Searcharxiv⌕ Search

arXiv subjects

Yuhang Li

Publications and source records attributed to Yuhang Li.

At least 109 records · Page 6Linked to original sources
arXiv2023

MINT: Multiplier-less INTeger Quantization for Energy Efficient Spiking Neural Networks

We propose Multiplier-less INTeger (MINT) quantization, a uniform quantization scheme that efficiently compresses weights and membrane potentials in spiking neural networks (SNNs). Unlike previous SNN quantization methods, MINT quantizes memory-intensive membrane potentials to an extremely low precision (2-bit), significantly reducing the memory footprint. MINT also shares the quantization scaling factor between weights and membrane potentials, eliminating the need for multipliers required in conventional uniform quantization. Experimental results show that our method matches the accuracy of full-precision models and other state-of-the-art SNN quantization techniques while surpassing them in memory footprint reduction and hardware cost efficiency at deployment. For example, 2-bit MINT VGG-16 achieves 90.6% accuracy on CIFAR-10, with roughly 93.8% reduction in memory footprint from the full-precision model and 90% reduction in computation energy compared to vanilla uniform quantization at deployment. The code is available at https://github.com/Intelligent-Computing-Lab-Yale/MINT-Quantization.

cs.NE↗
arXiv2023

GNN-Based Beamforming for Sum-Rate Maximization in MU-MISO Networks

The advantages of graph neural networks (GNNs) in leveraging the graph topology of wireless networks have drawn increasing attentions. This paper studies the GNN-based learning approach for the sum-rate maximization in multiple-user multiple-input single-output (MU-MISO) networks subject to the users' individual data rate requirements and the power budget of the base station. By modeling the MU-MISO network as a graph, a GNN-based architecture named CRGAT is proposed to directly map the channel state information to the beamforming vectors. The attention-enabled aggregation and the residual-assisted combination are adopted to enhance the learning capability and avoid the oversmoothing issue. Furthermore, a novel activation function is proposed for the constraint due to the limited power budget at the base station. The CRGAT is trained in an unsupervised learning manner with two proposed loss functions. An evaluation method is proposed for the learning-based approach, based on which the effectiveness of the proposed CRGAT is validated in comparison with several convex optimization and learning based approaches. Numerical results are provided to reveal the advantages of the CRGAT including the millisecond-level response with limited optimality performance loss, the scalability to different number of users and power budgets, and the adaptability to different system settings.

eess.SY↗
arXiv2023

Outlier Suppression+: Accurate quantization of large language models by equivalent and optimal shifting and scaling

Post-training quantization~(PTQ) of transformer language models faces significant challenges due to the existence of detrimental outliers in activations. We observe that these outliers are concentrated in specific channels and are asymmetric across channels. To address this issue, we propose the Outlier Suppression+~(OS+) framework, which contains the channel-wise shifting for asymmetry and channel-wise scaling for concentration. We show that these operations can be seamlessly migrated into subsequent modules while maintaining equivalence. Second, we propose a fast and stable scheme to calculate effective shifting and scaling values. The channel-wise shifting aligns the center of each channel for removal of outlier asymmetry. The channel-wise scaling quantitatively evaluates changes brought by migration and quantization for better quantization burden balance. We validate our OS+ under both standard and fine-grained quantization settings with models including BERT, OPT, BLOOM, BLOOMZ, and LLaMA. Comprehensive results across various tasks demonstrate the superiority of our approach. Especially, with standard quantization, OS+ can achieve near-floating-point performance on both small models and large language models on 8-bit and 6-bit. Besides, we establish a new state-of-the-art for 4-bit BERT with 15.5\% improvement. Our code is available at \url{https://github.com/ModelTC/Outlier_Suppression_Plus}.

cs.CL↗
arXiv2023

A Primal-Dual-Critic Algorithm for Offline Constrained Reinforcement Learning

Offline constrained reinforcement learning (RL) aims to learn a policy that maximizes the expected cumulative reward subject to constraints on expected cumulative cost using an existing dataset. In this paper, we propose Primal-Dual-Critic Algorithm (PDCA), a novel algorithm for offline constrained RL with general function approximation. PDCA runs a primal-dual algorithm on the Lagrangian function estimated by critics. The primal player employs a no-regret policy optimization oracle to maximize the Lagrangian estimate and the dual player acts greedily to minimize the Lagrangian estimate. We show that PDCA can successfully find a near saddle point of the Lagrangian, which is nearly optimal for the constrained RL problem. Unlike previous work that requires concentrability and a strong Bellman completeness assumption, PDCA only requires concentrability and realizability assumptions for sample-efficient learning.

