arXiv subjects

Jianhang Xie

Publications and source records attributed to Jianhang Xie.

9 recordsLinked to original sources

arXiv2026

ATLAS: Automated Approximation of Transformers for Efficient Homomorphic Inference in One Hour

Fully homomorphic encryption (FHE) lets a server run inference on encrypted data with strong privacy guarantees, but running a Transformer under FHE is expensive. Its non-linear operations, such as softmax, normalization, and activation, must be replaced with polynomial approximations that the CKKS scheme supports, and the depth of these approximations dominates inference cost. Existing FHE Transformers use hand-tuned approximation settings, such as iteration count and polynomial degree, applied uniformly across layers, models, and tasks. Hand-tuning is slow and error-prone. Even a single uniform setting has about $10^7$ choices, and manual search cannot exploit layer-wise variation. AutoFHE, the only automated method with multi-objective search, targets ReLU-only CNNs and needs full fine-tuning per candidate, which is too costly for Transformers. Per-layer settings also push the search space to about $10^{85}$ for BERT and ViT and $10^{228}$ for LLaMA3, beyond both manual and fine-tuning-based search. We present ATLAS, a training-free framework that automates this search by treating each layer's approximation setting as a multi-objective optimization over latency and accuracy. The problem is hard: the decision space is large (96 or 256 variables), each configuration takes 70 to 1,000 seconds to evaluate even in cleartext, and 85 to 90 percent of configurations are invalid. ATLAS handles this with a two-stage optimization strategy and a surrogate model, completing the search in about one hour. Compared to an iterative softmax baseline, ATLAS cuts multiplicative depth and end-to-end latency by about 35 percent with little accuracy loss, and works across encoder-only, decoder-only, and vision Transformers, complementing parallel work on packing and matrix multiplication.

cs.CR↗

arXiv2026

Dynamics of a Predator-Prey Model with Allee Effect and Interspecific Competition

This paper primarily discusses the dynamical properties of a class of Lotka-Volterra models featuring the Allee effect and interspecific competition within the predator population. The constructed models employ Holling II and Holling I response functions for the predator, respectively.The existence of boundary equilibrium points under various parameter conditions and internal equilibrium points under specific parameter conditions is discussed. The equilibrium points of the system may be stable or unstable nodes, saddle points, saddle-nodes, or cusp points with a codimension of 2. The parameter conditions under which internal equilibrium points possess one zero eigenvalue and two non-zero eigenvalues, one zero eigenvalue and a pair of purely imaginary eigenvalues, or two zero eigenvalues and one non-zero eigenvalue are analyzed.

math.DS↗

arXiv2026

GroupNL: Low-Resource and Robust CNN Design over Cloud and Device

Deploying Convolutional Neural Network (CNN) models on ubiquitous Internet of Things (IoT) devices in a cloud-assisted manner to provide users with a variety of high-quality services has become mainstream. Most existing studies speed up model cloud training/on-device inference by reducing the number of convolution (Conv) parameters and floating-point operations (FLOPs). However, they usually employ two or more lightweight operations (e.g., depthwise Conv, $1\times1$ cheap Conv) to replace a Conv, which can still affect the model's speedup even with fewer parameters and FLOPs. To this end, we propose the Grouped NonLinear transformation generation method (GroupNL), leveraging data-agnostic, hyperparameters-fixed, and lightweight Nonlinear Transformation Functions (NLFs) to generate diversified feature maps on demand via grouping, thereby reducing resource consumption while improving the robustness of CNNs. First, in a GroupNL Conv layer, a small set of feature maps, i.e., seed feature maps, are generated based on the seed Conv operation. Then, we split seed feature maps into several groups, each with a set of different NLFs, to generate the required number of diversified feature maps with tensor manipulation operators and nonlinear processing in a lightweight manner without additional Conv operations. We further introduce a sparse GroupNL Conv to speed up by reasonably designing the seed Conv groups between the number of input channels and seed feature maps. Experiments conducted on benchmarks and on-device resource measurements demonstrate that the GroupNL Conv is an impressive alternative to Conv layers in baseline models. Specifically, on Icons-50 dataset, the accuracy of GroupNL-ResNet-18 is 2.86% higher than ResNet-18; on ImageNet-C dataset, the accuracy of GroupNL-EfficientNet-ES achieves about 1.1% higher than EfficientNet-ES.

