arXiv subjects

Yu Feng

Publications and source records attributed to Yu Feng.

At least 127 records · Page 7Linked to original sources

arXiv2024

Existence and non-uniqueness of cone spherical metrics with prescribed singularities on a compact Riemann surface with positive genus

Cone spherical metrics, defined on compact Riemann surfaces, are conformal metrics with constant curvature one and finitely many cone singularities. Such a metric is termed \textit{reducible} if a developing map of the metric has monodromy in ${\rm U(1)}$, and \textit{irreducible} otherwise. Utilizing the polystable extensions of two line bundles on a compact Riemann surface $X$ with genus $g_X>0$, we establish the following three primary results concerning these metrics with cone angles in $2π{\mathbb Z}_{>1}$: \begin{itemize} \item[(1)] Given an effective divisor $D$ with an odd degree surpassing $2g_X$ on $X$, we find the existence of an effective divisor $D'$ in the complete linear system $|D|$ that can be represented by at least two distinct irreducible cone spherical metrics on $X$. \item[(2)] For a generic effective divisor $D$ with an even degree and $°D\geq 6g_X-2$ on $X$, we can identify an arcwise connected Borel subset in $|D|$ that demonstrates a Hausdorff dimension of no less than $\big(°D-4g_{X}+2\big)$. Within this subset, each divisor $D'$ can be distinctly represented by a family of reducible metrics, defined by a single real parameter. \item[(3)] For an effective divisor $D$ with $°D=2$ on an elliptic curve, we can identify a Borel subset in $|D|$ that is arcwise connected, showcasing a Hausdorff dimension of one. Within this subset, each divisor $D'$ can be distinctly represented by a family of reducible metrics, defined by a single real parameter.

math.DG↗

arXiv2024

Potamoi: Accelerating Neural Rendering via a Unified Streaming Architecture

Neural Radiance Field (NeRF) has emerged as a promising alternative for photorealistic rendering. Despite recent algorithmic advancements, achieving real-time performance on today's resource-constrained devices remains challenging. In this paper, we identify the primary bottlenecks in current NeRF algorithms and introduce a unified algorithm-architecture co-design, Potamoi, designed to accommodate various NeRF algorithms. Specifically, we introduce a runtime system featuring a plug-and-play algorithm, SpaRW, which significantly reduces the per-frame computational workload and alleviates compute inefficiencies. Furthermore, our unified streaming pipeline coupled with customized hardware support effectively tames both SRAM and DRAM inefficiencies by minimizing repetitive DRAM access and completely eliminating SRAM bank conflicts. When evaluated against a baseline utilizing a dedicated DNN accelerator, our framework demonstrates a speed-up and energy reduction of 53.1$\times$ and 67.7$\times$, respectively, all while maintaining high visual quality with less than a 1.0 dB reduction in peak signal-to-noise ratio.

cs.AR↗

arXiv2024

CP-Prompt: Composition-Based Cross-modal Prompting for Domain-Incremental Continual Learning

The key challenge of cross-modal domain-incremental learning (DIL) is to enable the learning model to continuously learn from novel data with different feature distributions under the same task without forgetting old ones. However, existing top-performing methods still cause high forgetting rates, by lacking intra-domain knowledge extraction and inter-domain common prompting strategy. In this paper, we propose a simple yet effective framework, CP-Prompt, by training limited parameters to instruct a pre-trained model to learn new domains and avoid forgetting existing feature distributions. CP-Prompt captures intra-domain knowledge by compositionally inserting personalized prompts on multi-head self-attention layers and then learns the inter-domain knowledge with a common prompting strategy. CP-Prompt shows superiority compared with state-of-the-art baselines among three widely evaluated DIL tasks. The source code is available at https://github.com/dannis97500/CP_Prompt.

cs.CL↗

arXiv2024

vTensor: Flexible Virtual Tensor Management for Efficient LLM Serving

Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests. This surge in demand poses significant challenges in optimizing throughput and latency while keeping costs manageable. The Key-Value (KV) cache, a standard method for retaining previous computations, makes LLM inference highly bounded by memory. While batching strategies can enhance performance, they frequently lead to significant memory fragmentation. Even though cutting-edge systems like vLLM mitigate KV cache fragmentation using paged Attention mechanisms, they still suffer from inefficient memory and computational operations due to the tightly coupled page management and computation kernels. This study introduces the vTensor, an innovative tensor structure for LLM inference based on GPU virtual memory management (VMM). vTensor addresses existing limitations by decoupling computation from memory defragmentation and offering dynamic extensibility. Our framework employs a CPU-GPU heterogeneous approach, ensuring efficient, fragmentation-free memory management while accommodating various computation kernels across different LLM architectures. Experimental results indicate that vTensor achieves an average speedup of 1.86x across different models, with up to 2.42x in multi-turn chat scenarios. Additionally, vTensor provides average speedups of 2.12x and 3.15x in kernel evaluation, reaching up to 3.92x and 3.27x compared to SGLang Triton prefix-prefilling kernels and vLLM paged Attention kernel, respectively. Furthermore, it frees approximately 71.25% (57GB) of memory on the NVIDIA A100 GPU compared to vLLM, enabling more memory-intensive workloads.

