arXiv subjects

Miao Sun

Publications and source records attributed to Miao Sun.

At least 19 recordsLinked to original sources

arXiv2026

Uni-SFU: Algorithm-HW Co-Design for Universal SFUs via Mixed-Degree Piecewise Approximation

Nonlinear activation functions are essential to modern deep neural networks (DNNs), but their hardware evaluation places significant pressure on the special-function units (SFUs) of GPUs and custom accelerators. Therefore, piecewise polynomial approximations are commonly used within allowed error bounds to improve computational efficiency. However, existing techniques often approximate each activation function in isolation using fixed-degree polynomials and uniform segments, leading to hardware redundancy and sub-optimal precision. To address these limitations, we present Uni-SFU, an algorithm-hardware co-design framework that jointly optimizes approximation accuracy and silicon area for a diverse set of activation functions. Uni-SFU leverages a joint search across all target functions to assign mixed-degree polynomials to nonuniform segments, guided by an RTL-derived area cost model. This approach identifies a unified hardware configuration to implement the target activation functions under given accuracy constraints. Validated across over 700 neural network variants and three Natural Language Processing (NLP) models, Uni-SFU achieves a superior Mean Squared Error (MSE) below 8.22x10^-8, limiting top-1 accuracy degradation to within 1.02% compared to floating-point baselines. The proposed design occupies only 6,800 um2 in GF 22nm CMOS technology, achieving a superior trade-off between silicon area and system-level accuracy compared to SOTA counterparts.

cs.AR↗

arXiv2026

ReVolt: Power Delivery Network-Aware Voltage Droop Control for 2.5D PIM Chiplet Architectures

Processing-in-memory (PIM)-based 2.5D multi-chiplet platforms are enablers for machine learning (ML) workloads. However, their performance is affected by the power delivery network (PDN), where varying chiplet-level current demand induces spatially and temporally varying voltage droop. These droop events lead to voltage violations, degrades system performance, and impact inference accuracy for ML workloads. In this work, we propose ReVolt, a dynamic operation unit (OU)-based framework for mitigating voltage droop in PIM-based multi-chiplet systems. ReVolt leverages an LSTM-based PDN surrogate to predict per-chiplet supply voltage trajectories at runtime, enabling proactive adjustment of OU size to mitigate droop events. By treating OU size as a control knob, ReVolt regulates chiplet-level current demand while maintaining computational accuracy. This approach prevents voltage droop violations and improves energy-delay product (EDP) while preserving ML model inference accuracy. Experimental results demonstrate that ReVolt prevents voltage droop violations while achieving an average 76x reduction in EDP compared to existing fixed and dynamic OU-based baselines, without compromising inference accuracy of ML models.

cs.AR↗

arXiv2026

End-to-End Markov State Sequence Learning for Auditory Attention Decoding

Auditory attention decoding (AAD) identifies the speaker a listener attends to from neural responses like electroencephalography (EEG), making it a key algorithm in neuro-steered hearing aids. However, most neural AAD models are trained as independent short-window classifiers, despite auditory attention being a temporally persistent cognitive state and short-window EEG--audio evidence often being noisy and ambiguous. We propose an end-to-end Markov AAD framework based on conditional random field (CRF) that trains window-level neural emissions under a two-state attention prior. The framework treats the logits of any AAD backbone as Markov emissions, learns the transition rate from a standard HMM initialization, and jointly optimizes cross-entropy and CRF objectives, allowing temporal continuity to guide representation learning rather than merely smoothing predictions after training. We also introduce ESCNet, an EEG--speech correlation backbone that preserves time-aligned features and converts the difference between two mean Pearson correlations into state logits. We evaluate the framework with four emission backbones spanning correlation-based, convolutional, recurrent, and attention-based designs. On the dynamic AVGC dataset, CRF training generally outperforms post-hoc HMM smoothing; with ESCNet, it achieves $86.5\%$ causal and $92.4\%$ non-causal accuracy using $1$s windows. On the static KUL and USTC datasets, it improves causal decoding over fixed-rate post-hoc HMM baselines by $5.6\%$ and $2.0\%$, respectively, showing the superiority of learning AAD as attention state sequence over isolated-window classification.

cs.SD↗

arXiv2026

LEXI: Lossless Exponent Coding for Efficient Inter-Chiplet Communication in Hybrid LLMs

