arXiv subjects

Yifei Zhang

Publications and source records attributed to Yifei Zhang.

At least 163 records · Page 9Linked to original sources

arXiv2024

Neuroscheduling for Remote Estimation

Many modern distributed systems consist of devices that generate more data than what can be transmitted via a communication link in near real time with high-fidelity. We consider the scheduling problem in which a device has access to multiple data sources, but at any moment, only one of them is revealed in real-time to a remote receiver. Even when the sources are Gaussian, and the fidelity criterion is the mean squared error, the globally optimal data selection strategy is not known. We propose a data-driven methodology to search for the elusive optimal solution using linear function approximation approach called neuroscheduling and establish necessary and sufficient conditions for the optimal scheduler to not over fit training data. Additionally, we present several numerical results that show that the globally optimal scheduler and estimator pair to the Gaussian case are nonlinear.

eess.SY↗

arXiv2024

On Joint Marginal Expected Shortfall and Associated Contribution Risk Measures

Systemic risk is the risk that a company- or industry-level risk could trigger a huge collapse of another or even the whole institution. Various systemic risk measures have been proposed in the literature to quantify the domino and (relative) spillover effects induced by systemic risks such as the well-known CoVaR, CoES, MES and CoD risk measures, and associated contribution measures. This paper proposes another new type of systemic risk measure, called the joint marginal expected shortfall (JMES), to measure whether the MES of one entity's risk-taking adds to another one or the overall risk conditioned on the event that the entity is already in some specified distress level. We further introduce two useful systemic risk contribution measures based on the difference function or relative ratio function of the JMES and the conventional ES, respectively. Some basic properties of these proposed measures are studied such as monotonicity, comonotonic additivity, non-identifiability and non-elicitability. For both risk measures and two different vectors of bivariate risks, we establish sufficient conditions imposed on copula structure, stress levels, and stochastic orders to compare these new measures. We further provide some numerical examples to illustrate our main findings. A real application in analyzing the risk contagion among several stock market indices is implemented to show the performances of our proposed measures compared with other commonly used measures including CoVaR, CoES, MES, and their associated contribution measures.

q-fin.RM↗

arXiv2024

Evaluate What You Can't Evaluate: Unassessable Quality for Generated Response

LLMs (large language models) such as ChatGPT have shown remarkable language understanding and generation capabilities. Although reference-free evaluators based on LLMs show better human alignment than traditional reference-based evaluators, there are many challenges in using reference-free evaluators based on LLMs. Reference-free evaluators are more suitable for open-ended examples with different semantics responses. But not all examples are open-ended. For closed-ended examples with unique correct semantic response, reference-free evaluators will still consider it high quality when giving a response that is inconsistent with the facts and the semantic of reference. In order to comprehensively evaluate the reliability of evaluators based on LLMs, we construct two adversarial meta-evaluation dialogue generation datasets KdConv-ADV and DSTC7-ADV based on KdConv and DSTC7-AVSD, respectively. Compared to previous meta-evaluation benchmarks, KdConv-ADV and DSTC7-ADV are much more challenging since they requires evaluators to be able to reasonably evaluate closed-ended examples with the help of external knowledge or even its own knowledge. Empirical results show that the ability of LLMs to identify unreasonable responses is insufficient. There are risks in using eference-free evaluators based on LLMs to evaluate the quality of dialogue responses.

cs.CL↗

arXiv2024

Spectrally Pruned Gaussian Fields with Neural Compensation

Recently, 3D Gaussian Splatting, as a novel 3D representation, has garnered attention for its fast rendering speed and high rendering quality. However, this comes with high memory consumption, e.g., a well-trained Gaussian field may utilize three million Gaussian primitives and over 700 MB of memory. We credit this high memory footprint to the lack of consideration for the relationship between primitives. In this paper, we propose a memory-efficient Gaussian field named SUNDAE with spectral pruning and neural compensation. On one hand, we construct a graph on the set of Gaussian primitives to model their relationship and design a spectral down-sampling module to prune out primitives while preserving desired signals. On the other hand, to compensate for the quality loss of pruning Gaussians, we exploit a lightweight neural network head to mix splatted features, which effectively compensates for quality losses while capturing the relationship between primitives in its weights. We demonstrate the performance of SUNDAE with extensive results. For example, SUNDAE can achieve 26.80 PSNR at 145 FPS using 104 MB memory while the vanilla Gaussian splatting algorithm achieves 25.60 PSNR at 160 FPS using 523 MB memory, on the Mip-NeRF360 dataset. Codes are publicly available at https://runyiyang.github.io/projects/SUNDAE/.

cs.CV↗

arXiv2024

Is Mamba Effective for Time Series Forecasting?

