arXiv subjects

Xinwei Chen

Publications and source records attributed to Xinwei Chen.

17 recordsLinked to original sources

arXiv2025

ASAP: an Agentic Solution to Auto-optimize Performance of Large-Scale LLM Training

Optimizing large-language model (LLM) training on distributed domain-specific accelerator systems presents significant challenges due to its complex optimization space. Existing optimization methods, however, rely on time-consuming manual tuning or resource-intensive black-box searches, which struggle to keep pace with the rapidly evolving LLM domain, leading to slow development and underutilized resources. To address this, we introduce ASAP, an Agentic Solution to Auto-optimize Performance of Large-Scale LLM Training. It is a multi-agent system, featuring Coordinator, Analyzer, and Proposal agents, which integrates LLM reasoning with insights from performance profiling tools, roofline analysis, and a knowledge base of best practices and successful past optimizations from human experts. Our proposed design can automate the diagnosis of performance bottlenecks and recommend optimized sharding configurations with reasoning, thus effectively improving the efficiency of distributed LLM training. Experiments have shown that the ASAP-generated sharding configurations can contribute up to 28% training step time reduction and 1.43 times throughput improvement. When combined with additional optimization from human experts, throughput can be further increased to 2.58 times. The proposed ASAP promises to provide a scalable and explainable methodology for AI-assisted performance engineering in large-scale LLM training.

cs.MA↗

arXiv2025

Global-State-Free Obstacle Avoidance for Quadrotor Control in Air-Ground Cooperation

CoNi-MPC provides an efficient framework for UAV control in air-ground cooperative tasks by relying exclusively on relative states, eliminating the need for global state estimation. However, its lack of environmental information poses significant challenges for obstacle avoidance. To address this issue, we propose a novel obstacle avoidance algorithm, Cooperative Non-inertial frame-based Obstacle Avoidance (CoNi-OA), designed explicitly for UAV-UGV cooperative scenarios without reliance on global state estimation or obstacle prediction. CoNi-OA uniquely utilizes a single frame of raw LiDAR data from the UAV to generate a modulation matrix, which directly adjusts the quadrotor's velocity to achieve obstacle avoidance. This modulation-based method enables real-time generation of collision-free trajectories within the UGV's non-inertial frame, significantly reducing computational demands (less than 5 ms per iteration) while maintaining safety in dynamic and unpredictable environments. The key contributions of this work include: (1) a modulation-based obstacle avoidance algorithm specifically tailored for UAV-UGV cooperation in non-inertial frames without global states; (2) rapid, real-time trajectory generation based solely on single-frame LiDAR data, removing the need for obstacle modeling or prediction; and (3) adaptability to both static and dynamic environments, thus extending applicability to featureless or unknown scenarios.

cs.RO↗

arXiv2025

To Start Up a Start-Up$-$Embedding Strategic Demand Development in Operational On-Demand Fulfillment via Reinforcement Learning with Information Shaping

The last few years have witnessed rapid growth in the on-demand delivery market, with many start-ups entering the field. However, not all of these start-ups have succeeded due to various reasons, among others, not being able to establish a large enough customer base. In this paper, we address this problem that many on-demand transportation start-ups face: how to establish themselves in a new market. When starting, such companies often have limited fleet resources to serve demand across a city. Depending on the use of the fleet, varying service quality is observed in different areas of the city, and in turn, the service quality impacts the respective growth of demand in each area. Thus, operational fulfillment decisions drive the longer-term demand development. To integrate strategic demand development into real-time fulfillment operations, we propose a two-step approach. First, we derive analytical insights into optimal allocation decisions for a stylized problem. Second, we use these insights to shape the training data of a reinforcement learning strategy for operational real-time fulfillment. Our experiments demonstrate that combining operational efficiency with long-term strategic planning is highly advantageous. Further, we show that the careful shaping of training data is essential for the successful development of demand.