Data movement overheads increase the inference latency of state-of-the-art large language models (LLMs). These models commonly use the bfloat16 (BF16) format for stable training. Floating-point standards allocate eight bits to the exponent, but our profiling reveals that exponent streams exhibit fewer than 3 bits Shannon entropy, indicating high inherent compressibility. To exploit this potential, we propose LEXI, a novel lossless exponent compression scheme based on Huffman coding. LEXI compresses activations and caches on the fly while storing compressed weights for just-in-time decompression near compute, without sacrificing system throughput and model accuracy. The codecs at the ingress and egress ports of network-on-chip routers sustain the maximum link bandwidth via multi-lane LUT decoders, incurring only 0.09 percent area and energy overheads with GF 22 nm technology. LEXI reduces inter-chiplet communication and end-to-end inference latencies by 33-45 percent and 30-35 percent on modern Jamba, Zamba, and Qwen LLMs implemented on a homogeneous chiplet architecture.

cs.AR↗

arXiv2025

eMamba: Efficient Acceleration Framework for Mamba Models in Edge Computing

State Space Model (SSM)-based machine learning architectures have recently gained significant attention for processing sequential data. Mamba, a recent sequence-to-sequence SSM, offers competitive accuracy with superior computational efficiency compared to state-of-the-art transformer models. While this advantage makes Mamba particularly promising for resource-constrained edge devices, no hardware acceleration frameworks are currently optimized for deploying it in such environments. This paper presents eMamba, a comprehensive end-to-end hardware acceleration framework explicitly designed for deploying Mamba models on edge platforms. eMamba maximizes computational efficiency by replacing complex normalization layers with lightweight hardware-aware alternatives and approximating expensive operations, such as SiLU activation and exponentiation, considering the target applications. Then, it performs an approximation-aware neural architecture search (NAS) to tune the learnable parameters used during approximation. Evaluations with Fashion-MNIST, CIFAR-10, and MARS, an open-source human pose estimation dataset, show eMamba achieves comparable accuracy to state-of-the-art techniques using 1.63-19.9$\times$ fewer parameters. In addition, it generalizes well to large-scale natural language tasks, demonstrating stable perplexity across varying sequence lengths on the WikiText2 dataset. We also quantize and implement the entire eMamba pipeline on an AMD ZCU102 FPGA and ASIC using GlobalFoundries (GF) 22 nm technology. Experimental results show 4.95-5.62$\times$ lower latency and 2.22-9.95$\times$ higher throughput, with 4.77$\times$ smaller area, 9.84$\times$ lower power, and 48.6$\times$ lower energy consumption than baseline solutions while maintaining competitive accuracy.

cs.LG↗

arXiv2024

A Lightweight and Real-Time Binaural Speech Enhancement Model with Spatial Cues Preservation

Binaural speech enhancement (BSE) aims to jointly improve the speech quality and intelligibility of noisy signals received by hearing devices and preserve the spatial cues of the target for natural listening. Existing methods often suffer from the compromise between noise reduction (NR) capacity and spatial cues preservation (SCP) accuracy and a high computational demand in complex acoustic scenes. In this work, we present a learning-based lightweight binaural complex convolutional network (LBCCN), which excels in NR by filtering low-frequency bands and keeping the rest. Additionally, our approach explicitly incorporates the estimation of interchannel relative acoustic transfer function to ensure the spatial cues fidelity and speech clarity. Results show that the proposed LBCCN can achieve a comparable NR performance to state-of-the-art methods under fixed-speaker conditions, but with a much lower computational cost and a certain degree of SCP capability. The reproducible code and audio examples are available at https://github.com/jywanng/LBCCN.

cs.SD↗

arXiv2024

SDformer: Efficient End-to-End Transformer for Depth Completion

Depth completion aims to predict dense depth maps with sparse depth measurements from a depth sensor. Currently, Convolutional Neural Network (CNN) based models are the most popular methods applied to depth completion tasks. However, despite the excellent high-end performance, they suffer from a limited representation area. To overcome the drawbacks of CNNs, a more effective and powerful method has been presented: the Transformer, which is an adaptive self-attention setting sequence-to-sequence model. While the standard Transformer quadratically increases the computational cost from the key-query dot-product of input resolution which improperly employs depth completion tasks. In this work, we propose a different window-based Transformer architecture for depth completion tasks named Sparse-to-Dense Transformer (SDformer). The network consists of an input module for the depth map and RGB image features extraction and concatenation, a U-shaped encoder-decoder Transformer for extracting deep features, and a refinement module. Specifically, we first concatenate the depth map features with the RGB image features through the input model. Then, instead of calculating self-attention with the whole feature maps, we apply different window sizes to extract the long-range depth dependencies. Finally, we refine the predicted features from the input module and the U-shaped encoder-decoder Transformer module to get the enriching depth features and employ a convolution layer to obtain the dense depth map. In practice, the SDformer obtains state-of-the-art results against the CNN-based depth completion models with lower computing loads and parameters on the NYU Depth V2 and KITTI DC datasets.

