arXiv subjects

Zhiyuan He

Publications and source records attributed to Zhiyuan He.

35 records · Page 2Linked to original sources

arXiv2025

Congestion Control System Optimization with Large Language Models

Congestion control is a fundamental component of Internet infrastructure, and researchers have dedicated considerable effort to developing improved congestion control algorithms. However, despite extensive study, existing algorithms continue to exhibit suboptimal performance across diverse network environments. In this paper, we introduce a novel approach that automatically optimizes congestion control algorithms using large language models (LLMs). Our framework consists of a structured algorithm generation process, an emulation-based evaluation pipeline covering a broad range of network conditions, and a statistically guided method to substantially reduce evaluation time. Empirical results from four distinct LLMs validate the effectiveness of our approach. We successfully identify algorithms that achieve up to 27% performance improvements over the original BBR algorithm in a production QUIC implementation. Our work demonstrates the potential of LLMs to accelerate the design of high-performance network algorithms and paves the way for broader applications in networking systems.

cs.NI↗

arXiv2025

Agent Lightning: Train ANY AI Agents with Reinforcement Learning

We present Agent Lightning, a flexible and extensible framework that enables Reinforcement Learning (RL)-based training of Large Language Models (LLMs) for any AI agent. Unlike existing methods that tightly couple RL training with agent or rely on sequence concatenation with masking, Agent Lightning achieves complete decoupling between agent execution and training, allowing seamless integration with existing agents developed via diverse ways (e.g., using frameworks like LangChain, OpenAI Agents SDK, AutoGen, and building from scratch) with almost ZERO code modifications. By formulating agent execution as Markov decision process, we define an unified data interface and propose a hierarchical RL algorithm, LightningRL, which contains a credit assignment module, allowing us to decompose trajectories generated by ANY agents into training transition. This enables RL to handle complex interaction logic, such as multi-agent scenarios and dynamic workflows. For the system design, we introduce a Training-Agent Disaggregation architecture, and brings agent observability frameworks into agent runtime, providing a standardized agent finetuning interface. Experiments across text-to-SQL, retrieval-augmented generation, and math tool-use tasks demonstrate stable, continuous improvements, showcasing the framework's potential for real-world agent training and deployment.

cs.AI↗

arXiv2025

LeanK: Learnable K Cache Channel Pruning for Efficient Decoding

Large language models (LLMs) enable long-context tasks but face efficiency challenges due to the growing key-value (KV) cache. We propose LeanK, a learning-based method that prunes unimportant key (K) cache channels by leveraging static channel sparsity. With a novel two-stage training process, LeanK learns channel-wise static mask that could satisfy specific sparsity ratio and hardware alignment requirement. LeanK reduces GPU memory and accelerates decoding without sacrificing accuracy. Experiments demonstrate up to 70% K cache and 16%-18% V cache memory reduction. Custom decoding kernel enables 1.3x speedup for attention computation. We also provide insights into model channels and attention heads during long-context inference by analyzing the learned importance distribution. Our code is available at https://aka.ms/LeanK.

cs.LG↗

arXiv2025

Do Not Let Low-Probability Tokens Over-Dominate in RL for LLMs

Reinforcement learning (RL) has become a cornerstone for enhancing the reasoning capabilities of large language models (LLMs), with recent innovations such as Group Relative Policy Optimization (GRPO) demonstrating exceptional effectiveness. In this study, we identify a critical yet underexplored issue in RL training: low-probability tokens disproportionately influence model updates due to their large gradient magnitudes. This dominance hinders the effective learning of high-probability tokens, whose gradients are essential for LLMs' performance but are substantially suppressed. To mitigate this interference, we propose two novel methods: Advantage Reweighting and Low-Probability Token Isolation (Lopti), both of which effectively attenuate gradients from low-probability tokens while emphasizing parameter updates driven by high-probability tokens. Our approaches promote balanced updates across tokens with varying probabilities, thereby enhancing the efficiency of RL training. Experimental results demonstrate that they substantially improve the performance of GRPO-trained LLMs, achieving up to a 46.2% improvement in K&K Logic Puzzle reasoning tasks. Our implementation is available at https://github.com/zhyang2226/AR-Lopti.