cs.LG↗
arXiv2023

SEENN: Towards Temporal Spiking Early-Exit Neural Networks

Spiking Neural Networks (SNNs) have recently become more popular as a biologically plausible substitute for traditional Artificial Neural Networks (ANNs). SNNs are cost-efficient and deployment-friendly because they process input in both spatial and temporal manner using binary spikes. However, we observe that the information capacity in SNNs is affected by the number of timesteps, leading to an accuracy-efficiency tradeoff. In this work, we study a fine-grained adjustment of the number of timesteps in SNNs. Specifically, we treat the number of timesteps as a variable conditioned on different input samples to reduce redundant timesteps for certain data. We call our method Spiking Early-Exit Neural Networks (SEENNs). To determine the appropriate number of timesteps, we propose SEENN-I which uses a confidence score thresholding to filter out the uncertain predictions, and SEENN-II which determines the number of timesteps by reinforcement learning. Moreover, we demonstrate that SEENN is compatible with both the directly trained SNN and the ANN-SNN conversion. By dynamically adjusting the number of timesteps, our SEENN achieves a remarkable reduction in the average number of timesteps during inference. For example, our SEENN-II ResNet-19 can achieve 96.1% accuracy with an average of 1.08 timesteps on the CIFAR-10 test dataset. Code is shared at https://github.com/Intelligent-Computing-Lab-Yale/SEENN.

cs.NE↗
arXiv2023

Artificial to Spiking Neural Networks Conversion for Scientific Machine Learning

We introduce a method to convert Physics-Informed Neural Networks (PINNs), commonly used in scientific machine learning, to Spiking Neural Networks (SNNs), which are expected to have higher energy efficiency compared to traditional Artificial Neural Networks (ANNs). We first extend the calibration technique of SNNs to arbitrary activation functions beyond ReLU, making it more versatile, and we prove a theorem that ensures the effectiveness of the calibration. We successfully convert PINNs to SNNs, enabling computational efficiency for diverse regression tasks in solving multiple differential equations, including the unsteady Navier-Stokes equations. We demonstrate great gains in terms of overall efficiency, including Separable PINNs (SPINNs), which accelerate the training process. Overall, this is the first work of this kind and the proposed method achieves relatively good accuracy with low spike rates.

cs.NE↗
arXiv2023

FinVis-GPT: A Multimodal Large Language Model for Financial Chart Analysis

In this paper, we propose FinVis-GPT, a novel multimodal large language model (LLM) specifically designed for financial chart analysis. By leveraging the power of LLMs and incorporating instruction tuning and multimodal capabilities, FinVis-GPT is capable of interpreting financial charts and providing valuable analysis. To train FinVis-GPT, a financial task oriented dataset was generated for pre-training alignment and instruction tuning, comprising various types of financial charts and their corresponding descriptions. We evaluate the model performance via several case studies due to the time limit, and the promising results demonstrated that FinVis-GPT is superior in various financial chart related tasks, including generating descriptions, answering questions and predicting future market trends, surpassing existing state-of-the-art multimodal LLMs. The proposed FinVis-GPT serves as a pioneering effort in utilizing multimodal LLMs in the finance domain and our generated dataset will be release for public use in the near future to speedup related research.

cs.CL↗
arXiv2023

Diffusion Models for Probabilistic Deconvolution of Galaxy Images

Telescopes capture images with a particular point spread function (PSF). Inferring what an image would have looked like with a much sharper PSF, a problem known as PSF deconvolution, is ill-posed because PSF convolution is not an invertible transformation. Deep generative models are appealing for PSF deconvolution because they can infer a posterior distribution over candidate images that, if convolved with the PSF, could have generated the observation. However, classical deep generative models such as VAEs and GANs often provide inadequate sample diversity. As an alternative, we propose a classifier-free conditional diffusion model for PSF deconvolution of galaxy images. We demonstrate that this diffusion model captures a greater diversity of possible deconvolutions compared to a conditional VAE.