cs.CV↗

arXiv2024

Sub-SA: Strengthen In-context Learning via Submodular Selective Annotation

In-context learning (ICL) leverages in-context examples as prompts for the predictions of Large Language Models (LLMs). These prompts play a crucial role in achieving strong performance. However, the selection of suitable prompts from a large pool of labeled examples often entails significant annotation costs. To address this challenge, we propose Sub-SA (Submodular Selective Annotation), a submodule-based selective annotation method. The aim of Sub-SA is to reduce annotation costs while improving the quality of in-context examples and minimizing the time consumption of the selection process. In Sub-SA, we design a submodular function that facilitates effective subset selection for annotation and demonstrates the characteristics of monotonically and submodularity from the theoretical perspective. Specifically, we propose RPR (Reward and Penalty Regularization) to better balance the diversity and representativeness of the unlabeled dataset attributed to a reward term and a penalty term, respectively. Consequently, the selection for annotations can be effectively addressed with a simple yet effective greedy search algorithm based on the submodular function. Finally, we apply the similarity prompt retrieval to get the examples for ICL.

cs.LG↗

arXiv2024

TimeLDM: Latent Diffusion Model for Unconditional Time Series Generation

Time series generation is a crucial research topic in the area of decision-making systems, which can be particularly important in domains like autonomous driving, healthcare, and, notably, robotics. Recent approaches focus on learning in the data space to model time series information. However, the data space often contains limited observations and noisy features. In this paper, we propose TimeLDM, a novel latent diffusion model for high-quality time series generation. TimeLDM is composed of a variational autoencoder that encodes time series into an informative and smoothed latent content and a latent diffusion model operating in the latent space to generate latent information. We evaluate the ability of our method to generate synthetic time series with simulated and real-world datasets and benchmark the performance against existing state-of-the-art methods. Qualitatively and quantitatively, we find that the proposed TimeLDM persistently delivers high-quality generated time series. For example, TimeLDM achieves new state-of-the-art results on the simulated benchmarks and an average improvement of 55% in Discriminative score with all benchmarks. Further studies demonstrate that our method yields more robust outcomes across various lengths of time series data generation. Especially, for the Context-FID score and Discriminative score, TimeLDM realizes significant improvements of 80% and 50%, respectively. The code will be released after publication.

cs.LG↗

arXiv2024

LiDAR-PTQ: Post-Training Quantization for Point Cloud 3D Object Detection

Due to highly constrained computing power and memory, deploying 3D lidar-based detectors on edge devices equipped in autonomous vehicles and robots poses a crucial challenge. Being a convenient and straightforward model compression approach, Post-Training Quantization (PTQ) has been widely adopted in 2D vision tasks. However, applying it directly to 3D lidar-based tasks inevitably leads to performance degradation. As a remedy, we propose an effective PTQ method called LiDAR-PTQ, which is particularly curated for 3D lidar detection (both SPConv-based and SPConv-free). Our LiDAR-PTQ features three main components, \textbf{(1)} a sparsity-based calibration method to determine the initialization of quantization parameters, \textbf{(2)} a Task-guided Global Positive Loss (TGPL) to reduce the disparity between the final predictions before and after quantization, \textbf{(3)} an adaptive rounding-to-nearest operation to minimize the layerwise reconstruction error. Extensive experiments demonstrate that our LiDAR-PTQ can achieve state-of-the-art quantization performance when applied to CenterPoint (both Pillar-based and Voxel-based). To our knowledge, for the very first time in lidar-based 3D detection tasks, the PTQ INT8 model's accuracy is almost the same as the FP32 model while enjoying $3\times$ inference speedup. Moreover, our LiDAR-PTQ is cost-effective being $30\times$ faster than the quantization-aware training method. Code will be released at \url{https://github.com/StiphyJay/LiDAR-PTQ}.

