Selected work

Representative Papers

Bayesian policy gradient and actor-critic algorithms

Apr 30, 2026

This work addresses the high variance and poor sample efficiency inherent in traditional policy gradient methods, which rely on Monte Carlo estimates, leading to slow convergence. To overcome these limitations, the paper introduces a Bayesian nonparametric approach to policy gradients for the first time, proposing a Gaussian process–based Bayesian actor-critic framework. In this framework, the policy gradient is modeled as a Gaussian process, and a nonparametric Bayesian critic enables analytical posterior computation and explicit quantification of gradient uncertainty. Furthermore, the formulation naturally supports natural gradient updates. Empirical results demonstrate that the proposed method substantially improves gradient estimation accuracy, significantly reduces sample complexity, and accelerates convergence across multiple reinforcement learning tasks.

0 citationsRead paper

Under-Canopy Terrain Reconstruction in Dense Forests Using RGB Imaging and Neural 3D Reconstruction

Jan 30, 2026

This study addresses the challenge of severe canopy occlusion in dense forests, which significantly impedes ground and understory terrain observation and hinders applications such as search-and-rescue and resource surveys. The authors propose a novel neural radiance field (NeRF) approach that relies solely on standard RGB images to achieve high-fidelity 3D reconstruction of forest understory terrain. By optimizing image acquisition under low-light conditions, designing a dedicated low-light loss function, and explicitly removing occluding elements during ray integration, the method eliminates the need for LiDAR or thermal imaging equipment. Evaluated on personnel detection tasks, it outperforms thermal imaging in terms of AOS metrics and demonstrates practical utility through successful tree counting, highlighting its cost-effectiveness and real-world applicability.

0 citationsRead paper

Blank Space: Adaptive Causal Coding for Streaming Communications Over Multi-Hop Networks

Feb 17, 2025

To address the inherent three-way trade-off among throughput, end-to-end latency, and resource efficiency in multi-hop networked streaming, this paper proposes an Adaptive Causal Random Linear Network Coding (AC-RLNC) framework. The method introduces two key innovations: (1) Blank Space periodic scheduling and (2) a “No-New No-FEC” dual suspension mechanism—enabling lightweight, node-autonomous causal-constrained recoding that alleviates channel bottlenecks while preserving data availability. The framework integrates dynamic FEC rate adaptation, idle-period-aware scheduling, and a low-overhead recoding algorithm (NET). Experimental results demonstrate that, compared to standard RLNC baselines, AC-RLNC reduces channel occupancy by 20%, maintains comparable throughput and end-to-end latency, and significantly improves spectral and computational resource utilization efficiency.

0 citationsRead paper

Rule-ATT&CK Mapper (RAM): Mapping SIEM Rules to TTPs Using LLMs

Feb 04, 2025

Manual mapping of SIEM rules to MITRE ATT&CK techniques (TTPs) is inefficient and error-prone, while existing machine learning approaches struggle to handle the structured nature of SIEM rules. Method: We propose the first multi-stage prompt-chaining LLM framework specifically designed for SIEM rule-to-TTP mapping—requiring no pretraining or fine-tuning. It leverages structured prompt engineering and injection of external cybersecurity knowledge to enhance domain-specific TTP identification capabilities of large language models (e.g., GPT-4-Turbo, Qwen, Granite, Mistral). Contribution/Results: Evaluated on the Splunk Security Content dataset, GPT-4-Turbo achieves the highest accuracy. Ablation studies confirm that external knowledge substantially compensates for LLMs’ deficiencies in implicit cybersecurity knowledge. This work establishes a new paradigm for automated, interpretable, and scalable annotation of threat-detection rules.

0 citationsRead paper

Remote Inference over Dynamic Links via Adaptive Rate Deep Task-Oriented Vector Quantization

Jan 05, 2025

To address the challenge that static compression mechanisms fail to adapt to time-varying links in remote inference over dynamic bandwidth-constrained channels, this paper proposes Adaptive Rate Task-Oriented Vector Quantization (ARTOVeQ). Methodologically, ARTOVeQ introduces a nested codebook architecture coupled with a progressive learning algorithm, enabling parallel multi-resolution transmission and successive refinement of inference outputs. It integrates end-to-end joint optimization, task-oriented distortion metrics, successive refinement coding, and nested vector quantization. Experimental results demonstrate that ARTOVeQ achieves near-single-rate performance across multiple bitrates while supporting a wide range of bitrate budgets. Inference quality monotonically improves with increasing transmitted bits, and end-to-end latency is significantly reduced. Overall, ARTOVeQ achieves synergistic optimization of communication efficiency and task accuracy under dynamic channel conditions.