cs.CL↗

arXiv2025

Orlando's flask: detection of a lost-and-found valley on the Moon

High angular resolution holds the key to extending our knowledge in several domains of astronomical research. In addition to the development of new instruments, advancements in post-processing algorithms can enhance the performances attainable in an observation, turning archival observations into a treasure. We developed a machine-learning tool, named zoom-in, that is able to improve the angular resolution of an astronomical image by a factor of $\sim 100$ by optimally recombining short-cadence sequences of images. After training our model on real-life photographs, we tested our method on archival images of the Moon taken through ESO instruments. We were able to achieve a remarkable spatial resolution of $\sim 1$ m of the lunar surface. While analyzing one of the fields from the sample, we discovered structures of clear anthropic origin inside the Aristarchus crater. The features appear to be consistent with ancient ruins of cities and castles. A thorough analysis of the relevant literature allowed us to conclude that this valley corresponds to the one described in Ludovico Ariosto's "Orlando Furioso": a place where all the items lost by humans gather and pile up. Analyses of the surface brightness from our images, indicating an abnormally high albedo of $\sim 0.25$, further corroborate this idea suggesting a conspicuous presence of glass. We infer the presence of >1 billion flasks of human wits on the lunar surface, whose origin we investigate in detail. We urge for a dedicated mission, astolfo, to be carried out by Artemis astronauts in order to recover human wits and bring them back to the Earth.

astro-ph.EP↗

arXiv2024

Designing Network Algorithms via Large Language Models

We introduce NADA, the first framework to autonomously design network algorithms by leveraging the generative capabilities of large language models (LLMs). Starting with an existing algorithm implementation, NADA enables LLMs to create a wide variety of alternative designs in the form of code blocks. It then efficiently identifies the top-performing designs through a series of filtering techniques, minimizing the need for full-scale evaluations and significantly reducing computational costs. Using adaptive bitrate (ABR) streaming as a case study, we demonstrate that NADA produces novel ABR algorithms -- previously unknown to human developers -- that consistently outperform the original algorithm in diverse network environments, including broadband, satellite, 4G, and 5G.

cs.NI↗

arXiv2024

Position Engineering: Boosting Large Language Models through Positional Information Manipulation

The performance of large language models (LLMs) is significantly influenced by the quality of the prompts provided. In response, researchers have developed enormous prompt engineering strategies aimed at modifying the prompt text to enhance task performance. In this paper, we introduce a novel technique termed position engineering, which offers a more efficient way to guide large language models. Unlike prompt engineering, which requires substantial effort to modify the text provided to LLMs, position engineering merely involves altering the positional information in the prompt without modifying the text itself. We have evaluated position engineering in two widely-used LLM scenarios: retrieval-augmented generation (RAG) and in-context learning (ICL). Our findings show that position engineering substantially improves upon the baseline in both cases. Position engineering thus represents a promising new strategy for exploiting the capabilities of large language models.

cs.CL↗

arXiv2024

Retrieval Augmented Generation (RAG) and Beyond: A Comprehensive Survey on How to Make your LLMs use External Data More Wisely

Large language models (LLMs) augmented with external data have demonstrated remarkable capabilities in completing real-world tasks. Techniques for integrating external data into LLMs, such as Retrieval-Augmented Generation (RAG) and fine-tuning, are gaining increasing attention and widespread application. Nonetheless, the effective deployment of data-augmented LLMs across various specialized fields presents substantial challenges. These challenges encompass a wide range of issues, from retrieving relevant data and accurately interpreting user intent to fully harnessing the reasoning capabilities of LLMs for complex tasks. We believe that there is no one-size-fits-all solution for data-augmented LLM applications. In practice, underperformance often arises from a failure to correctly identify the core focus of a task or because the task inherently requires a blend of multiple capabilities that must be disentangled for better resolution. In this survey, we propose a RAG task categorization method, classifying user queries into four levels based on the type of external data required and primary focus of the task: explicit fact queries, implicit fact queries, interpretable rationale queries, and hidden rationale queries. We define these levels of queries, provide relevant datasets, and summarize the key challenges and most effective techniques for addressing these challenges. Finally, we discuss three main forms of integrating external data into LLMs: context, small model, and fine-tuning, highlighting their respective strengths, limitations, and the types of problems they are suited to solve. This work aims to help readers thoroughly understand and decompose the data requirements and key bottlenecks in building LLM applications, offering solutions to the different challenges and serving as a guide to systematically developing such applications.