cs.CV↗

arXiv2025

NestQuant: Post-Training Integer-Nesting Quantization for On-Device DNN

Deploying quantized deep neural network (DNN) models with resource adaptation capabilities on ubiquitous Internet of Things (IoT) devices to provide high-quality AI services can leverage the benefits of compression and meet multi-scenario resource requirements. However, existing dynamic/mixed precision quantization requires retraining or special hardware, whereas post-training quantization (PTQ) has two limitations for resource adaptation: (i) The state-of-the-art PTQ methods only provide one fixed bitwidth model, which makes it challenging to adapt to the dynamic resources of IoT devices; (ii) Deploying multiple PTQ models with diverse bitwidths consumes large storage resources and switching overheads. To this end, this paper introduces a resource-friendly post-training integer-nesting quantization, i.e., NestQuant, for on-device quantized model switching on IoT devices. The proposed NestQuant incorporates the integer weight decomposition, which bit-wise splits quantized weights into higher-bit and lower-bit weights of integer data types. It also contains a decomposed weights nesting mechanism to optimize the higher-bit weights by adaptive rounding and nest them into the original quantized weights. In deployment, we can send and store only one NestQuant model and switch between the full-bit/part-bit model by paging in/out lower-bit weights to adapt to resource changes and reduce consumption. Experimental results on the ImageNet-1K pretrained DNNs demonstrated that the NestQuant model can achieve high performance in top-1 accuracy, and reduce in terms of data transmission, storage consumption, and switching overheads. In particular, the ResNet-101 with INT8 nesting INT6 can achieve 78.1% and 77.9% accuracy for full-bit and part-bit models, respectively, and reduce switching overheads by approximately 78.1% compared with diverse bitwidths PTQ models.

cs.LG↗

arXiv2025

Complex dynamics of a predator-prey model with constant-yield prey harvesting and Allee effect in predator

This paper investigates the dynamical behaviors of a Holling type I Leslie-Gower predator-prey model where the predator exhibits an Allee effect and is subjected to constant harvesting. The model demonstrates three types of equilibrium points under different parameter conditions, which could be either stable or unstable nodes (foci), saddle nodes, weak centers, or cusps. The system exhibits a saddle-node bifurcation near the saddle-node point and a Hopf bifurcation near the weak center. By calculating the first Lyapunov coefficient, the conditions for the occurrence of both supercritical and subcritical Hopf bifurcations are derived. Finally, it is proven that when the predator growth rate and the prey capture coefficient vary within a specific small neighborhood, the system undergoes a codimension-2 Bogdanov-Takens bifurcation near the cusp point.