astro-ph.IM↗
arXiv2023

SysNoise: Exploring and Benchmarking Training-Deployment System Inconsistency

Extensive studies have shown that deep learning models are vulnerable to adversarial and natural noises, yet little is known about model robustness on noises caused by different system implementations. In this paper, we for the first time introduce SysNoise, a frequently occurred but often overlooked noise in the deep learning training-deployment cycle. In particular, SysNoise happens when the source training system switches to a disparate target system in deployments, where various tiny system mismatch adds up to a non-negligible difference. We first identify and classify SysNoise into three categories based on the inference stage; we then build a holistic benchmark to quantitatively measure the impact of SysNoise on 20+ models, comprehending image classification, object detection, instance segmentation and natural language processing tasks. Our extensive experiments revealed that SysNoise could bring certain impacts on model robustness across different tasks and common mitigations like data augmentation and adversarial training show limited effects on it. Together, our findings open a new research topic and we hope this work will raise research attention to deep learning deployment systems accounting for model performance. We have open-sourced the benchmark and framework at https://modeltc.github.io/systemnoise_web.

cs.LG↗
arXiv2023

Manipulating chiral-spin transport with ferroelectric polarization

A collective excitation of the spin structure in a magnetic insulator can transmit spin-angular momentum with negligible dissipation. This quantum of a spin wave, introduced more than nine decades ago, has always been manipulated through magnetic dipoles, (i.e., timereversal symmetry). Here, we report the experimental observation of chiral-spin transport in multiferroic BiFeO3, where the spin transport is controlled by reversing the ferroelectric polarization (i.e., spatial inversion symmetry). The ferroelectrically controlled magnons produce an unprecedented ratio of up to 18% rectification at room temperature. The spin torque that the magnons in BiFeO3 carry can be used to efficiently switch the magnetization of adja-cent magnets, with a spin-torque efficiency being comparable to the spin Hall effect in heavy metals. Utilizing such a controllable magnon generation and transmission in BiFeO3, an alloxide, energy-scalable logic is demonstrated composed of spin-orbit injection, detection, and magnetoelectric control. This observation opens a new chapter of multiferroic magnons and paves an alternative pathway towards low-dissipation nanoelectronics.

physics.app-ph↗
arXiv2023

Input-Aware Dynamic Timestep Spiking Neural Networks for Efficient In-Memory Computing

Spiking Neural Networks (SNNs) have recently attracted widespread research interest as an efficient alternative to traditional Artificial Neural Networks (ANNs) because of their capability to process sparse and binary spike information and avoid expensive multiplication operations. Although the efficiency of SNNs can be realized on the In-Memory Computing (IMC) architecture, we show that the energy cost and latency of SNNs scale linearly with the number of timesteps used on IMC hardware. Therefore, in order to maximize the efficiency of SNNs, we propose input-aware Dynamic Timestep SNN (DT-SNN), a novel algorithmic solution to dynamically determine the number of timesteps during inference on an input-dependent basis. By calculating the entropy of the accumulated output after each timestep, we can compare it to a predefined threshold and decide if the information processed at the current timestep is sufficient for a confident prediction. We deploy DT-SNN on an IMC architecture and show that it incurs negligible computational overhead. We demonstrate that our method only uses 1.46 average timesteps to achieve the accuracy of a 4-timestep static SNN while reducing the energy-delay-product by 80%.

cs.NE↗
arXiv2023

Sharing Leaky-Integrate-and-Fire Neurons for Memory-Efficient Spiking Neural Networks

Spiking Neural Networks (SNNs) have gained increasing attention as energy-efficient neural networks owing to their binary and asynchronous computation. However, their non-linear activation, that is Leaky-Integrate-and-Fire (LIF) neuron, requires additional memory to store a membrane voltage to capture the temporal dynamics of spikes. Although the required memory cost for LIF neurons significantly increases as the input dimension goes larger, a technique to reduce memory for LIF neurons has not been explored so far. To address this, we propose a simple and effective solution, EfficientLIF-Net, which shares the LIF neurons across different layers and channels. Our EfficientLIF-Net achieves comparable accuracy with the standard SNNs while bringing up to ~4.3X forward memory efficiency and ~21.9X backward memory efficiency for LIF neurons. We conduct experiments on various datasets including CIFAR10, CIFAR100, TinyImageNet, ImageNet-100, and N-Caltech101. Furthermore, we show that our approach also offers advantages on Human Activity Recognition (HAR) datasets, which heavily rely on temporal information.