cs.CV↗

arXiv2020

Design of on-chip plasmonic modulator with vanadium-dioxide in hybrid orthogonal junctions on Silicon-on-Insulator

We present a plasmonic modulator based on hybrid orthogonal silver junctions using vanadium dioxide as the modulating material on the silicon on insulator. The modulator has an ultra-compact footprint of 1.8μm x 1μm with a 100nm x 100nm modulating section based on the orthogonal geometry. We take advantage of large change in the refractive index of vanadium dioxide during its phase transition to achieve a high modulation depth of 46.89dB/μm. We also provide a fabrication strategy for the development of this device. The device geometry has potential applications in the development of next generation high frequency photonic modulators for optical communications which require a nanometer scale footprint, large modulation depth and small insertion losses.

physics.optics↗

arXiv2019

Graph-Time Spectral Analysis for Atrial Fibrillation

Atrial fibrillation is a clinical arrhythmia with multifactorial mechanisms still unresolved. Time-frequency analysis of epicardial electrograms has been investigated to study atrial fibrillation. However, deeper understanding of atrial fibrillation can be achieved if the spatial dimension can be incorporated. Unfortunately, the physical models describing the spatial relations of atrial fibrillation signals are complex and non-linear; hence, the conventional signal processing techniques to study electrograms in the joint space, time, and frequency domain are less suitable. In this study, we wish to put forward a radically different approach to analyze atrial fibrillation with a higher-level model. This approach relies on graph signal processing to represent the spatial relations between epicardial electrograms and put forward a graph-time spectral analysis for atrial fibrillation. To capture the frequency content along both the time and graph domain, we proposed the joint graph and short-time Fourier transform. The latter allows us to analyze the spatial variability of the electrogram temporal frequencies. With this technique, we have found that the spatial variation of the atrial electrograms decreases during atrial fibrillation due to the reduction of the high temporal frequencies of the atrial waves. The proposed analysis further confirms that the ventricular activity is smoother over the atrial area compared with the atrial activity. Besides using the proposed graph-time analysis to conduct a first study on atrial fibrillation, we applied it to the cancellation of ventricular activity from atrial electrograms. Experimental results on simulated and real data further corroborate the findings in this atrial fibrillation study.

eess.SP↗

arXiv2017

EddyNet: A Deep Neural Network For Pixel-Wise Classification of Oceanic Eddies

This work presents EddyNet, a deep learning based architecture for automated eddy detection and classification from Sea Surface Height (SSH) maps provided by the Copernicus Marine and Environment Monitoring Service (CMEMS). EddyNet is a U-Net like network that consists of a convolutional encoder-decoder followed by a pixel-wise classification layer. The output is a map with the same size of the input where pixels have the following labels \{'0': Non eddy, '1': anticyclonic eddy, '2': cyclonic eddy\}. We investigate the use of SELU activation function instead of the classical ReLU+BN and we use an overlap based loss function instead of the cross entropy loss. Keras Python code, the training datasets and EddyNet weights files are open-source and freely available on https://github.com/redouanelg/EddyNet.

cs.CV↗

arXiv2017

ProLanGO: Protein Function Prediction Using Neural~Machine Translation Based on a Recurrent Neural Network

With the development of next generation sequencing techniques, it is fast and cheap to determine protein sequences but relatively slow and expensive to extract useful information from protein sequences because of limitations of traditional biological experimental techniques. Protein function prediction has been a long standing challenge to fill the gap between the huge amount of protein sequences and the known function. In this paper, we propose a novel method to convert the protein function problem into a language translation problem by the new proposed protein sequence language "ProLan" to the protein function language "GOLan", and build a neural machine translation model based on recurrent neural networks to translate "ProLan" language to "GOLan" language. We blindly tested our method by attending the latest third Critical Assessment of Function Annotation (CAFA 3) in 2016, and also evaluate the performance of our methods on selected proteins whose function was released after CAFA competition. The good performance on the training and testing datasets demonstrates that our new proposed method is a promising direction for protein function prediction. In summary, we first time propose a method which converts the protein function prediction problem to a language translation problem and applies a neural machine translation model for protein function prediction.

