Boyuan Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Boyuan Wang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Wenbin Shang
University of Glasgow, Glasgow, U.K.
2 shared publications - Wenyu Zhao
Microsoft, Beijing, China
2 shared publications - Zimeng Wang
Brandeis University, Waltham, MA, USA
2 shared publications - Fengyuan Zhang
Tandon School of Engineering, New York University, New York, NY 10010, USA
1 shared publication - Han Lin
University of WisconsinMadison, Madison, WI, USA
1 shared publication - Shuxin Zhang
University of California, Berkeley, Berkeley, CA, USA
1 shared publication
Latest Publications
- Open Access
- 19 August 2026
- Published by Elsevier in SSRN Electronic Journal
SafeRL-Converge: Risk-Aware Deep Reinforcement Learning for Fast and Reliable Routing Convergence in Large-Scale Data Center Networks
Fast control-plane recovery after a link failure is difficult in a large data center because a logically centralized controller cannot make all forwarding-state changes appear simultaneously. A large update batch reduces the number of control rounds, but mixed old/new forwarding states can transiently route traffic onto the failed link, create blackholes or loops, or overload a link even when both endpoint configurations are individually acceptable. This paper presents SafeRL-Converge, a concrete learning-and-certification algorithm for device-granularity routing convergence. The policy is a graph neural network trained with proximal policy optimization to rank pending switches under a fixed per-round programming budget. A deterministic safety projector then certifies the proposed batch against two explicit constraints: non-increasing traffic-weighted forwarding failure/loop mass and a tail-load budget expressed as CVaR 0.9 (MLU). When a ranked subset cannot be certified, SafeRL-Convergeconstructs a device-dependency graph induced by destination next hops, extracts sink strongly connected components as dependency-closed update groups, and tests these groups before declaring a safe stall. Thus learning is used to reduce convergence cost, while safety does not depend on the neural policy being correct. We evaluate the method using public, downloadable traffic matrices from the SDN-TMprediction testbed dataset and, without fine-tuning, a public G ÉANT matrix set for cross-dataset trace-driven robustness. Canonical k = 4, 6, 8 fat-tree/Clos fabrics are synthesized because the public matrices do not include production data-center switch-level topology. The model is trained only on k = 4 and k = 6 and evaluated on a fixed 30-scenario test set per topology; k = 8 is therefore an unseen scale. Under the predefined risk budget, 89 of 90 main target configurations are feasible. On all feasible scenarios, the full method completes with zero committed topology or CVaR violations. On unseen k = 8, it converges in 16.50 rounds versus 16.63 for Random+Shield while reducing transient disruption area by 80.5%; versus Impact+Shield it uses 25.9% fewer rounds and reduces disruption by 69.9%. Removing the CVaR shield causes risk violations in 6.7% of k = 8 scenarios; removing the topology shield causes topology violations in 30.0%; replacing the GNN with a node-wise MLP increases the mean convergence time from 16.50 to 22.20 rounds. Three independently trained seeds remain violation-free. A controlled CPU benchmark reports 16.22 ms per decision at k = 8 (80 switches). All reported numbers are produced by the accompanying code and CSV files; no experimental value is manually fabricated.
- 19 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessSafeRL-Converge: Risk-Aware Deep Reinforcement Learning for Fast and Reliable Routing Convergence in Large-Scale Data Center Networks
Fast control-plane recovery after a link failure is difficult in a large data center because a logically centralized controller cannot make all forwarding-state changes appear simultaneously. A large update batch reduces the number of control rounds, but mixed old/new forwarding states can transiently route traffic onto the failed link, create blackholes or loops, or overload a link even when both endpoint configurations are individually acceptable. This paper presents SafeRL-Converge, a concrete learning-and-certification algorithm for device-granularity routing convergence. The policy is a graph neural network trained with proximal policy optimization to rank pending switches under a fixed per-round programming budget. A deterministic safety projector then certifies the proposed batch against two explicit constraints: non-increasing traffic-weighted forwarding failure/loop mass and a tail-load budget expressed as CVaR 0.9 (MLU). When a ranked subset cannot be certified, SafeRL-Convergeconstructs a device-dependency graph induced by destination next hops, extracts sink strongly connected components as dependency-closed update groups, and tests these groups before declaring a safe stall. Thus learning is used to reduce convergence cost, while safety does not depend on the neural policy being correct. We evaluate the method using public, downloadable traffic matrices from the SDN-TMprediction testbed dataset and, without fine-tuning, a public G ÉANT matrix set for cross-dataset trace-driven robustness. Canonical k = 4, 6, 8 fat-tree/Clos fabrics are synthesized because the public matrices do not include production data-center switch-level topology. The model is trained only on k = 4 and k = 6 and evaluated on a fixed 30-scenario test set per topology; k = 8 is therefore an unseen scale. Under the predefined risk budget, 89 of 90 main target configurations are feasible. On all feasible scenarios, the full method completes with zero committed topology or CVaR violations. On unseen k = 8, it converges in 16.50 rounds versus 16.63 for Random+Shield while reducing transient disruption area by 80.5%; versus Impact+Shield it uses 25.9% fewer rounds and reduces disruption by 69.9%. Removing the CVaR shield causes risk violations in 6.7% of k = 8 scenarios; removing the topology shield causes topology violations in 30.0%; replacing the GNN with a node-wise MLP increases the mean convergence time from 16.50 to 22.20 rounds. Three independently trained seeds remain violation-free. A controlled CPU benchmark reports 16.22 ms per decision at k = 8 (80 switches). All reported numbers are produced by the accompanying code and CSV files; no experimental value is manually fabricated.