In the realm of time series forecasting (TSF), it is imperative for models to adeptly discern and distill hidden patterns within historical time series data to forecast future states. Transformer-based models exhibit formidable efficacy in TSF, primarily attributed to their advantage in apprehending these patterns. However, the quadratic complexity of the Transformer leads to low computational efficiency and high costs, which somewhat hinders the deployment of the TSF model in real-world scenarios. Recently, Mamba, a selective state space model, has gained traction due to its ability to process dependencies in sequences while maintaining near-linear complexity. For TSF tasks, these characteristics enable Mamba to comprehend hidden patterns as the Transformer and reduce computational overhead compared to the Transformer. Therefore, we propose a Mamba-based model named Simple-Mamba (S-Mamba) for TSF. Specifically, we tokenize the time points of each variate autonomously via a linear layer. A bidirectional Mamba layer is utilized to extract inter-variate correlations and a Feed-Forward Network is set to learn temporal dependencies. Finally, the generation of forecast outcomes through a linear mapping layer. Experiments on thirteen public datasets prove that S-Mamba maintains low computational overhead and achieves leading performance. Furthermore, we conduct extensive experiments to explore Mamba's potential in TSF tasks. Our code is available at https://github.com/wzhwzhwzh0921/S-D-Mamba.

cs.LG↗

arXiv2024

Visual Attention Prompted Prediction and Learning

Visual explanation (attention)-guided learning uses not only labels but also explanations to guide model reasoning process. While visual attention-guided learning has shown promising results, it requires a large number of explanation annotations that are time-consuming to prepare. However, in many real-world situations, it is usually desired to prompt the model with visual attention without model retraining. For example, when doing AI-assisted cancer classification on a medical image, users (e.g., clinicians) can provide the AI model with visual attention prompt on which areas are indispensable and which are precluded. Despite its promising objectives, achieving visual attention-prompted prediction presents several major challenges: 1) How can the visual prompt be effectively integrated into the model's reasoning process? 2) How should the model handle samples that lack visual prompts? 3) What is the impact on the model's performance when a visual prompt is imperfect? This paper introduces a novel framework for attention-prompted prediction and learning, utilizing visual prompts to steer the model's reasoning process. To improve performance in non-prompted situations and align it with prompted scenarios, we propose a co-training approach for both non-prompted and prompted models, ensuring they share similar parameters and activations. Additionally, for instances where the visual prompt does not encompass the entire input image, we have developed innovative attention prompt refinement methods. These methods interpolate the incomplete prompts while maintaining alignment with the model's explanations. Extensive experiments on four datasets demonstrate the effectiveness of our proposed framework in enhancing predictions for samples both with and without prompt.

cs.CV↗

arXiv2024

Contribution of coherent electron production to measurements of heavy-flavor decayed electrons in heavy-ion collisions

Heavy quarks, produced at early stages of heavy-ion collisions, are an excellent probe of the Quark-Gluon Plasma (QGP) also created in these collisions. Electrons from open heavy-flavor hadron decays (HFE) are good proxies for heavy quarks, and have been measured extensively in the last two decades to study QGP properties. These measurements are traditionally carried out by subtracting all known background sources from the inclusive electron sample. More recently, a significant enhancement of $e^+e^-$ pair production at very low transverse momenta was observed in peripheral heavy-ion collisions. The production characteristics is consistent with coherent photon-photon interactions, which should also constitute a background source to the HFE measurements. In this article, we provide theoretical predictions for the contribution of coherent electron production to HFE as a function of transverse momentum, centrality and collision energy in Au+Au and Pb+Pb collisions.