cs.CL↗

arXiv2024

RIGID: A Training-free and Model-Agnostic Framework for Robust AI-Generated Image Detection

The rapid advances in generative AI models have empowered the creation of highly realistic images with arbitrary content, raising concerns about potential misuse and harm, such as Deepfakes. Current research focuses on training detectors using large datasets of generated images. However, these training-based solutions are often computationally expensive and show limited generalization to unseen generated images. In this paper, we propose a training-free method to distinguish between real and AI-generated images. We first observe that real images are more robust to tiny noise perturbations than AI-generated images in the representation space of vision foundation models. Based on this observation, we propose RIGID, a training-free and model-agnostic method for robust AI-generated image detection. RIGID is a simple yet effective approach that identifies whether an image is AI-generated by comparing the representation similarity between the original and the noise-perturbed counterpart. Our evaluation on a diverse set of AI-generated images and benchmarks shows that RIGID significantly outperforms existing trainingbased and training-free detectors. In particular, the average performance of RIGID exceeds the current best training-free method by more than 25%. Importantly, RIGID exhibits strong generalization across different image generation methods and robustness to image corruptions.

cs.CV↗

arXiv2024

Transfer Learning-Enhanced Instantaneous Multi-Person Indoor Localization by CSI

Passive indoor localization, integral to smart buildings, emergency response, and indoor navigation, has traditionally been limited by a focus on single-target localization and reliance on multi-packet CSI. We introduce a novel Multi-target loss, notably enhancing multi-person localization. Utilizing this loss function, our instantaneous CSI-ResNet achieves an impressive 99.21% accuracy at 0.6m precision with single-timestamp CSI. A preprocessing algorithm is implemented to counteract WiFi-induced variability, thereby augmenting robustness. Furthermore, we incorporate Nuclear Norm-Based Transfer Pre-Training, ensuring adaptability in diverse environments, which provides a new paradigm for indoor multi-person localization. Additionally, we have developed an extensive dataset, surpassing existing ones in scope and diversity, to underscore the efficacy of our method and facilitate future fingerprint-based localization research.

eess.SP↗

arXiv2023

T3P: Demystifying Low-Earth Orbit Satellite Broadband

The Internet is going through a massive infrastructural revolution with the advent of low-flying satellite networks, 5/6G, WiFi7, and hollow-core fiber deployments. While these networks could unleash enhanced connectivity and new capabilities, it is critical to understand the performance characteristics to efficiently drive applications over them. Low-Earth orbit (LEO) satellite mega-constellations like SpaceX Starlink aim to offer broad coverage and low latencies at the expense of high orbital dynamics leading to continuous latency changes and frequent satellite hand-offs. This paper aims to quantify Starlink's latency and its variations and components using a real testbed spanning multiple latitudes from the North to the South of Europe. We identify tail latencies as a problem. We develop predictors for latency and throughput and show their utility in improving application performance by up to 25%. We also explore how transport protocols can be optimized for LEO networks and show that this can improve throughput by up to 115% (with only a 5% increase in latency). Also, our measurement testbed with a footprint across multiple locations offers unique trigger-based scheduling capabilities that are necessary to quantify the impact of LEO dynamics.

cs.NI↗

arXiv2023

TSNet-SAC: Leveraging Transformers for Efficient Task Scheduling

In future 6G Mobile Edge Computing (MEC), autopilot systems require the capability of processing multimodal data with strong interdependencies. However, traditional heuristic algorithms are inadequate for real-time scheduling due to their requirement for multiple iterations to derive the optimal scheme. We propose a novel TSNet-SAC based on Transformer, that utilizes heuristic algorithms solely to guide the training of TSNet. Additionally, a Sliding Augment Component (SAC) is introduced to enhance the robustness and resolve algorithm defects. Furthermore, the Extender component is designed to handle multi-scale training data and provide network scalability, enabling TSNet to adapt to different access scenarios. Simulation demonstrates that TSNet-SAC outperforms existing networks in accuracy and robustness, achieving superior scheduling-making latency compared to heuristic algorithms.

