arXiv subjects

Bowen Xie

Publications and source records attributed to Bowen Xie.

11 recordsLinked to original sources

arXiv2026

Direct Translation between Sign Languages

The field of sign language translation has witnessed significant progress in the translation between sign and spoken languages, but the translation between sign languages remains largely unexplored and out of reach. The latter can help 1.5 billion deaf and hard-of-hearing (DHH) people worldwide communicate across language barriers without relying on hearing interpreters or written-language fluency. The cascade approach composing separate sign-to-text, text-to-text, and text-to-sign systems suffers from error propagation and extra latency as well as the loss of information unique in the visual modality. We aim to develop direct sign-to-sign translation. However, a large-scale open-domain parallel corpus has not been curated between sign languages. To enable direct translation between sign language utterances, we use back-translation to produce synthetic sign-sign pairs from unaligned individual language utterance-sign corpora. Using this data, we jointly train a single MBART-based model for both text->sign (T2S) and sign->sign (S2S). On synthetically generated paired sets between American Sign Language (ASL), Chinese Sign Language (CSL), and German Sign Language (DGS), our direct S2S method outperforms the cascaded baseline on geometric sign error metrics (20% lower DTW-aligned MPJPE) and language matching metrics after predicted sign utterances are translated back to sentences (50% high BLEU-4) while achieving a roughly 2.3* speedup. On a small set of pre-existing cross-lingual sign data, we find similar improvements for our proposed method.

cs.CL↗

arXiv2026

Joint Pricing and Innovation Control in Regulated Recycling-Rate Diffusion

We introduce a regulated stochastic diffusion model for the recycling rate and formulate a joint control problem over production and process innovation via the dynamics of recycling investment and product pricing. The resulting stochastic control problem captures the system manager's trade-off between product-price decisions and investment expenditures under an infinite-horizon discounted cost structure. Owing to the recycling-rate specification, we incorporate two regulated state processes, which induce additional policy-driven cost components in the value function consistent with green-economy regulations. We resolve the jointly regulated stochastic production and process-innovation admission control problem by introducing the associated Hamilton-Jacobi-Bellman (HJB) equation and providing rigorous proofs that establish the correspondence between the HJB solution and the value function of the underlying control problem. The HJB equation is analyzed under mild, practically motivated assumptions on the system parameters. We further present numerical experiments and sensitivity analyses to illustrate the tractability of the HJB characterization and to assess the practical relevance of the imposed parameter conditions.

math.OC↗

arXiv2026

Designing RNAs with Language Models

RNA design, the task of finding a sequence that folds into a target secondary structure, has broad biological and biomedical impact but remains computationally challenging due to the exponentially large sequence space and exponentially many competing folds. Traditional approaches treat it as an optimization problem, relying on per-instance heuristics or constraint-based search. We instead reframe RNA design as conditional sequence generation and introduce a reusable neural approximator, instantiated as an autoregressive language model (LM), that maps target structures directly to sequences. We first train our model in a supervised setting on random-induced structure-sequence pairs, and then use reinforcement learning (RL) to optimize end-to-end metrics. We also propose methods to select a small subset for RL that greatly improves RL efficiency and quality. Across four datasets, our approach outperforms state-of-the-art systems on key metrics such as Boltzmann probability while being 1.7x faster, establishing conditional LM generation as a scalable, task-agnostic alternative to per-instance optimization for RNA design. Our code and data are available at https://github.com/KuNyaa/RNA-Design-LM.