hep-ph↗

arXiv2024

DreamLIP: Language-Image Pre-training with Long Captions

Language-image pre-training largely relies on how precisely and thoroughly a text describes its paired image. In practice, however, the contents of an image can be so rich that well describing them requires lengthy captions (e.g., with 10 sentences), which are usually missing in existing datasets. Consequently, there are currently no clear evidences on whether and how language-image pre-training could benefit from long captions. To figure this out, we first re-caption 30M images with detailed descriptions using a pre-trained Multi-modality Large Language Model (MLLM), and then study the usage of the resulting captions under a contrastive learning framework. We observe that, each sentence within a long caption is very likely to describe the image partially (e.g., an object). Motivated by this, we propose to dynamically sample sub-captions from the text label to construct multiple positive pairs, and introduce a grouping loss to match the embeddings of each sub-caption with its corresponding local image patches in a self-supervised manner. Experimental results on a wide rage of downstream tasks demonstrate the consistent superiority of our method, termed DreamLIP, over previous alternatives, highlighting its fine-grained representational capacity. It is noteworthy that, on the tasks of image-text retrieval and semantic segmentation, our model trained with 30M image-text pairs achieves on par or even better performance than CLIP trained with 400M pairs. Project page is available at https://zyf0619sjtu.github.io/dream-lip.

cs.CV↗

arXiv2024

On the role of network structure in learning to coordinate with bounded rationality

Many socioeconomic phenomena, such as technology adoption, collaborative problem-solving, and content engagement, involve a collection of agents coordinating to take a common action, aligning their decisions to maximize their individual goals. We consider a model for networked interactions where agents learn to coordinate their binary actions under a strict bound on their rationality. We first prove that our model is a potential game and that the optimal action profile is always to achieve perfect alignment at one of the two possible actions, regardless of the network structure. Using a stochastic learning algorithm known as Log Linear Learning, where agents have the same finite rationality parameter, we show that the probability of agents successfully agreeing on the correct decision is monotonically increasing in the number of network links. Therefore, more connectivity improves the accuracy of collective decision-making, as predicted by the phenomenon known as Wisdom of Crowds. Finally, we show that for a fixed number of links, a regular network maximizes the probability of success. We conclude that when using a network of irrational agents, promoting more homogeneous connectivity improves the accuracy of collective decision-making.

physics.soc-ph↗

arXiv2024

DUE: Dynamic Uncertainty-Aware Explanation Supervision via 3D Imputation

Explanation supervision aims to enhance deep learning models by integrating additional signals to guide the generation of model explanations, showcasing notable improvements in both the predictability and explainability of the model. However, the application of explanation supervision to higher-dimensional data, such as 3D medical images, remains an under-explored domain. Challenges associated with supervising visual explanations in the presence of an additional dimension include: 1) spatial correlation changed, 2) lack of direct 3D annotations, and 3) uncertainty varies across different parts of the explanation. To address these challenges, we propose a Dynamic Uncertainty-aware Explanation supervision (DUE) framework for 3D explanation supervision that ensures uncertainty-aware explanation guidance when dealing with sparsely annotated 3D data with diffusion-based 3D interpolation. Our proposed framework is validated through comprehensive experiments on diverse real-world medical imaging datasets. The results demonstrate the effectiveness of our framework in enhancing the predictability and explainability of deep learning models in the context of medical imaging diagnosis applications.

cs.CV↗

arXiv2024

FastMAC: Stochastic Spectral Sampling of Correspondence Graph

3D correspondence, i.e., a pair of 3D points, is a fundamental concept in computer vision. A set of 3D correspondences, when equipped with compatibility edges, forms a correspondence graph. This graph is a critical component in several state-of-the-art 3D point cloud registration approaches, e.g., the one based on maximal cliques (MAC). However, its properties have not been well understood. So we present the first study that introduces graph signal processing into the domain of correspondence graph. We exploit the generalized degree signal on correspondence graph and pursue sampling strategies that preserve high-frequency components of this signal. To address time-consuming singular value decomposition in deterministic sampling, we resort to a stochastic approximate sampling strategy. As such, the core of our method is the stochastic spectral sampling of correspondence graph. As an application, we build a complete 3D registration algorithm termed as FastMAC, that reaches real-time speed while leading to little to none performance drop. Through extensive experiments, we validate that FastMAC works for both indoor and outdoor benchmarks. For example, FastMAC can accelerate MAC by 80 times while maintaining high registration success rate on KITTI. Codes are publicly available at https://github.com/Forrest-110/FastMAC.