cs.LG↗

arXiv2025

Solving Situation Puzzles with Large Language Model and External Reformulation

In recent years, large language models (LLMs) have shown an impressive ability to perform arithmetic and symbolic reasoning tasks. However, we found that LLMs (e.g., ChatGPT) cannot perform well on reasoning that requires multiple rounds of dialogue, especially when solving situation puzzles. Specifically, LLMs intend to ask very detailed questions focusing on a specific aspect or same/similar questions after several rounds of Q&As. To help LLMs get out of the above dilemma, we propose a novel external reformulation methodology, where the situation puzzle will be reformulated after several rounds of Q&A or when the LLMs raise an incorrect guess. Experiments show superior performance (e.g., win rate, number of question/guess attempts) of our method than directly using LLMs for solving situation puzzles, highlighting the potential of strategic problem reformulation to enhance the reasoning capabilities of LLMs in complex interactive scenarios.

cs.LG↗

arXiv2024

GPTDrawer: Enhancing Visual Synthesis through ChatGPT

In the burgeoning field of AI-driven image generation, the quest for precision and relevance in response to textual prompts remains paramount. This paper introduces GPTDrawer, an innovative pipeline that leverages the generative prowess of GPT-based models to enhance the visual synthesis process. Our methodology employs a novel algorithm that iteratively refines input prompts using keyword extraction, semantic analysis, and image-text congruence evaluation. By integrating ChatGPT for natural language processing and Stable Diffusion for image generation, GPTDrawer produces a batch of images that undergo successive refinement cycles, guided by cosine similarity metrics until a threshold of semantic alignment is attained. The results demonstrate a marked improvement in the fidelity of images generated in accordance with user-defined prompts, showcasing the system's ability to interpret and visualize complex semantic constructs. The implications of this work extend to various applications, from creative arts to design automation, setting a new benchmark for AI-assisted creative processes.

cs.CV↗

arXiv2024

High-Efficiency Urban 3D Radio Map Estimation Based on Sparse Measurements

Recent widespread applications for unmanned aerial vehicles (UAVs) -- from infrastructure inspection to urban logistics -- have prompted an urgent need for high-accuracy three-dimensional (3D) radio maps. However, existing methods designed for two-dimensional radio maps face challenges of high measurement costs and limited data availability when extended to 3D scenarios. To tackle these challenges, we first build a real-world large-scale 3D radio map dataset, covering over 4.2 million m^3 and over 4 thousand data points in complex urban environments. We propose a Gaussian Process Regression-based scheme for 3D radio map estimation, allowing us to realize more accurate map recovery with a lower RMSE than state-of-the-art schemes by over 2.5 dB. To further enhance data efficiency, we propose two methods for training point selection, including an offline clustering-based method and an online maximum a posterior (MAP)-based method. Extensive experiments demonstrate that the proposed scheme not only achieves full-map recovery with only 2% of UAV measurements, but also sheds light on future studies on 3D radio maps.

cs.IT↗

arXiv2024

Energy-Guided Diffusion Sampling for Offline-to-Online Reinforcement Learning

Combining offline and online reinforcement learning (RL) techniques is indeed crucial for achieving efficient and safe learning where data acquisition is expensive. Existing methods replay offline data directly in the online phase, resulting in a significant challenge of data distribution shift and subsequently causing inefficiency in online fine-tuning. To address this issue, we introduce an innovative approach, \textbf{E}nergy-guided \textbf{DI}ffusion \textbf{S}ampling (EDIS), which utilizes a diffusion model to extract prior knowledge from the offline dataset and employs energy functions to distill this knowledge for enhanced data generation in the online phase. The theoretical analysis demonstrates that EDIS exhibits reduced suboptimality compared to solely utilizing online data or directly reusing offline data. EDIS is a plug-in approach and can be combined with existing methods in offline-to-online RL setting. By implementing EDIS to off-the-shelf methods Cal-QL and IQL, we observe a notable 20% average improvement in empirical performance on MuJoCo, AntMaze, and Adroit environments. Code is available at \url{https://github.com/liuxhym/EDIS}.

cs.LG↗

arXiv2024

Region-level labels in ice charts can produce pixel-level segmentation for Sea Ice types

Fully supervised deep learning approaches have demonstrated impressive accuracy in sea ice classification, but their dependence on high-resolution labels presents a significant challenge due to the difficulty of obtaining such data. In response, our weakly supervised learning method provides a compelling alternative by utilizing lower-resolution regional labels from expert-annotated ice charts. This approach achieves exceptional pixel-level classification performance by introducing regional loss representations during training to measure the disparity between predicted and ice chart-derived sea ice type distributions. Leveraging the AI4Arctic Sea Ice Challenge Dataset, our method outperforms the fully supervised U-Net benchmark, the top solution of the AutoIce challenge, in both mapping resolution and class-wise accuracy, marking a significant advancement in automated operational sea ice mapping.