cs.DC↗

arXiv2024

AutoVCoder: A Systematic Framework for Automated Verilog Code Generation using LLMs

Recently, the use of large language models (LLMs) for software code generation, e.g., C/C++ and Python, has proven a great success. However, LLMs still suffer from low syntactic and functional correctness when it comes to the generation of register-transfer level (RTL) code, such as Verilog. To address this issue, in this paper, we develop AutoVCoder, a systematic open-source framework that significantly improves the LLMs' correctness of generating Verilog code and enhances the quality of its output at the same time. Our framework integrates three novel techniques, including a high-quality hardware dataset generation approach, a two-round LLM fine-tuning method and a domain-specific retrieval-augmented generation (RAG) mechanism. Experimental results demonstrate that AutoVCoder outperforms both industrial and academic LLMs in Verilog code generation. Specifically, AutoVCoder shows a 0.5% and 2.2% improvement in functional correctness on the EvalMachine and EvalHuman benchmarks compared with BetterV, and also achieves a 3.4% increase in syntax correctness and a 3.4% increase in functional correctness on the RTLLM benchmark compared with RTLCoder.

cs.AR↗

arXiv2024

Logarithmic correlation functions in 2D critical percolation

It is believed that the large-scale geometric properties of two-dimensional critical percolation are described by a logarithmic conformal field theory, but it has been challenging to exhibit concrete examples of logarithmic singularities and to find an explanation and a physical interpretation, in terms of lattice observables, for their appearance. We show that certain percolation correlation functions receive independent contributions from a large number of similar connectivity events happening at different scales. Combined with scale invariance, this leads to logarithmic divergences. We study several logarithmic correlation functions for critical percolation in the bulk and in the presence of a boundary, including the four-point function of the density (spin) field. Our analysis confirms previous findings, provides new explicit calculations and explains, in terms of lattice observables, the physical mechanism that leads to the logarithmic singularities we discover. Although we adopt conformal field theory (CFT) terminology to present our results, the core of our analysis relies on probabilistic arguments and recent rigorous results on the scaling limit of critical percolation and does not assume a priori the existence of a percolation CFT. As a consequence, our results provide strong support for the validity of a CFT description of critical percolation and a step in the direction of a mathematically rigorous formulation of a logarithmic CFT of two-dimensional critical percolation.

math-ph↗

arXiv2024

Gap Completion in Point Cloud Scene occluded by Vehicles using SGC-Net

Recent advances in mobile mapping systems have greatly enhanced the efficiency and convenience of acquiring urban 3D data. These systems utilize LiDAR sensors mounted on vehicles to capture vast cityscapes. However, a significant challenge arises due to occlusions caused by roadside parked vehicles, leading to the loss of scene information, particularly on the roads, sidewalks, curbs, and the lower sections of buildings. In this study, we present a novel approach that leverages deep neural networks to learn a model capable of filling gaps in urban scenes that are obscured by vehicle occlusion. We have developed an innovative technique where we place virtual vehicle models along road boundaries in the gap-free scene and utilize a ray-casting algorithm to create a new scene with occluded gaps. This allows us to generate diverse and realistic urban point cloud scenes with and without vehicle occlusion, surpassing the limitations of real-world training data collection and annotation. Furthermore, we introduce the Scene Gap Completion Network (SGC-Net), an end-to-end model that can generate well-defined shape boundaries and smooth surfaces within occluded gaps. The experiment results reveal that 97.66% of the filled points fall within a range of 5 centimeters relative to the high-density ground truth point cloud scene. These findings underscore the efficacy of our proposed model in gap completion and reconstructing urban scenes affected by vehicle occlusions.