cs.CV↗

arXiv2024

oneDNN Graph Compiler: A Hybrid Approach for High-Performance Deep Learning Compilation

With the rapid development of deep learning models and hardware support for dense computing, the deep learning workload characteristics changed significantly from a few hot spots on compute-intensive operations to a broad range of operations scattered across the models. Accelerating a few compute-intensive operations using the expert-tuned implementation of primitives does not fully exploit the performance potential of AI hardware. Various efforts have been made to compile a full deep neural network (DNN) graph. One of the biggest challenges is to achieve high-performance tensor compilation by generating expert level performance code for the dense compute-intensive operations and applying compilation optimization at the scope of DNN computation graph across multiple compute-intensive operations. We present oneDNN Graph Compiler, a tensor compiler that employs a hybrid approach of using techniques from both compiler optimization and expert-tuned kernels for high performance code generation of the deep neural network graph. oneDNN Graph Compiler addresses unique optimization challenges in the deep learning domain, such as low-precision computation, aggressive fusion of graph operations, optimization for static tensor shapes and memory layout, constant weight optimization, and memory buffer reuse. Experimental results demonstrate significant performance gains over existing tensor compiler and primitives library for performance-critical DNN computation graphs and end-to-end models on Intel Xeon Scalable Processors.

cs.LG↗

arXiv2024

ELAD: Explanation-Guided Large Language Models Active Distillation

The deployment and application of Large Language Models (LLMs) is hindered by their memory inefficiency, computational demands, and the high costs of API inferences. Traditional distillation methods, which transfer the capabilities of LLMs to smaller models, often fail to determine whether the knowledge has been sufficiently transferred, potentially resulting in high costs or incomplete distillation. In this paper, we propose an Explanation-Guided LLMs Active Distillation (ELAD) framework that employs an active learning strategy to optimize the balance between annotation costs and model performance. To improve efficient sample selection, we introduce an explanation-guided sample selection method that identifies samples challenging its reasoning by exploiting uncertainties in explanation steps. Additionally, we present a customized LLM-annotated explanation revision technique where the teacher model detects and corrects flaws in the student model's reasoning. Our experiments across various reasoning datasets demonstrate that our framework significantly enhances the efficiency of LLM knowledge distillation.

cs.CL↗

arXiv2024

STICKERCONV: Generating Multimodal Empathetic Responses from Scratch

Stickers, while widely recognized for enhancing empathetic communication in online interactions, remain underexplored in current empathetic dialogue research, notably due to the challenge of a lack of comprehensive datasets. In this paper, we introduce the Agent for STICKERCONV (Agent4SC), which uses collaborative agent interactions to realistically simulate human behavior with sticker usage, thereby enhancing multimodal empathetic communication. Building on this foundation, we develop a multimodal empathetic dialogue dataset, STICKERCONV, comprising 12.9K dialogue sessions, 5.8K unique stickers, and 2K diverse conversational scenarios. This dataset serves as a benchmark for multimodal empathetic generation. To advance further, we propose PErceive and Generate Stickers (PEGS), a multimodal empathetic response generation framework, complemented by a comprehensive set of empathy evaluation metrics based on LLM. Our experiments demonstrate PEGS's effectiveness in generating contextually relevant and emotionally resonant multimodal empathetic responses, contributing to the advancement of more nuanced and engaging empathetic dialogue systems.

cs.CL↗

arXiv2024

AI Empowered Channel Semantic Acquisition for 6G Integrated Sensing and Communication Networks