- Top Cited PapersOpen Access
- 24 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Computer Society
- Vol. 7, 849-863
DRL-Adapt: Deep Reinforcement Learning for Adaptive Routing Convergence Optimization in Large-Scale Networks
Large-scale fully-meshed networks often experience prolonged routing convergence after failures because routing protocol parameters are statically configured and cannot adapt to changing network conditions. To address this limitation, this paper proposes DRL-Adapt, a deep reinforcement learning-based framework that dynamically tunes routing protocol parameters using network event-stream observations. The framework employs a dueling prioritized deep Q-network agent to jointly adjust Hello/Dead interval timers and routing weights according to real-time network states. We evaluate the proposed method using a high-fidelity simulator constructed from CAIDA topologies under diverse failure scenarios and network scales. Experimental results demonstrate that DRL-Adapt significantly accelerates recovery and reduces control overhead compared with conventional static configurations. On average, the proposed approach decreases fault convergence time by 32.7% and lowers control-plane overhead by 24.3%, while preserving routing stability. Moreover, the learned policy generalizes across networks ranging from 50 to 500 nodes and incurs minimal inference latency ($< $ 15 ms/decision), indicating practicality for practical deployment.
- 24 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of the Computer Society
- Vol. 7, 849-863
Top Cited PapersOpen AccessDRL-Adapt: Deep Reinforcement Learning for Adaptive Routing Convergence Optimization in Large-Scale Networks
Large-scale fully-meshed networks often experience prolonged routing convergence after failures because routing protocol parameters are statically configured and cannot adapt to changing network conditions. To address this limitation, this paper proposes DRL-Adapt, a deep reinforcement learning-based framework that dynamically tunes routing protocol parameters using network event-stream observations. The framework employs a dueling prioritized deep Q-network agent to jointly adjust Hello/Dead interval timers and routing weights according to real-time network states. We evaluate the proposed method using a high-fidelity simulator constructed from CAIDA topologies under diverse failure scenarios and network scales. Experimental results demonstrate that DRL-Adapt significantly accelerates recovery and reduces control overhead compared with conventional static configurations. On average, the proposed approach decreases fault convergence time by 32.7% and lowers control-plane overhead by 24.3%, while preserving routing stability. Moreover, the learned policy generalizes across networks ranging from 50 to 500 nodes and incurs minimal inference latency ($< $ 15 ms/decision), indicating practicality for practical deployment.
- Top Cited PapersOpen Access
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 38544-38562
CAGR: A Cross-Accelerator Graph Optimization Framework for Efficient Recommender System Inference
Recommender systems have become ubiquitous in modern online services, yet their deployment across diverse hardware accelerators remains challenging due to significant performance variations. Contemporary deep learning recommendation models (DLRMs), such as DeepFM and NGCF, exhibit substantial inference latency differences when executed on NVIDIA GPUs,AMDGPUs, and Google TPUs, primarily due to architectural disparities and vendor-specific optimization strategies. Existing graph optimization frameworks are typically designed for specific hardware backends, lacking the flexibility to generate portable high-performance implementations across heterogeneous accelerators. This paper presents CAGR (Cross-Accelerator Graph Rewriting), a novel framework that achieves performance-portable inference optimization for recommendation models through three key innovations: (1) a hardware-aware graph rewriting engine that dynamically selects optimal operator implementations by analyzing compute-to-memory bandwidth ratios and operator density characteristics; (2) a reinforcement learning-based transformation policy that learns cross-platform optimization strategies without exhaustive search; and (3) a heterogeneous pipeline architecture enabling “optimize once, deploy across supported backends” semantics. We implement CAGR with support for multiple kernel backends including Triton, cuBLAS, MIOpen, and oneDNN, and demonstrate its effectiveness on the Avazu CTR prediction dataset. Experimental results show that CAGR achieves 1.8-3.2× speedup over baseline implementations across NVIDIA V100, AMD MI100, and Google TPU v3 platforms, while reducing optimization time by 67% compared to platform-specific auto-tuning approaches. Furthermore, CAGR maintains 92-96% of reference optimized performance with zero manual intervention, demonstrating practical viability for production recommendation systems requiring multi-vendor deployment.