cs.CV↗

arXiv2024

BLINK: Multimodal Large Language Models Can See but Not Perceive

We introduce Blink, a new benchmark for multimodal language models (LLMs) that focuses on core visual perception abilities not found in other evaluations. Most of the Blink tasks can be solved by humans "within a blink" (e.g., relative depth estimation, visual correspondence, forensics detection, and multi-view reasoning). However, we find these perception-demanding tasks cast significant challenges for current multimodal LLMs because they resist mediation through natural language. Blink reformats 14 classic computer vision tasks into 3,807 multiple-choice questions, paired with single or multiple images and visual prompting. While humans get 95.70% accuracy on average, Blink is surprisingly challenging for existing multimodal LLMs: even the best-performing GPT-4V and Gemini achieve accuracies of 51.26% and 45.72%, only 13.17% and 7.63% higher than random guessing, indicating that such perception abilities have not "emerged" yet in recent multimodal LLMs. Our analysis also highlights that specialist CV models could solve these problems much better, suggesting potential pathways for future improvements. We believe Blink will stimulate the community to help multimodal LLMs catch up with human-level visual perception.

cs.CV↗

arXiv2024

Connection Probabilities of Multiple FK-Ising Interfaces

We find the scaling limits of a general class of boundary-to-boundary connection probabilities and multiple interfaces in the critical planar FK-Ising model, thus verifying predictions from the physics literature. We also discuss conjectural formulas using Coulomb gas integrals for the corresponding quantities in general critical planar random-cluster models with cluster-weight $q \in [1,4)$. Thus far, proofs for convergence, including ours, rely on discrete complex analysis techniques and are beyond reach for other values of $q$ than the FK-Ising model ($q=2$). Given the convergence of interfaces, the conjectural formulas for other values of $q$ could be verified similarly with relatively minor technical work. The limit interfaces are variants of $\mathrm{SLE}_κ$ curves (with $κ= 16/3$ for $q=2$). Their partition functions, that give the connection probabilities, also satisfy properties predicted for correlation functions in conformal field theory (CFT), expected to describe scaling limits of critical random-cluster models. We verify these properties for all $q \in [1,4)$, thus providing further evidence of the expected CFT description of these models.

math.PR↗

arXiv2024

PNG-UNITsims: Halo clustering response to primordial non-Gaussianities as a function of mass

We present the largest full N-body simulation to date with local primordial non-Gaussianities (L-PNG), the \textsc{PNG-UNITsim}. It tracks the evolution of $4096^3$ particles within a periodic box with $L_{\rm box} = 1 \; h^{-1}\,{\rm Gpc}$, leading to a mass resolution of $m_{p} = 1.24\times 10^{9}\; h^{-1}\,M_\odot$. This is enough to resolve galaxies targeted by stage-IV spectroscopic surveys. The \textsc{PNG-UNIT} has \textit{Fixed} initial conditions whose phases are also \textit{Matched} to the pre-existing \textsc{UNIT} simulation. These two features in the simulations reduce our uncertainty significantly so we use 100 \textsc{FastPM} mocks to estimate this reduction. The amplitude of the non-Gaussianities used to set the initial conditions of this new simulation is $f_{\rm NL}^{\rm local} = 100$. In this first study, we use mass selected dark matter haloes from the \textsc{PNG-UNIT} simulation to constrain the local PNG parameters. PNG induce a scale dependent bias, parameterised through \bp or $p$, which might depend on the type of cosmological tracer. Those cases when $p=1$ are referred to as the {\it universality relation}. We measure $p$ as a function of the halo mass. Haloes with masses between $1\times 10^{12}$ and $2\times 10^{13} \, h^{-1} M_\odot$ are well described by the {\it universality relation}. For haloes with masses between $2\times 10^{10}$ and $1\times 10^{12} \, h^{-1} M_\odot$ we find that $p<1$ at $3σ$. Combining all the mass bins, we find $p$ consistent with a value of $0.955\pm0.013$, which is $3σ$ away from \textit{universality}, as low mass haloes are more numerous. We also study the effect of using priors on $p$ when constraining $f_{\rm NL}$. Using the values we obtain for $b_ϕ$ as priors, we forecast that a DESI-like (stage-IV) survey will be able to constrain $f_{\rm NL}$ better than if the universality relation is assumed.

astro-ph.CO↗

arXiv2024

Estimating Earthquake Early Warning Effectiveness via Blind Zone Sizes: A Case Study of the Planned Seismic Network in Chinese Mainland