cs.LG↗

arXiv2025

Admission Control for A Single Server Waiting Time Process in Heavy Traffic

We address a single server queue control problem (QCP) in heavy traffic originating from Lee and Weerasinghe (2011). The state process represents the offered waiting time, the customer arrival has a state-dependent intensity, and the customers' service and patience times are i.i.d with general distributions. We introduce an infinite-horizon discounted cost functional consisting of a control cost generated from the use of control and a penalty for idleness cost. Our primary goal is to tackle the QCP, taking into account a non-trivial control cost and a non-increasing cost function resulting from the control mechanisms in the waiting time. Under mild assumptions, the heavy traffic limit of the QCP yields a stochastic control problem described by a diffusion process, which we call a diffusion control problem (DCP). We find the optimal control of the associated DCP by incorporating the Legendre-Fenchel transform and a formal Hamilton-Jacobi-Bellman (HJB) equation. Then, we ``translate'' this optimal strategy to the QCP, of which we obtain an asymptotically optimal policy. Apart from theoretical results, we also examine the REINFORCE algorithm, a Reinforcement learning (RL) approach, for solving stochastic controls motivated by recent literature. We highlight the advantages and limitations of simulation from theoretical results and data-driven algorithms.

math.OC↗

arXiv2025

An infinitesimal generator approach on weak convergence of regulated multi-class matching systems

We consider a regulated multi-class instantaneous matching system with reneging, in which each event requires $K \geq 2$ distinct impatient agents who wait in their respective queues. Each agent class is subject to a buffer capacity, allowing for the special case without buffers. Due to the instantaneous matching behavior, at any give time, at least one category has an empty queue. Under the Markovian assumption, the system dynamics are described by a Markov chain with innovative rate matrices that capture all possible queue configurations across all classes. To effectively circumvent the structural challenges introduced by instantaneous matching, we establish a non-trivial yet tractable diffusion approximation under heavy traffic conditions by leveraging the infinitesimal generator in conjunction with appropriate regulation and boundary conditions. This asymptotic analysis offers a direct explanation of the dynamics of the regulated coupled heavy-traffic limiting process. Furthermore, we demonstrate the connection between the diffusion-scaled limit derived from the generator approach and the one established in the literature. The latter is typically described by a regulated coupled stochastic integral equation.

math.PR↗

arXiv2024

AEPHORA: AI/ML-Based Energy-Efficient Proactive Handover and Resource Allocation

Future Vehicle-to-Everything (V2X) scenarios require high-speed, low-latency, and ultra-reliable communication services, particularly for applications such as autonomous driving and in-vehicle infotainment. Dense heterogeneous cellular networks, which incorporate both macro and micro base stations, can effectively address these demands. However, they introduce more frequent handovers and higher energy consumption. Proactive handover (PHO) mechanisms can significantly reduce handover delays and failure rates caused by frequent handovers, especially with the mobility prediction capabilities enhanced by artificial intelligence and machine learning (AI/ML) technologies. Nonetheless, the energy-efficient joint optimization of PHO and resource allocation (RA) remains underexplored. In this paper, we propose the AEPHORA framework, which leverages AI/ML-based predictions of vehicular mobility to jointly optimize PHO and RA decisions. This framework aims to minimize the average system transmission power while satisfying quality of service (QoS) constraints on communication delay and reliability. Simulation results demonstrate the effectiveness of the AEPHORA framework in balancing energy efficiency with QoS requirements in high-demand V2X environments.

eess.SY↗

arXiv2024

Multi-component Matching Queues in Heavy Traffic

We consider multi-component matching systems in heavy traffic consisting of $K\geq 2$ distinct perishable components which arrive randomly over time at high speed at the assemble-to-order station, and they wait in their respective queues according to their categories until matched or their ``patience" runs out. An instantaneous match occurs if all categories are available, and the matched components leave immediately thereafter. For a sequence of such systems parameterized by $n$, we establish an explicit definition for the matching completion process, and when all the arrival rates tend to infinity in concert as $n\to\infty$, we obtain a heavy traffic limit of the appropriately scaled queue lengths under mild assumptions, which is characterized by a coupled stochastic integral equation with a scalar-valued non-linear term. We demonstrate some crucial properties for certain coupled equations and exhibit numerical case studies. Moreover, we establish an asymptotic Little's law, which reveals the asymptotic relationship between the queue length and its virtual waiting time. Motivated by the cost structure of blood bank drives, we formulate an infinite-horizon discounted cost functional and show that the expected value of this cost functional for the nth system converges to that of the heavy traffic limiting process as n tends to infinity.