- 9 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 38544-38562
Top Cited PapersOpen AccessCAGR: A Cross-Accelerator Graph Optimization Framework for Efficient Recommender System Inference
Recommender systems have become ubiquitous in modern online services, yet their deployment across diverse hardware accelerators remains challenging due to significant performance variations. Contemporary deep learning recommendation models (DLRMs), such as DeepFM and NGCF, exhibit substantial inference latency differences when executed on NVIDIA GPUs,AMDGPUs, and Google TPUs, primarily due to architectural disparities and vendor-specific optimization strategies. Existing graph optimization frameworks are typically designed for specific hardware backends, lacking the flexibility to generate portable high-performance implementations across heterogeneous accelerators. This paper presents CAGR (Cross-Accelerator Graph Rewriting), a novel framework that achieves performance-portable inference optimization for recommendation models through three key innovations: (1) a hardware-aware graph rewriting engine that dynamically selects optimal operator implementations by analyzing compute-to-memory bandwidth ratios and operator density characteristics; (2) a reinforcement learning-based transformation policy that learns cross-platform optimization strategies without exhaustive search; and (3) a heterogeneous pipeline architecture enabling “optimize once, deploy across supported backends” semantics. We implement CAGR with support for multiple kernel backends including Triton, cuBLAS, MIOpen, and oneDNN, and demonstrate its effectiveness on the Avazu CTR prediction dataset. Experimental results show that CAGR achieves 1.8-3.2× speedup over baseline implementations across NVIDIA V100, AMD MI100, and Google TPU v3 platforms, while reducing optimization time by 67% compared to platform-specific auto-tuning approaches. Furthermore, CAGR maintains 92-96% of reference optimized performance with zero manual intervention, demonstrating practical viability for production recommendation systems requiring multi-vendor deployment.
- Top Cited PapersOpen Access
- 13 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 9319-9339
Adaptive Robust Watermarking for Large Language Models via Dynamic Token Embedding Perturbation
Large Language Models (LLMs) have demonstrated remarkable capabilities in generating high-quality text, raising significant concerns regarding copyright protection and content provenance verification. However, most existing watermarking techniques rely on uniform perturbation or rule-based token biasing schemes, which exhibit critical vulnerabilities under adversarial attacks such as paraphrasing, translation, and content truncation, often failing to maintain detection reliability in real-world deployment scenarios. To address these challenges, this paper introduces a novel context-aware robust watermarking framework that dynamically adjusts watermark embedding strength according to contextual semantic characteristics during text generation. The proposed approach incorporates a token-level semantic modulation mechanism that strategically intensifies watermark signals in copyright-sensitive segments while minimizing perturbations in semantically neutral regions, achieving an improved balance between imperceptibility and robustness. Furthermore, an adaptive threshold estimation algorithm is developed for watermark detection, which automatically calibrates detection boundaries based on noise statistics, significantly enhancing resilience against diverse attack vectors. Extensive experiments on the WaterBench benchmark demonstrate superior performance over state-of-the-art baselines, maintaining high detection accuracy with a 95.3% true positive rate (TPR) under clean conditions and strong robustness under severe perturbations, including paraphrasing attacks (82.7% TPR), translation attacks (78.4% TPR), and content truncation (88.9% TPR at 50% retention). Meanwhile, the proposed method reduces false positive rates by 43.2% compared with existing approaches while preserving text quality with negligible perplexity increase (1.8%). These results establish a new paradigm for practical and scalable LLM watermarking in real-world copyright-sensitive deployment scenarios.
- 13 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 9319-9339
Top Cited PapersOpen AccessAdaptive Robust Watermarking for Large Language Models via Dynamic Token Embedding Perturbation
Large Language Models (LLMs) have demonstrated remarkable capabilities in generating high-quality text, raising significant concerns regarding copyright protection and content provenance verification. However, most existing watermarking techniques rely on uniform perturbation or rule-based token biasing schemes, which exhibit critical vulnerabilities under adversarial attacks such as paraphrasing, translation, and content truncation, often failing to maintain detection reliability in real-world deployment scenarios. To address these challenges, this paper introduces a novel context-aware robust watermarking framework that dynamically adjusts watermark embedding strength according to contextual semantic characteristics during text generation. The proposed approach incorporates a token-level semantic modulation mechanism that strategically intensifies watermark signals in copyright-sensitive segments while minimizing perturbations in semantically neutral regions, achieving an improved balance between imperceptibility and robustness. Furthermore, an adaptive threshold estimation algorithm is developed for watermark detection, which automatically calibrates detection boundaries based on noise statistics, significantly enhancing resilience against diverse attack vectors. Extensive experiments on the WaterBench benchmark demonstrate superior performance over state-of-the-art baselines, maintaining high detection accuracy with a 95.3% true positive rate (TPR) under clean conditions and strong robustness under severe perturbations, including paraphrasing attacks (82.7% TPR), translation attacks (78.4% TPR), and content truncation (88.9% TPR at 50% retention). Meanwhile, the proposed method reduces false positive rates by 43.2% compared with existing approaches while preserving text quality with negligible perplexity increase (1.8%). These results establish a new paradigm for practical and scalable LLM watermarking in real-world copyright-sensitive deployment scenarios.