cs.CV↗

arXiv2024

Mix of Experts Language Model for Named Entity Recognition

Named Entity Recognition (NER) is an essential steppingstone in the field of natural language processing. Although promising performance has been achieved by various distantly supervised models, we argue that distant supervision inevitably introduces incomplete and noisy annotations, which may mislead the model training process. To address this issue, we propose a robust NER model named BOND-MoE based on Mixture of Experts (MoE). Instead of relying on a single model for NER prediction, multiple models are trained and ensembled under the Expectation-Maximization (EM) framework, so that noisy supervision can be dramatically alleviated. In addition, we introduce a fair assignment module to balance the document-model assignment process. Extensive experiments on real-world datasets show that the proposed method achieves state-of-the-art performance compared with other distantly supervised NER.

cs.CL↗

arXiv2024

Few-shot Name Entity Recognition on StackOverflow

StackOverflow, with its vast question repository and limited labeled examples, raise an annotation challenge for us. We address this gap by proposing RoBERTa+MAML, a few-shot named entity recognition (NER) method leveraging meta-learning. Our approach, evaluated on the StackOverflow NER corpus (27 entity types), achieves a 5% F1 score improvement over the baseline. We improved the results further domain-specific phrase processing enhance results.

cs.CL↗

arXiv2023

CoNi-MPC: Cooperative Non-inertial Frame Based Model Predictive Control

This paper presents a novel solution for UAV control in cooperative multi-robot systems, which can be used in various scenarios such as leader-following, landing on a moving base, or specific relative motion with a target. Unlike classical methods that tackle UAV control in the world frame, we directly control the UAV in the target coordinate frame, without making motion assumptions about the target. In detail, we formulate a non-linear model predictive controller of a UAV, referred to as the agent, within a non-inertial frame (i.e., the target frame). The system requires the relative states (pose and velocity), the angular velocity and the accelerations of the target, which can be obtained by relative localization methods and ubiquitous MEMS IMU sensors, respectively. This framework eliminates dependencies that are vital in classical solutions, such as accurate state estimation for both the agent and target, prior knowledge of the target motion model, and continuous trajectory re-planning for some complex tasks. We have performed extensive simulations to investigate the control performance with varying motion characteristics of the target. Furthermore, we conducted real robot experiments, employing either simulated relative pose estimation from motion capture systems indoors or directly from our previous relative pose estimation devices outdoors, to validate the applicability and feasibility of the proposed approach.

cs.RO↗

arXiv2021

Same-Day Delivery with Fairness

The demand for same-day delivery (SDD) has increased rapidly in the last few years and has particularly boomed during the COVID-19 pandemic. The fast growth is not without its challenge. In 2016, due to low concentrations of memberships and far distance from the depot, certain minority neighborhoods were excluded from receiving Amazon's SDD service, raising concerns about fairness. In this paper, we study the problem of offering fair SDD-service to customers. The service area is partitioned into different regions. Over the course of a day, customers request for SDD service, and the timing of requests and delivery locations are not known in advance. The dispatcher dynamically assigns vehicles to make deliveries to accepted customers before their delivery deadline. In addition to the overall service rate (utility), we maximize the minimal regional service rate across all regions (fairness). We model the problem as a multi-objective Markov decision process and develop a deep Q-learning solution approach. We introduce a novel transformation of learning from rates to actual services, which creates a stable and efficient learning process. Computational results demonstrate the effectiveness of our approach in alleviating unfairness both spatially and temporally in different customer geographies. We also show this effectiveness is valid with different depot locations, providing businesses with an opportunity to achieve better fairness from any location. Further, we consider the impact of ignoring fairness in service, and results show that our policies eventually outperform the utility-driven baseline when customers have a high expectation on service level.

cs.LG↗

arXiv2021

Measuring Sample Efficiency and Generalization in Reinforcement Learning Benchmarks: NeurIPS 2020 Procgen Benchmark