Motivated by the need for increased spectral efficiency and the proliferation of intelligent applications, the sixth-generation (6G) mobile network is anticipated to integrate the dual-functions of communication and sensing (C&S). Although the millimeter wave (mmWave) communication and mmWave radar share similar multiple-input multiple-output (MIMO) architecture for integration, the full potential of dual-function synergy remains to be exploited. In this paper, we commence by overviewing state-of-the-art schemes from the aspects of waveform design and signal processing. Nevertheless, these approaches face the dilemma of mutual compromise between C&S performance. To this end, we reveal and exploit the synergy between C&S. In the proposed framework, we introduce a two-stage frame structure and resort artificial intelligence (AI) to achieve the synergistic gain by designing a joint C&S channel semantic extraction and reconstruction network (JCASCasterNet). With just a cost-effective and energy-efficient single sensing antenna, the proposed scheme achieves enhanced overall performance while requiring only limited pilot and feedback signaling overhead. In the end, we outline the challenges that lie ahead in the future development of integrated sensing and communication networks, along with promising directions for further research.

cs.IT↗

arXiv2024

InRank: Incremental Low-Rank Learning

The theory of greedy low-rank learning (GLRL) aims to explain the impressive generalization capabilities of deep learning. It proves that stochastic gradient-based training implicitly regularizes neural networks towards low-rank solutions through a gradual increase of the rank during training. However, there is a gap between theory and practice since GLRL requires an infinitesimal initialization of the weights, which is not practical due to the fact that it is a saddle point. In this work, we remove the assumption of infinitesimal initialization by focusing on cumulative weight updates. We prove the cumulative weight updates follow an incremental low-rank trajectory for arbitrary orthogonal initialization of weights in a three-layer linear network. Empirically, we demonstrate that our theory holds on a broad range of neural networks (e.g., transformers) and standard training algorithms (e.g., SGD, Adam). However, existing training algorithms do not exploit the low-rank property to improve computational efficiency as the networks are not parameterized in low-rank. To remedy this, we design a new training algorithm Incremental Low-Rank Learning (InRank), which explicitly expresses cumulative weight updates as low-rank matrices while incrementally augmenting their ranks during training. We evaluate InRank on GPT-2, and our results indicate that InRank achieves comparable prediction performance as the full-rank counterpart while requiring at most 33% of the total ranks throughout training. We also propose an efficient version of InRank that achieves a reduction of 37% in total training time and 36% in model size when training GPT-medium on WikiText-103 from scratch.

cs.LG↗

arXiv2023

Electrically reconfigurable phase-change transmissive metasurface

Programmable and reconfigurable optics hold significant potential for transforming a broad spectrum of applications, spanning space explorations to biomedical imaging, gas sensing, and optical cloaking. The ability to adjust the optical properties of components like filters, lenses, and beam steering devices could result in dramatic reductions in size, weight, and power consumption in future optoelectronic devices. Among the potential candidates for reconfigurable optics, chalcogenide-based phase change materials (PCMs) offer great promise due to their non-volatile and analogue switching characteristics. Although PCM have found widespread use in electronic data storage, these memory devices are deeply sub-micron-sized. To incorporate phase change materials into free-space optical components, it is essential to scale them up to beyond several hundreds of microns while maintaining reliable switching characteristics. This study demonstrated a non-mechanical, non-volatile transmissive filter based on low-loss PCMs with a 200 $μ$m$ \times $200 $μ$m switching area. The device/metafilter can be consistently switched between low- and high-transmission states using electrical pulses with a switching contrast ratio of 5.5 dB. The device was reversibly switched for 1250 cycles before accelerated degradation took place. The work represents an important step toward realizing free-space reconfigurable optics based on PCMs.

physics.optics↗

arXiv2023

Insight-HXMT on-orbit thermal control status and thermal deformation impact analysis

Purpose: The Hard X-ray Modulation Telescope is China's first X-ray astronomy satellite launched on June 15th, 2017, dubbed Insight-HXMT. Active and passive thermal control measures are employed to keep devices at suitable temperatures. In this paper, we analyzed the on-orbit thermal monitoring data of the first 5 years and investigated the effect of thermal deformation on the point spread function (PSF) of the telescopes. Methods: We examined the data of the on-orbit temperatures measured using 157 thermistors placed on the collimators, detectors and their support structures and compared the results with the thermal control requirements. The thermal deformation was evaluated by the relative orientation of the two star sensors installed on the main support structure. its effect was estimated with evolution of the PSF obtained with calibration scanning observations of the Crab nebula. Conclusion: The on-orbit temperatures met the thermal control requirements thus far, and the effect of thermal deformation on the PSF was negligible after the on-orbit pointing calibration.

astro-ph.IM↗