q-bio.QM↗

arXiv2017

Age Group and Gender Estimation in the Wild with Deep RoR Architecture

Automatically predicting age group and gender from face images acquired in unconstrained conditions is an important and challenging task in many real-world applications. Nevertheless, the conventional methods with manually-designed features on in-the-wild benchmarks are unsatisfactory because of incompetency to tackle large variations in unconstrained images. This difficulty is alleviated to some degree through Convolutional Neural Networks (CNN) for its powerful feature representation. In this paper, we propose a new CNN based method for age group and gender estimation leveraging Residual Networks of Residual Networks (RoR), which exhibits better optimization ability for age group and gender classification than other CNN architectures.Moreover, two modest mechanisms based on observation of the characteristics of age group are presented to further improve the performance of age estimation.In order to further improve the performance and alleviate over-fitting problem, RoR model is pre-trained on ImageNet firstly, and then it is fune-tuned on the IMDB-WIKI-101 data set for further learning the features of face images, finally, it is used to fine-tune on Adience data set. Our experiments illustrate the effectiveness of RoR method for age and gender estimation in the wild, where it achieves better performance than other CNN methods. Finally, the RoR-152+IMDB-WIKI-101 with two mechanisms achieves new state-of-the-art results on Adience benchmark.

cs.CV↗

arXiv2017

Residual Networks of Residual Networks: Multilevel Residual Networks

A residual-networks family with hundreds or even thousands of layers dominates major image recognition tasks, but building a network by simply stacking residual blocks inevitably limits its optimization ability. This paper proposes a novel residual-network architecture, Residual networks of Residual networks (RoR), to dig the optimization ability of residual networks. RoR substitutes optimizing residual mapping of residual mapping for optimizing original residual mapping. In particular, RoR adds level-wise shortcut connections upon original residual networks to promote the learning capability of residual networks. More importantly, RoR can be applied to various kinds of residual networks (ResNets, Pre-ResNets and WRN) and significantly boost their performance. Our experiments demonstrate the effectiveness and versatility of RoR, where it achieves the best performance in all residual-network-like structures. Our RoR-3-WRN58-4+SD models achieve new state-of-the-art results on CIFAR-10, CIFAR-100 and SVHN, with test errors 3.77%, 19.73% and 1.59%, respectively. RoR-3 models also achieve state-of-the-art results compared to ResNets on ImageNet data set.

cs.CV↗

arXiv2016

Latent Model Ensemble with Auto-localization

Deep Convolutional Neural Networks (CNN) have exhibited superior performance in many visual recognition tasks including image classification, object detection, and scene label- ing, due to their large learning capacity and resistance to overfit. For the image classification task, most of the current deep CNN- based approaches take the whole size-normalized image as input and have achieved quite promising results. Compared with the previously dominating approaches based on feature extraction, pooling, and classification, the deep CNN-based approaches mainly rely on the learning capability of deep CNN to achieve superior results: the burden of minimizing intra-class variation while maximizing inter-class difference is entirely dependent on the implicit feature learning component of deep CNN; we rely upon the implicitly learned filters and pooling component to select the discriminative regions, which correspond to the activated neurons. However, if the irrelevant regions constitute a large portion of the image of interest, the classification performance of the deep CNN, which takes the whole image as input, can be heavily affected. To solve this issue, we propose a novel latent CNN framework, which treats the most discriminate region as a latent variable. We can jointly learn the global CNN with the latent CNN to avoid the aforementioned big irrelevant region issue, and our experimental results show the evident advantage of the proposed latent CNN over traditional deep CNN: latent CNN outperforms the state-of-the-art performance of deep CNN on standard benchmark datasets including the CIFAR-10, CIFAR- 100, MNIST and PASCAL VOC 2007 Classification dataset.

cs.CV↗

arXiv2016

Multiple Instance Learning Convolutional Neural Networks for Object Recognition

Convolutional Neural Networks (CNN) have demon- strated its successful applications in computer vision, speech recognition, and natural language processing. For object recog- nition, CNNs might be limited by its strict label requirement and an implicit assumption that images are supposed to be target- object-dominated for optimal solutions. However, the labeling procedure, necessitating laying out the locations of target ob- jects, is very tedious, making high-quality large-scale dataset prohibitively expensive. Data augmentation schemes are widely used when deep networks suffer the insufficient training data problem. All the images produced through data augmentation share the same label, which may be problematic since not all data augmentation methods are label-preserving. In this paper, we propose a weakly supervised CNN framework named Multiple Instance Learning Convolutional Neural Networks (MILCNN) to solve this problem. We apply MILCNN framework to object recognition and report state-of-the-art performance on three benchmark datasets: CIFAR10, CIFAR100 and ILSVRC2015 classification dataset.

cs.CV↗