The NeurIPS 2020 Procgen Competition was designed as a centralized benchmark with clearly defined tasks for measuring Sample Efficiency and Generalization in Reinforcement Learning. Generalization remains one of the most fundamental challenges in deep reinforcement learning, and yet we do not have enough benchmarks to measure the progress of the community on Generalization in Reinforcement Learning. We present the design of a centralized benchmark for Reinforcement Learning which can help measure Sample Efficiency and Generalization in Reinforcement Learning by doing end to end evaluation of the training and rollout phases of thousands of user submitted code bases in a scalable way. We designed the benchmark on top of the already existing Procgen Benchmark by defining clear tasks and standardizing the end to end evaluation setups. The design aims to maximize the flexibility available for researchers who wish to design future iterations of such benchmarks, and yet imposes necessary practical constraints to allow for a system like this to scale. This paper presents the competition setup and the details and analysis of the top solutions identified through this setup in context of 2020 iteration of the competition at NeurIPS.

cs.LG↗

arXiv2021

Deep Q-Learning for Same-Day Delivery with Vehicles and Drones

In this paper, we consider same-day delivery with vehicles and drones. Customers make delivery requests over the course of the day, and the dispatcher dynamically dispatches vehicles and drones to deliver the goods to customers before their delivery deadline. Vehicles can deliver multiple packages in one route but travel relatively slowly due to the urban traffic. Drones travel faster, but they have limited capacity and require charging or battery swaps. To exploit the different strengths of the fleets, we propose a deep Q-learning approach. Our method learns the value of assigning a new customer to either drones or vehicles as well as the option to not offer service at all. In a systematic computational analysis, we show the superiority of our policy compared to benchmark policies and the effectiveness of our deep Q-learning approach. We also show that our policy can maintain effectiveness when the fleet size changes moderately. Experiments on data drawn from varied spatial/temporal distributions demonstrate that our trained policies can cope with changes in the input data.

cs.LG↗

arXiv2020

Deep Learning for Content-based Personalized Viewport Prediction of 360-Degree VR Videos

In this paper, the problem of head movement prediction for virtual reality videos is studied. In the considered model, a deep learning network is introduced to leverage position data as well as video frame content to predict future head movement. For optimizing data input into this neural network, data sample rate, reduced data, and long-period prediction length are also explored for this model. Simulation results show that the proposed approach yields 16.1\% improvement in terms of prediction accuracy compared to a baseline approach that relies only on the position data.

cs.CV↗

arXiv2016

IPA in the Loop: Control Design for Throughput Regulation in Computer Processors

A new technique for performance regulation in event-driven systems, recently proposed by the authors, consists of an adaptive-gain integral control. The gain is adjusted in the control loop by a real-time estimation of the derivative of the plant-function with respect to the control input. This estimation is carried out by Infinitesimal Perturbation Analysis (IPA). The main motivation comes from applications to throughput regulation in computer processors, where to-date, testing and assessment of the proposed control technique has been assessed by simulation. The purpose of this paper is to report on its implementation on a machine, namely an Intel Haswell microprocessor, and compare its performance to that obtained from cycle-level, full system simulation environment. The intrinsic contribution of the paper to the Workshop on Discrete Event System is in describing the process of taking an IPA-based design and simulation to a concrete implementation, thereby providing a bridge between theory and applications.

math.OC↗

arXiv2016

Performance Regulation of Event-Driven Dynamical Systems Using Infinitesimal Perturbation Analysis

This paper presents a performance-regulation method for a class of stochastic timed event-driven systems aimed at output tracking of a given reference setpoint. The systems are either Discrete Event Dynamic Systems (DEDS) such as queueing networks or Petri nets, or Hybrid Systems (HS) with time-driven dynamics and event-driven dynamics, like fluid queues and hybrid Petri nets. The regulator, designed for simplicity and speed of computation, is comprised of a single integrator having a variable gain to ensure effective tracking under time-varying plants. The gain's computation is based on the Infinitesimal Perturbation Analysis (IPA) gradient of the plant function with respect to the control variable, and the resultant tracking can be quite robust with respect to modeling inaccuracies and gradient-estimation errors. The proposed technique is tested on examples taken from various application areas and modeled with different formalisms, including queueing models, Petri-net model of a production-inventory control system, and a stochastic DEDS model of a multicore chip control. Simulation results are presented in support of the proposed approach.

math.OC↗