The China Earthquake Administration (CEA) has launched an ambitious nationwide earthquake early warning (EEW) system project currently under development, which will include approximately 15,000 seismic stations and be the largest EEW system in the world. The new EEW system is planned to go online at the end of 2023. In approximately 50%, 30% and 20% of Chinese mainland, the inter-station distance will soon be smaller than 50 km, 25 km and 15 km, respectively. The expected effectiveness of this EEW system can be quantified via the metric determined from the radius of the blind zone, which refers to the area near the epicenter where there is insufficient time to issue a warning before the arrival of strong S- and surface waves. This study uses a theoretical network-based method together with Monte Carlo simulation to obtain the spatial distribution of the blind zone radii and their associated uncertainties for the new seismic network based on its configuration. We find that the densified new seismic network is expected to have excellent EEW performance as the area covered by small blind zones with radius less than 30 km increases dramatically from approximately 2% to 22%, that is by 2.4 million km2 inside Chinese mainland. We also find that every 1,000,000 RMB (about 146,000 USD) invested to densify the planned network will lead to an areal increase of 3,000 km2 of small blind zones. Continuing to increase the density of stations in some key regions with blind zone radii ranging from 15 to 40 km is still necessary to control the unexpected expansion of blind zones due to possible (and common) stations failure. Our work provides insights into the expected performance of the upcoming EEW network in Chinese mainland, and our proposed evaluation approach is broadly applicable for predicting the performance of EEW systems during their planning, design, and implementation stages.

physics.geo-ph↗

arXiv2024

BlissCam: Boosting Eye Tracking Efficiency with Learned In-Sensor Sparse Sampling

Eye tracking is becoming an increasingly important task domain in emerging computing platforms such as Augmented/Virtual Reality (AR/VR). Today's eye tracking system suffers from long end-to-end tracking latency and can easily eat up half of the power budget of a mobile VR device. Most existing optimization efforts exclusively focus on the computation pipeline by optimizing the algorithm and/or designing dedicated accelerators while largely ignoring the front-end of any eye tracking pipeline: the image sensor. This paper makes a case for co-designing the imaging system with the computing system. In particular, we propose the notion of "in-sensor sparse sampling", whereby the pixels are drastically downsampled (by 20x) within the sensor. Such in-sensor sampling enhances the overall tracking efficiency by significantly reducing 1) the power consumption of the sensor readout chain and sensor-host communication interfaces, two major power contributors, and 2) the work done on the host, which receives and operates on far fewer pixels. With careful reuse of existing pixel circuitry, our proposed BLISSCAM requires little hardware augmentation to support the in-sensor operations. Our synthesis results show up to 8.2x energy reduction and 1.4x latency reduction over existing eye tracking pipelines.

cs.AR↗

arXiv2024

Test-Time Training on Graphs with Large Language Models (LLMs)

Graph Neural Networks have demonstrated great success in various fields of multimedia. However, the distribution shift between the training and test data challenges the effectiveness of GNNs. To mitigate this challenge, Test-Time Training (TTT) has been proposed as a promising approach. Traditional TTT methods require a demanding unsupervised training strategy to capture the information from test to benefit the main task. Inspired by the great annotation ability of Large Language Models (LLMs) on Text-Attributed Graphs (TAGs), we propose to enhance the test-time training on graphs with LLMs as annotators. In this paper, we design a novel Test-Time Training pipeline, LLMTTT, which conducts the test-time adaptation under the annotations by LLMs on a carefully-selected node set. Specifically, LLMTTT introduces a hybrid active node selection strategy that considers not only node diversity and representativeness, but also prediction signals from the pre-trained model. Given annotations from LLMs, a two-stage training strategy is designed to tailor the test-time model with the limited and noisy labels. A theoretical analysis ensures the validity of our method and extensive experiments demonstrate that the proposed LLMTTT can achieve a significant performance improvement compared to existing Out-of-Distribution (OOD) generalization methods.

cs.LG↗

arXiv2024

Cicero: Addressing Algorithmic and Architectural Bottlenecks in Neural Rendering by Radiance Warping and Memory Optimizations

Neural Radiance Field (NeRF) is widely seen as an alternative to traditional physically-based rendering. However, NeRF has not yet seen its adoption in resource-limited mobile systems such as Virtual and Augmented Reality (VR/AR), because it is simply extremely slow. On a mobile Volta GPU, even the state-of-the-art NeRF models generally execute only at 0.8 FPS. We show that the main performance bottlenecks are both algorithmic and architectural. We introduce, CICERO, to tame both forms of inefficiencies. We first introduce two algorithms, one fundamentally reduces the amount of work any NeRF model has to execute, and the other eliminates irregular DRAM accesses. We then describe an on-chip data layout strategy that eliminates SRAM bank conflicts. A pure software implementation of CICERO offers an 8.0x speed-up and 7.9x energy saving over a mobile Volta GPU. When compared to a baseline with a dedicated DNN accelerator, our speed-up and energy reduction increase to 28.2x and 37.8x, respectively - all with minimal quality loss (less than 1.0 dB peak signal-to-noise ratio reduction).