cs.NI↗

arXiv2023

A Novel Channel-Constrained Model for 6G Vehicular Networks with Traffic Spikes

Mobile Edge Computing (MEC) holds excellent potential in Congestion Management (CM) of 6G vehicular networks. A reasonable schedule of MEC ensures a more reliable and efficient CM system. Unfortunately, existing parallel and sequential models cannot cope with scarce computing resources and constrained channels, especially during traffic rush hour. In this paper, we propose a channel-constrained multi-core sequential model (CCMSM) for task offloading and resource allocation. The CCMSM incorporates a utility index that couples system energy consumption and delay, applying Genetic Algorithm combining Sparrow Search Algorithm (GA-SSA) in the branching optimization. Furthermore, we prove that the system delay is the shortest with the FCFS computing strategy in the MEC server. Simulation demonstrates that the proposed CCMSM achieves a higher optimization level and exhibits better robustness and resilient scalability for traffic spikes.

cs.NI↗

arXiv2022

Be Your Own Neighborhood: Detecting Adversarial Example by the Neighborhood Relations Built on Self-Supervised Learning

Deep Neural Networks (DNNs) have achieved excellent performance in various fields. However, DNNs' vulnerability to Adversarial Examples (AE) hinders their deployments to safety-critical applications. This paper presents a novel AE detection framework, named BEYOND, for trustworthy predictions. BEYOND performs the detection by distinguishing the AE's abnormal relation with its augmented versions, i.e. neighbors, from two prospects: representation similarity and label consistency. An off-the-shelf Self-Supervised Learning (SSL) model is used to extract the representation and predict the label for its highly informative representation capacity compared to supervised learning models. For clean samples, their representations and predictions are closely consistent with their neighbors, whereas those of AEs differ greatly. Furthermore, we explain this observation and show that by leveraging this discrepancy BEYOND can effectively detect AEs. We develop a rigorous justification for the effectiveness of BEYOND. Furthermore, as a plug-and-play model, BEYOND can easily cooperate with the Adversarial Trained Classifier (ATC), achieving the state-of-the-art (SOTA) robustness accuracy. Experimental results show that BEYOND outperforms baselines by a large margin, especially under adaptive attacks. Empowered by the robust relation net built on SSL, we found that BEYOND outperforms baselines in terms of both detection ability and speed. Our code will be publicly available.

cs.LG↗

arXiv2019

Gradient Boosting Machine: A Survey

In this survey, we discuss several different types of gradient boosting algorithms and illustrate their mathematical frameworks in detail: 1. introduction of gradient boosting leads to 2. objective function optimization, 3. loss function estimations, and 4. model constructions. 5. application of boosting in ranking.

stat.ML↗

arXiv2018

Unsupervised Discovery of Object Landmarks as Structural Representations

Deep neural networks can model images with rich latent representations, but they cannot naturally conceptualize structures of object categories in a human-perceptible way. This paper addresses the problem of learning object structures in an image modeling process without supervision. We propose an autoencoding formulation to discover landmarks as explicit structural representations. The encoding module outputs landmark coordinates, whose validity is ensured by constraints that reflect the necessary properties for landmarks. The decoding module takes the landmarks as a part of the learnable input representations in an end-to-end differentiable framework. Our discovered landmarks are semantically meaningful and more predictive of manually annotated landmarks than those discovered by previous methods. The coordinates of our landmarks are also complementary features to pretrained deep-neural-network representations in recognizing visual attributes. In addition, the proposed method naturally creates an unsupervised, perceptible interface to manipulate object shapes and decode images with controllable structures. The project webpage is at http://ytzhang.net/projects/lmdis-rep

cs.CV↗

arXiv2017

Discriminative Bimodal Networks for Visual Localization and Detection with Natural Language Queries

Associating image regions with text queries has been recently explored as a new way to bridge visual and linguistic representations. A few pioneering approaches have been proposed based on recurrent neural language models trained generatively (e.g., generating captions), but achieving somewhat limited localization accuracy. To better address natural-language-based visual entity localization, we propose a discriminative approach. We formulate a discriminative bimodal neural network (DBNet), which can be trained by a classifier with extensive use of negative samples. Our training objective encourages better localization on single images, incorporates text phrases in a broad range, and properly pairs image regions with text phrases into positive and negative examples. Experiments on the Visual Genome dataset demonstrate the proposed DBNet significantly outperforms previous state-of-the-art methods both for localization on single images and for detection on multiple images. We we also establish an evaluation protocol for natural-language visual detection.

cs.CV↗