arXiv subjects

Wei Geng

Publications and source records attributed to Wei Geng.

8 recordsLinked to original sources

arXiv2026

Budget-Adaptive Routing: Skipping the Weak When the Strong Answers Anyway

Edge-cloud inference collaborations are often designed with a routing estimator that decides whether to offload each frame from weak models at the edge to stronger models in the cloud. Existing systems place the routing estimator after the weak detector, so the weak forward pass still runs even on frames that are later offloaded. In this paper, we argue that this weak-conditioned design can be suboptimal when the offload budget varies. First, we present a competitive weak-skipping estimator (0.153 GFLOPs, about 29x lighter than the weak detector at 4.49 GFLOPs) that extracts routing signal from raw pixels, outperforming the common after-weak placement weak-conditioned baselines. Second, we show that neither weak-skipping nor weak-conditioned placement dominates across the full operating curve, and we propose budget-adaptive routing, which selects between them by offload budget via two offline-tuned thresholds. On PASCAL VOC, our budget-adaptive router traces the upper accuracy envelope of both fixed placements across the operating range. Our method reduces per-frame latency by up to 19.1 ms (about 30% lower at rho = 0.9). Besides outperforming SOTA methods, it is surprisingly stronger than the strong model (+1.7 pp over the strong model's peak mAP) at some operating points with far less compute. Artifacts are available at https://github.com/ViGeng/bgt-ada

cs.NI↗

arXiv2025

{\tau}-Ring: A Smart Ring Platform for Multimodal Physiological and Behavioral Sensing

Smart rings have emerged as uniquely convenient devices for continuous physiological and behavioral sensing, offering unobtrusive, constant access to metrics such as heart rate, motion, and skin temperature. Yet most commercial solutions remain proprietary, hindering reproducibility and slowing innovation in wearable research. We introduce {\tau}-Ring, a commercial-ready platform that bridges this gap through: (i) accessible hardware combining time-synchronized multi-channel PPG, 6-axis IMU, temperature sensing, NFC, and on-board storage; (ii) adjustable firmware that lets researchers rapidly reconfigure sampling rates, power modes, and wireless protocols; and (iii) a fully open-source Android software suite that supports both real-time streaming and 8-hour offline logging. Together, these features enable out-of-the-box, reproducible acquisition of rich physiological and behavioral datasets, accelerating prototyping and standardizing experimentation. We validate the platform with demonstration studies in heart-rate monitoring and ring-based handwriting recognition. Source code is available at GitHub: https://github.com/thuhci/OpenRing.

cs.CE↗

arXiv2024

Noise Distribution Decomposition based Multi-Agent Distributional Reinforcement Learning

Generally, Reinforcement Learning (RL) agent updates its policy by repetitively interacting with the environment, contingent on the received rewards to observed states and undertaken actions. However, the environmental disturbance, commonly leading to noisy observations (e.g., rewards and states), could significantly shape the performance of agent. Furthermore, the learning performance of Multi-Agent Reinforcement Learning (MARL) is more susceptible to noise due to the interference among intelligent agents. Therefore, it becomes imperative to revolutionize the design of MARL, so as to capably ameliorate the annoying impact of noisy rewards. In this paper, we propose a novel decomposition-based multi-agent distributional RL method by approximating the globally shared noisy reward by a Gaussian mixture model (GMM) and decomposing it into the combination of individual distributional local rewards, with which each agent can be updated locally through distributional RL. Moreover, a diffusion model (DM) is leveraged for reward generation in order to mitigate the issue of costly interaction expenditure for learning distributions. Furthermore, the optimality of the distribution decomposition is theoretically validated, while the design of loss function is carefully calibrated to avoid the decomposition ambiguity. We also verify the effectiveness of the proposed method through extensive simulation experiments with noisy rewards. Besides, different risk-sensitive policies are evaluated in order to demonstrate the superiority of distributional RL in different MARL tasks.

cs.AI↗

arXiv2023

SoK: Distributed Computing in ICN

Information-Centric Networking (ICN), with its data-oriented operation and generally more powerful forwarding layer, provides an attractive platform for distributed computing. This paper provides a systematic overview and categorization of different distributed computing approaches in ICN encompassing fundamental design principles, frameworks and orchestration, protocols, enablers, and applications. We discuss current pain points in legacy distributed computing, attractive ICN features, and how different systems use them. This paper also provides a discussion of potential future work for distributed computing in ICN.