0 citationsRead paper

Recent publications

Latest Papers

Bayesian policy gradient and actor-critic algorithms

Apr 30, 2026

This work addresses the high variance and poor sample efficiency inherent in traditional policy gradient methods, which rely on Monte Carlo estimates, leading to slow convergence. To overcome these limitations, the paper introduces a Bayesian nonparametric approach to policy gradients for the first time, proposing a Gaussian process–based Bayesian actor-critic framework. In this framework, the policy gradient is modeled as a Gaussian process, and a nonparametric Bayesian critic enables analytical posterior computation and explicit quantification of gradient uncertainty. Furthermore, the formulation naturally supports natural gradient updates. Empirical results demonstrate that the proposed method substantially improves gradient estimation accuracy, significantly reduces sample complexity, and accelerates convergence across multiple reinforcement learning tasks.

0 citationsRead paper

Under-Canopy Terrain Reconstruction in Dense Forests Using RGB Imaging and Neural 3D Reconstruction

Jan 30, 2026

This study addresses the challenge of severe canopy occlusion in dense forests, which significantly impedes ground and understory terrain observation and hinders applications such as search-and-rescue and resource surveys. The authors propose a novel neural radiance field (NeRF) approach that relies solely on standard RGB images to achieve high-fidelity 3D reconstruction of forest understory terrain. By optimizing image acquisition under low-light conditions, designing a dedicated low-light loss function, and explicitly removing occluding elements during ray integration, the method eliminates the need for LiDAR or thermal imaging equipment. Evaluated on personnel detection tasks, it outperforms thermal imaging in terms of AOS metrics and demonstrates practical utility through successful tree counting, highlighting its cost-effectiveness and real-world applicability.

0 citationsRead paper

Blank Space: Adaptive Causal Coding for Streaming Communications Over Multi-Hop Networks

Feb 17, 2025

To address the inherent three-way trade-off among throughput, end-to-end latency, and resource efficiency in multi-hop networked streaming, this paper proposes an Adaptive Causal Random Linear Network Coding (AC-RLNC) framework. The method introduces two key innovations: (1) Blank Space periodic scheduling and (2) a “No-New No-FEC” dual suspension mechanism—enabling lightweight, node-autonomous causal-constrained recoding that alleviates channel bottlenecks while preserving data availability. The framework integrates dynamic FEC rate adaptation, idle-period-aware scheduling, and a low-overhead recoding algorithm (NET). Experimental results demonstrate that, compared to standard RLNC baselines, AC-RLNC reduces channel occupancy by 20%, maintains comparable throughput and end-to-end latency, and significantly improves spectral and computational resource utilization efficiency.

0 citationsRead paper

Rule-ATT&CK Mapper (RAM): Mapping SIEM Rules to TTPs Using LLMs

Feb 04, 2025

Manual mapping of SIEM rules to MITRE ATT&CK techniques (TTPs) is inefficient and error-prone, while existing machine learning approaches struggle to handle the structured nature of SIEM rules. Method: We propose the first multi-stage prompt-chaining LLM framework specifically designed for SIEM rule-to-TTP mapping—requiring no pretraining or fine-tuning. It leverages structured prompt engineering and injection of external cybersecurity knowledge to enhance domain-specific TTP identification capabilities of large language models (e.g., GPT-4-Turbo, Qwen, Granite, Mistral). Contribution/Results: Evaluated on the Splunk Security Content dataset, GPT-4-Turbo achieves the highest accuracy. Ablation studies confirm that external knowledge substantially compensates for LLMs’ deficiencies in implicit cybersecurity knowledge. This work establishes a new paradigm for automated, interpretable, and scalable annotation of threat-detection rules.

0 citationsRead paper

Remote Inference over Dynamic Links via Adaptive Rate Deep Task-Oriented Vector Quantization

Jan 05, 2025

To address the challenge that static compression mechanisms fail to adapt to time-varying links in remote inference over dynamic bandwidth-constrained channels, this paper proposes Adaptive Rate Task-Oriented Vector Quantization (ARTOVeQ). Methodologically, ARTOVeQ introduces a nested codebook architecture coupled with a progressive learning algorithm, enabling parallel multi-resolution transmission and successive refinement of inference outputs. It integrates end-to-end joint optimization, task-oriented distortion metrics, successive refinement coding, and nested vector quantization. Experimental results demonstrate that ARTOVeQ achieves near-single-rate performance across multiple bitrates while supporting a wide range of bitrate budgets. Inference quality monotonically improves with increasing transmitted bits, and end-to-end latency is significantly reduced. Overall, ARTOVeQ achieves synergistic optimization of communication efficiency and task accuracy under dynamic channel conditions.

0 citationsRead paper