cs.NE↗
arXiv2023

Broadband nonlinear modulation of incoherent light using a transparent optoelectronic neuron array

Nonlinear optical processing of ambient natural light is highly desired in computational imaging and sensing applications. A strong optical nonlinear response that can work under weak broadband incoherent light is essential for this purpose. Here we introduce an optoelectronic nonlinear filter array that can address this emerging need. By merging 2D transparent phototransistors (TPTs) with liquid crystal (LC) modulators, we create an optoelectronic neuron array that allows self-amplitude modulation of spatially incoherent light, achieving a large nonlinear contrast over a broad spectrum at orders-of-magnitude lower intensity than what is achievable in most optical nonlinear materials. For a proof-of-concept demonstration, we fabricated a 10,000-pixel array of optoelectronic neurons, each serving as a nonlinear filter, and experimentally demonstrated an intelligent imaging system that uses the nonlinear response to instantly reduce input glares while retaining the weaker-intensity objects within the field of view of a cellphone camera. This intelligent glare-reduction capability is important for various imaging applications, including autonomous driving, machine vision, and security cameras. Beyond imaging and sensing, this optoelectronic neuron array, with its rapid nonlinear modulation for processing incoherent broadband light, might also find applications in optical computing, where nonlinear activation functions that can work under ambient light conditions are highly sought.

physics.optics↗
arXiv2023

Uncovering the Representation of Spiking Neural Networks Trained with Surrogate Gradient

Spiking Neural Networks (SNNs) are recognized as the candidate for the next-generation neural networks due to their bio-plausibility and energy efficiency. Recently, researchers have demonstrated that SNNs are able to achieve nearly state-of-the-art performance in image recognition tasks using surrogate gradient training. However, some essential questions exist pertaining to SNNs that are little studied: Do SNNs trained with surrogate gradient learn different representations from traditional Artificial Neural Networks (ANNs)? Does the time dimension in SNNs provide unique representation power? In this paper, we aim to answer these questions by conducting a representation similarity analysis between SNNs and ANNs using Centered Kernel Alignment (CKA). We start by analyzing the spatial dimension of the networks, including both the width and the depth. Furthermore, our analysis of residual connections shows that SNNs learn a periodic pattern, which rectifies the representations in SNNs to be ANN-like. We additionally investigate the effect of the time dimension on SNN representation, finding that deeper layers encourage more dynamics along the time dimension. We also investigate the impact of input data such as event-stream data and adversarial attacks. Our work uncovers a host of new findings of representations in SNNs. We hope this work will inspire future research to fully comprehend the representation power of SNNs. Code is released at https://github.com/Intelligent-Computing-Lab-Yale/SNNCKA.

cs.LG↗
arXiv2023

Universal Polarization Transformations: Spatial programming of polarization scattering matrices using a deep learning-designed diffractive polarization transformer

We demonstrate universal polarization transformers based on an engineered diffractive volume, which can synthesize a large set of arbitrarily-selected, complex-valued polarization scattering matrices between the polarization states at different positions within its input and output field-of-views (FOVs). This framework comprises 2D arrays of linear polarizers with diverse angles, which are positioned between isotropic diffractive layers, each containing tens of thousands of diffractive features with optimizable transmission coefficients. We demonstrate that, after its deep learning-based training, this diffractive polarization transformer could successfully implement N_i x N_o = 10,000 different spatially-encoded polarization scattering matrices with negligible error within a single diffractive volume, where N_i and N_o represent the number of pixels in the input and output FOVs, respectively. We experimentally validated this universal polarization transformation framework in the terahertz part of the spectrum by fabricating wire-grid polarizers and integrating them with 3D-printed diffractive layers to form a physical polarization transformer operating at 0.75 mm wavelength. Through this set-up, we demonstrated an all-optical polarization permutation operation of spatially-varying polarization fields, and simultaneously implemented distinct spatially-encoded polarization scattering matrices between the input and output FOVs of a compact diffractive processor that axially spans 200 wavelengths. This framework opens up new avenues for developing novel optical devices for universal polarization control, and may find various applications in, e.g., remote sensing, medical imaging, security, material inspection and machine vision.