cs.NI↗

arXiv2023

Error Analysis of an Approximate Optimal Policy for a Non-stationary Inventory System with Setup Costs

In this paper, we consider a finite horizon non-stationary inventory system with setup costs. We detail an algorithm to find an approximate optimal policy and the basic idea is to use numerical procedure for computing integrals involved in the standard method. We provide analytical error bounds, which converge to zero, between the costs of an approximate optimal policy found in this paper and the optimal policy, which are the main contributions of this paper. To the best of our knowledge, the error bound results are not found in the literature on finite horizon inventory systems with setup costs. A convergence result for the approximate optimal policy is also provided. In this paper, algorithms to find the above error bounds are also provided. Several numerical examples show that the performance of Algorithms in this paper is satisfactory.

math.OC↗

arXiv2023

Bright Semiconductor Single-Photon Sources Pumped by Heterogeneously Integrated Micropillar lasers with Electrical Injections

The emerging hybrid integrated quantum photonics combines advantages of different functional components into a single chip to meet the stringent requirements for quantum information processing. Despite the tremendous progress in hybrid integrations of III-V quantum emitters with silicon-based photonic circuits and superconducting single-photon detectors, on-chip optical excitations of quantum emitters via miniaturized lasers towards single-photon sources (SPSs) with low power consumptions, small device footprints and excellent coherence properties is highly desirable yet illusive. In this work, we present realizations of bright semiconductor singe-photon sources heterogeneously integrated with on-chip electrically-injected microlasers. Different from previous one-by-one transfer printing technique implemented in hybrid quantum dot (QD) photonic devices, multiple deterministically coupled QD-circular Bragg Grating (CBG) SPSs were integrated with electrically-injected micropillar lasers at one time via a potentially scalable transfer printing process assisted by the wide-field photoluminescence (PL) imaging technique. Optically pumped by electrically-injected microlasers, pure single photons are generated with a high-brightness of a count rate of 3.8 M/s and an extraction efficiency of 25.44%. Such a high-brightness is due to the enhancement by the cavity mode of the CBG, which is confirmed by a Purcell factor of 2.5. Our work provides a powerful tool for advancing hybrid integrated quantum photonics in general and boosts the developments for realizing highly-compact, energy-efficient and coherent SPSs in particular.

quant-ph↗

arXiv2020

Quantum Annealing for Semi-Supervised Learning

Recent advances in quantum technology have led to the development and the manufacturing of programmable quantum annealers that promise to solve certain combinatorial optimization problems faster than their classical counterparts. Semi-supervised learning is a machine learning technique that makes use of both labeled and unlabeled data for training, which enables a good classifier with only a small amount of labeled data. In this paper, we propose and theoretically analyze a graph-based semi-supervised learning method with the aid of the quantum annealing technique, which efficiently utilize the quantum resources while maintaining a good accuracy. We illustrate two classification examples, suggesting the feasibility of this method even with a small portion (20%) of labeled data is involved.

quant-ph↗

arXiv2018

Plasmon-emitter interaction using integrated ring grating-nanoantenna structures

Overcoming the diffraction limit to achieve high optical resolution is one of the main challenges in the fields of plasmonics, nanooptics and nanophotonics. In this work, we introduce novel plasmonic structures consisting of nanoantennas (nanoprisms, single bowtie nanoantennas and double bowtie nanoantennas) integrated in the center of ring diffraction gratings. Propagating surface plasmon polaritons (SPPs) are generated by the ring grating and coupled with localized surface plasmons (LSPs) at the nanoantennas exciting emitters placed in their gap. SPPs are widely used for optical waveguiding but provide low resolution due to their weak spatial confinement. Oppositely, LSPs provide excellent sub-wavelength confinement but induce large losses. The phenomenon of SPP-LSP coupling witnessed in our structures allows achieving more precise focusing at the nanoscale, causing an increase in the fluorescence emission of the emitters. FDTD simulations as well as experimental fabrication and optical characterization results are presented to study plasmon-emitter coupling between an ensemble of dye molecules and our integrated plasmonic structures. A comparison is given to highlight the importance of each structure on the photoluminescence and radiative decay enhancement of the molecules.

cond-mat.mes-hall↗