cs.AR↗

arXiv2024

Multiple Ising interfaces in annulus and $2N$-sided radial SLE

We consider critical planar Ising model in annulus with alternating boundary conditions on the outer boundary and free boundary conditions in the inner boundary. As the size of the inner hole goes to zero, the event that all interfaces get close to the inner hole before they meet each other is a rare event. We prove that the law of the collection of the interfaces conditional on this rare event converges in total variation distance to the so-called $2N$-sided radial SLE$_3$, introduced by~[HL21]. The proof relies crucially on an estimate for multiple chordal SLE. Suppose $(γ_1, \ldots, γ_N)$ is chordal $N$-SLE$_κ$ with $κ\in (0,4]$ in the unit disc, and we consider the probability that all $N$ curves get close to the origin. We prove that the limit $\lim_{r\to 0+}r^{-A_{2N}}\mathbb{P}[\mathrm{dist}(0,γ_j)<r, 1\le j\le N]$ exists, where $A_{2N}$ is the so-called $2N$-arm exponents and $\mathrm{dist}$ is Euclidean distance. We call the limit Green's function for chordal $N$-SLE$_κ$. This estimate is a generalization of previous conclusions with $N=1$ and $N=2$ proved in~[LR12, LR15] and~[Zha20] respectively.

math.PR↗

arXiv2024

DiffPoint: Single and Multi-view Point Cloud Reconstruction with ViT Based Diffusion Model

As the task of 2D-to-3D reconstruction has gained significant attention in various real-world scenarios, it becomes crucial to be able to generate high-quality point clouds. Despite the recent success of deep learning models in generating point clouds, there are still challenges in producing high-fidelity results due to the disparities between images and point clouds. While vision transformers (ViT) and diffusion models have shown promise in various vision tasks, their benefits for reconstructing point clouds from images have not been demonstrated yet. In this paper, we first propose a neat and powerful architecture called DiffPoint that combines ViT and diffusion models for the task of point cloud reconstruction. At each diffusion step, we divide the noisy point clouds into irregular patches. Then, using a standard ViT backbone that treats all inputs as tokens (including time information, image embeddings, and noisy patches), we train our model to predict target points based on input images. We evaluate DiffPoint on both single-view and multi-view reconstruction tasks and achieve state-of-the-art results. Additionally, we introduce a unified and flexible feature fusion module for aggregating image features from single or multiple input images. Furthermore, our work demonstrates the feasibility of applying unified architectures across languages and images to improve 3D reconstruction tasks.

cs.CV↗

arXiv2024

A unified Bayesian inversion approach for a class of tumor growth models with different pressure laws

In this paper, we use the Bayesian inversion approach to study the data assimilation problem for a family of tumor growth models described by porous-medium type equations. The models contain uncertain parameters and are indexed by a physical parameter $m$, which characterizes the constitutive relation between density and pressure. Based on these models, we employ the Bayesian inversion framework to infer parametric and nonparametric unknowns that affect tumor growth from noisy observations of tumor cell density. We establish the well-posedness and the stability theories for the Bayesian inversion problem and further prove the convergence of the posterior distribution in the so-called incompressible limit, $m \rightarrow \infty$. Since the posterior distribution across the index regime $m\in[2,\infty)$ can thus be treated in a unified manner, such theoretical results also guide the design of the numerical inference for the unknown. We propose a generic computational framework for such inverse problems, which consists of a typical sampling algorithm and an asymptotic preserving solver for the forward problem. With extensive numerical tests, we demonstrate that the proposed method achieves satisfactory accuracy in the Bayesian inference of the tumor growth models, which is uniform with respect to the constitutive relation.

math.NA↗

arXiv2024

Differentiable Cosmological Simulation with Adjoint Method

Rapid advances in deep learning have brought not only myriad powerful neural networks, but also breakthroughs that benefit established scientific research. In particular, automatic differentiation (AD) tools and computational accelerators like GPUs have facilitated forward modeling of the Universe with differentiable simulations. Based on analytic or automatic backpropagation, current differentiable cosmological simulations are limited by memory, and thus are subject to a trade-off between time and space/mass resolution, usually sacrificing both. We present a new approach free of such constraints, using the adjoint method and reverse time integration. It enables larger and more accurate forward modeling at the field level, and will improve gradient based optimization and inference. We implement it in an open-source particle-mesh (PM) $N$-body library pmwd (particle-mesh with derivatives). Based on the powerful AD system JAX, pmwd is fully differentiable, and is highly performant on GPUs.

astro-ph.IM↗