physics.optics↗
arXiv2023

Optical information transfer through random unknown diffusers using electronic encoding and diffractive decoding

Free-space optical information transfer through diffusive media is critical in many applications, such as biomedical devices and optical communication, but remains challenging due to random, unknown perturbations in the optical path. In this work, we demonstrate an optical diffractive decoder with electronic encoding to accurately transfer the optical information of interest, corresponding to, e.g., any arbitrary input object or message, through unknown random phase diffusers along the optical path. This hybrid electronic-optical model, trained using supervised learning, comprises a convolutional neural network (CNN) based electronic encoder and successive passive diffractive layers that are jointly optimized. After their joint training using deep learning, our hybrid model can transfer optical information through unknown phase diffusers, demonstrating generalization to new random diffusers never seen before. The resulting electronic-encoder and the optical-decoder model were experimentally validated using a 3D-printed diffractive network that axially spans less than 70 x lambda, where lambda = 0.75 mm is the illumination wavelength in the terahertz spectrum, carrying the desired optical information through random unknown diffusers. The presented framework can be physically scaled to operate at different parts of the electromagnetic spectrum, without retraining its components, and would offer low-power and compact solutions for optical information transfer in free space through unknown random diffusive media.

physics.optics↗
arXiv2023

Rapid Sensing of Hidden Objects and Defects using a Single-Pixel Diffractive Terahertz Processor

Terahertz waves offer numerous advantages for the nondestructive detection of hidden objects/defects in materials, as they can penetrate through most optically-opaque materials. However, existing terahertz inspection systems are restricted in their throughput and accuracy (especially for detecting small features) due to their limited speed and resolution. Furthermore, machine vision-based continuous sensing systems that use large-pixel-count imaging are generally bottlenecked due to their digital storage, data transmission and image processing requirements. Here, we report a diffractive processor that rapidly detects hidden defects/objects within a target sample using a single-pixel spectroscopic terahertz detector, without scanning the sample or forming/processing its image. This terahertz processor consists of passive diffractive layers that are optimized using deep learning to modify the spectrum of the terahertz radiation according to the absence/presence of hidden structures or defects. After its fabrication, the resulting diffractive processor all-optically probes the structural information of the sample volume and outputs a spectrum that directly indicates the presence or absence of hidden structures, not visible from outside. As a proof-of-concept, we trained a diffractive terahertz processor to sense hidden defects (including subwavelength features) inside test samples, and evaluated its performance by analyzing the detection sensitivity as a function of the size and position of the unknown defects. We validated its feasibility using a single-pixel terahertz time-domain spectroscopy setup and 3D-printed diffractive layers, successfully detecting hidden defects using pulsed terahertz illumination. This technique will be valuable for various applications, e.g., security screening, biomedical sensing, quality control, anti-counterfeiting measures and cultural heritage protection.

physics.optics↗
arXiv2023

QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantization

Recently, post-training quantization (PTQ) has driven much attention to produce efficient neural networks without long-time retraining. Despite its low cost, current PTQ works tend to fail under the extremely low-bit setting. In this study, we pioneeringly confirm that properly incorporating activation quantization into the PTQ reconstruction benefits the final accuracy. To deeply understand the inherent reason, a theoretical framework is established, indicating that the flatness of the optimized low-bit model on calibration and test data is crucial. Based on the conclusion, a simple yet effective approach dubbed as QDROP is proposed, which randomly drops the quantization of activations during PTQ. Extensive experiments on various tasks including computer vision (image classification, object detection) and natural language processing (text classification and question answering) prove its superiority. With QDROP, the limit of PTQ is pushed to the 2-bit activation for the first time and the accuracy boost can be up to 51.49%. Without bells and whistles, QDROP establishes a new state of the art for PTQ. Our code is available at https://github.com/wimh966/QDrop and has been integrated into MQBench (https://github.com/ModelTC/MQBench)

cs.CV↗