalar ratio of metric perturbation. By specifying the cosmic background with a typical realization of cosmic reheating, we establish a relation between DM particle mass $m_χ$ and the tensor-to-scalar ratio $r$ in the WIMP paradigm, which also contains two reheating parameters, the reheating temperature $T_{R_f}$ and the dissipative constant $Γ_0$. Notably, for a sizeable parameter region of WIMP candidate and cosmic reheating, this relation predicts a smaller value of $r$ in comparing with the conventional expectation obtained by assuming DM starts out in thermal equilibrium. Once the suppression of $r$ is measured in future observations of primordial gravitational wave in CMB, this relation can be used to constrain $m_χ$, $T_{R_f}$ and $Γ_0$ in principle.

astro-ph.CO↗