math.PR↗

arXiv2023

Conditional Image Generation with Pretrained Generative Model

In recent years, diffusion models have gained popularity for their ability to generate higher-quality images in comparison to GAN models. However, like any other large generative models, these models require a huge amount of data, computational resources, and meticulous tuning for successful training. This poses a significant challenge, rendering it infeasible for most individuals. As a result, the research community has devised methods to leverage pre-trained unconditional diffusion models with additional guidance for the purpose of conditional image generative. These methods enable conditional image generations on diverse inputs and, most importantly, circumvent the need for training the diffusion model. In this paper, our objective is to reduce the time-required and computational overhead introduced by the addition of guidance in diffusion models -- while maintaining comparable image quality. We propose a set of methods based on our empirical analysis, demonstrating a reduction in computation time by approximately threefold.

cs.CV↗

arXiv2022

MOB-FL: Mobility-Aware Federated Learning for Intelligent Connected Vehicles

Federated learning (FL) is a promising approach to enable the future Internet of vehicles consisting of intelligent connected vehicles (ICVs) with powerful sensing, computing and communication capabilities. We consider a base station (BS) coordinating nearby ICVs to train a neural network in a collaborative yet distributed manner, in order to limit data traffic and privacy leakage. However, due to the mobility of vehicles, the connections between the BS and ICVs are short-lived, which affects the resource utilization of ICVs, and thus, the convergence speed of the training process. In this paper, we propose an accelerated FL-ICV framework, by optimizing the duration of each training round and the number of local iterations, for better convergence performance of FL. We propose a mobility-aware optimization algorithm called MOB-FL, which aims at maximizing the resource utilization of ICVs under short-lived wireless connections, so as to increase the convergence speed. Simulation results based on the beam selection and the trajectory prediction tasks verify the effectiveness of the proposed solution.

cs.LG↗

arXiv2022

MEET: Mobility-Enhanced Edge inTelligence for Smart and Green 6G Networks

Edge intelligence is an emerging paradigm for real-time training and inference at the wireless edge, thus enabling mission-critical applications. Accordingly, base stations (BSs) and edge servers (ESs) need to be densely deployed, leading to huge deployment and operation costs, in particular the energy costs. In this article, we propose a new framework called Mobility-Enhanced Edge inTelligence (MEET), which exploits the sensing, communication, computing, and self-powering capabilities of intelligent connected vehicles for the smart and green 6G networks. Specifically, the operators can incorporate infrastructural vehicles as movable BSs or ESs, and schedule them in a more flexible way to align with the communication and computation traffic fluctuations. Meanwhile, the remaining compute resources of opportunistic vehicles are exploited for edge training and inference, where mobility can further enhance edge intelligence by bringing more compute resources, communication opportunities, and diverse data. In this way, the deployment and operation costs are spread over the vastly available vehicles, so that the edge intelligence is realized cost-effectively and sustainably. Furthermore, these vehicles can be either powered by renewable energy to reduce carbon emissions, or charged more flexibly during off-peak hours to cut electricity bills.

cs.NI↗

arXiv2021

Classification of Diabetic Retinopathy Severity in Fundus Images with DenseNet121 and ResNet50

In this work, deep learning algorithms are used to classify fundus images in terms of diabetic retinopathy severity. Six different combinations of two model architectures, the Dense Convolutional Network-121 and the Residual Neural Network-50 and three image types, RGB, Green, and High Contrast, were tested to find the highest performing combination. We achieved an average validation loss of 0.17 and a max validation accuracy of 85 percent. By testing out multiple combinations, certain combinations of parameters performed better than others, though minimal variance was found overall. Green filtration was shown to perform the poorest, while amplified contrast appeared to have a negligible effect in comparison to RGB analysis. ResNet50 proved to be less of a robust model as opposed to DenseNet121.

eess.IV↗