math.DS↗

arXiv2025

A study on a class of predator-prey models with Allee effect

math.DS↗

arXiv2024

LoRA-C: Parameter-Efficient Fine-Tuning of Robust CNN for IoT Devices

Efficient fine-tuning of pre-trained convolutional neural network (CNN) models using local data is essential for providing high-quality services to users using ubiquitous and resource-limited Internet of Things (IoT) devices. Low-Rank Adaptation (LoRA) fine-tuning has attracted widespread attention from industry and academia because it is simple, efficient, and does not incur any additional reasoning burden. However, most of the existing advanced methods use LoRA to fine-tune Transformer, and there are few studies on using LoRA to fine-tune CNN. The CNN model is widely deployed on IoT devices for application due to its advantages in comprehensive resource occupancy and performance. Moreover, IoT devices are widely deployed outdoors and usually process data affected by the environment (such as fog, snow, rain, etc.). The goal of this paper is to use LoRA technology to efficiently improve the robustness of the CNN model. To this end, this paper first proposes a strong, robust CNN fine-tuning method for IoT devices, LoRA-C, which performs low-rank decomposition in convolutional layers rather than kernel units to reduce the number of fine-tuning parameters. Then, this paper analyzes two different rank settings in detail and observes that the best performance is usually achieved when $α/{r}$ is a constant in either standard data or corrupted data. This discovery provides experience for the widespread application of LoRA-C. Finally, this paper conducts many experiments based on pre-trained models. Experimental results on CIFAR-10, CIFAR-100, CIFAR-10-C, and Icons50 datasets show that the proposed LoRA-Cs outperforms standard ResNets. Specifically, on the CIFAR-10-C dataset, the accuracy of LoRA-C-ResNet-101 achieves 83.44% accuracy, surpassing the standard ResNet-101 result by +9.5%.

cs.DC↗

arXiv2023

Estimating Stellar Parameters and Identifying Very Metal-poor Stars Using Convolutional Neural Networks for Low-resolution Spectra (R~200)

Very metal-poor (VMP, [Fe/H]<-2.0) stars offer a wealth of information on the nature and evolution of elemental production in the early galaxy and universe. The upcoming China Space Station Telescope (CSST) will provide us with a large amount of spectroscopic data that may contain plenty of VMP stars, and thus it is crucial to determine the stellar atmospheric parameters ($T_{eff}$, $\log g$, and [Fe/H]) for low-resolution spectra similar to the CSST spectra (R~200). In this paper, a two-dimensional Convolutional Neural Network (CNN) model with three convolutional layers and two fully connected layers is constructed. The principal aim of this work is to measure the ability of this model to estimate stellar parameters on low-resolution (R~200) spectra and to identify VMP stars so that we can better search for VMP stars in the spectra observed by CSST.We mainly use 10,008 observed spectra of VMP stars from LAMOST DR3, and 16,638 spectra of common stars ([Fe/H]>-2.0) from LAMOST DR8 for the experiment and make comparisons. All spectra are reduced to R~200 to match the resolution of the CSST and are preprocessed and collapsed into two-dimensional spectra for input to the CNN model. The results show that the MAE values are 99.40 K for $T_{eff}$, 0.22 dex for $\log g$, 0.14 dex for [Fe/H], and 0.26 dex for [C/Fe], respectively. Besides, the CNN model efficiently identifies VMP stars with a precision of 94.77%. The validation and practicality of this model are also tested on the MARCS synthetic spectra. This paper powerfully demonstrates the effectiveness of the proposed CNN model in estimating stellar parameters for low-resolution spectra (R~200) and recognizing VMP stars that are of interest for stellar population and galactic evolution work.

astro-ph.SR↗

arXiv2021

Determine the Masses and Ages of Red Giant Branch Stars from Low-resolution LAMOST Spectra Using DenseNet

We propose a new model to determine the ages and masses of red giant branch (RGB) stars from the low-resolution large sky area multi-object fiber spectroscopic telescope (LAMOST) spectra. The ages of RGB stars are difficult to determine using classical isochrone fitting techniques in the Hertzsprung-Russell diagram, because isochrones of RGB stars are tightly crowned. With the help of the asteroseismic method, we can determine the masses and ages of RGB stars accurately. Using the ages derived from the asteroseismic method, we train a deep learning model based on DenseNet to calculate the ages of RGB stars directly from their spectra. We then apply this model to determine the ages of 512 272 RGB stars from LAMOST DR7 spectra (see http://dr7.lamost.org/). The results show that our model can estimate the ages of RGB stars from low-resolution spectra with an accuracy of 24.3%. The results on the open clusters M 67, Berkeley 32, and NGC 2420 show that our model performs well in estimating the ages of RGB stars. Through comparison, we find that our method performs better than other methods in determining the ages of RGB stars. The proposed method can be used in the stellar parameter pipeline of upcoming large surveys such as 4MOST, WEAVES, and MOONS.

astro-ph.SR↗