Selected work

Representative Papers

Characterizing Nonlinearities in IM-DD Links via the Best Linear Approximation: Distortion Analysis and Modulation Optimization

Aug 12, 2026

This work addresses the challenge of modeling and compensating nonlinear distortions in intensity-modulation direct-detection (IM-DD) optical links, which significantly limits system performance. By introducing the Best Linear Approximation (BLA) framework, the study reveals—for the first time—that the orthogonal component of the nonlinear distortion exhibits non-Gaussian statistical characteristics. Leveraging this insight, the authors develop an optimization strategy for modulation depth tailored to various equalizer architectures based on BLA analysis. This approach effectively characterizes the nonlinear impairments and enables near-optimal modulation depth selection across different equalizers, thereby substantially enhancing overall system performance.

0 citationsRead paper

A Photonic-CXL Memory Appliance for Scalable KV Cache Management in LLM Inference

Jul 29, 2026

This work addresses the memory capacity and bandwidth limitations of KV caching in large language model inference, which hinder simultaneous support for terabyte-scale memory and hundreds of gigabytes per second throughput. The authors propose a novel optoelectronic hybrid CXL memory architecture that introduces passive optical interconnects into a CXL memory pool, enabling a switchless full-mesh topology that allows 16 hosts to share 32 TB of memory. By eliminating electrical switches, the design circumvents bottlenecks in latency, reach, and power consumption, yielding a scalable low-latency shared memory system. Experimental results demonstrate over 50% lower system latency compared to an all-electrical CXL baseline and a 6.6× speedup in first-token generation time under multi-turn dialogue scenarios, effectively mitigating the performance cliff caused by cache eviction.

0 citationsRead paper

Simulating Unified Tensor Resharding in heterogeneous AI systems

Jun 25, 2026

This work addresses the limitation of existing AI training simulators, which often assume homogeneous compute and network infrastructure and thus fail to accurately model real-world heterogeneous training environments. To overcome this, the authors propose Xsim—the first distributed simulator supporting heterogeneous large model training—that innovatively unifies the simulation of non-uniform tensor resharding and pipeline parallelism. Xsim introduces heterogeneity-aware communication modeling, reusable parallel algorithm abstractions, and a flexible deployment mechanism. It integrates techniques such as custom ring construction, block partitioning, and plugins from NS-3 and htsim to achieve high-fidelity, scalable performance simulation. Experimental results demonstrate that Xsim achieves training time prediction errors below 5% across diverse heterogeneous configurations, with pipeline communication modeling errors around 2%, while also providing key metrics such as pipeline bubble duration and straggler waiting time.

0 citationsRead paper

Simulating LLM training workloads for heterogeneous compute and network infrastructure

Aug 07, 2025

Existing LLM training simulators commonly assume hardware homogeneity, failing to capture real-world performance deviations arising from resource sharing, device-generation heterogeneity, and on-chip interconnect diversity in cloud environments. To address this, we propose the first distributed LLM training simulator supporting fine-grained device heterogeneity. Our approach systematically models non-uniform computation and communication capabilities, enabling customizable hardware topologies and parallelism strategy abstractions. We further design a non-uniform workload partitioning mechanism that integrates computation-communication overlap modeling, fine-grained pipeline scheduling, and heterogeneous resource mapping. Experimental evaluation demonstrates that our simulator faithfully reproduces actual training time trends—achieving significantly higher fidelity than homogeneous baselines. This enables reliable performance assessment for training optimization and architecture co-design in heterogeneous systems.

0 citationsRead paper

The 2nd Place Solution from the 3D Semantic Segmentation Track in the 2024 Waymo Open Dataset Challenge

Jan 06, 2025

To address poor generalization on long-tailed classes and insufficient training data diversity in LiDAR point cloud 3D semantic segmentation, this paper proposes MixSeg3D—a novel framework built upon MinkUNet. It is the first to jointly integrate LaserMix (mixing along the elevation angle dimension) and PolarMix (mixing along the azimuth angle dimension) in the vehicle coordinate system at the scene level. The framework further incorporates multi-scale feature fusion and fine-grained preprocessing tailored for the Waymo Open Dataset. This cross-dimensional mixing strategy significantly improves model robustness and generalization, particularly for sparse categories and complex urban scenes. Evaluated on the 2024 Waymo Open Dataset 3D semantic segmentation benchmark, MixSeg3D achieves second place, with a substantial improvement in mean Intersection-over-Union (mIoU) over both the baseline and prior state-of-the-art methods—empirically validating the efficacy of the proposed hybrid augmentation paradigm.

0 citationsRead paper

Recent publications

Latest Papers

Characterizing Nonlinearities in IM-DD Links via the Best Linear Approximation: Distortion Analysis and Modulation Optimization

Aug 12, 2026

This work addresses the challenge of modeling and compensating nonlinear distortions in intensity-modulation direct-detection (IM-DD) optical links, which significantly limits system performance. By introducing the Best Linear Approximation (BLA) framework, the study reveals—for the first time—that the orthogonal component of the nonlinear distortion exhibits non-Gaussian statistical characteristics. Leveraging this insight, the authors develop an optimization strategy for modulation depth tailored to various equalizer architectures based on BLA analysis. This approach effectively characterizes the nonlinear impairments and enables near-optimal modulation depth selection across different equalizers, thereby substantially enhancing overall system performance.

0 citationsRead paper

A Photonic-CXL Memory Appliance for Scalable KV Cache Management in LLM Inference

Jul 29, 2026

This work addresses the memory capacity and bandwidth limitations of KV caching in large language model inference, which hinder simultaneous support for terabyte-scale memory and hundreds of gigabytes per second throughput. The authors propose a novel optoelectronic hybrid CXL memory architecture that introduces passive optical interconnects into a CXL memory pool, enabling a switchless full-mesh topology that allows 16 hosts to share 32 TB of memory. By eliminating electrical switches, the design circumvents bottlenecks in latency, reach, and power consumption, yielding a scalable low-latency shared memory system. Experimental results demonstrate over 50% lower system latency compared to an all-electrical CXL baseline and a 6.6× speedup in first-token generation time under multi-turn dialogue scenarios, effectively mitigating the performance cliff caused by cache eviction.

0 citationsRead paper

Simulating Unified Tensor Resharding in heterogeneous AI systems

Jun 25, 2026

This work addresses the limitation of existing AI training simulators, which often assume homogeneous compute and network infrastructure and thus fail to accurately model real-world heterogeneous training environments. To overcome this, the authors propose Xsim—the first distributed simulator supporting heterogeneous large model training—that innovatively unifies the simulation of non-uniform tensor resharding and pipeline parallelism. Xsim introduces heterogeneity-aware communication modeling, reusable parallel algorithm abstractions, and a flexible deployment mechanism. It integrates techniques such as custom ring construction, block partitioning, and plugins from NS-3 and htsim to achieve high-fidelity, scalable performance simulation. Experimental results demonstrate that Xsim achieves training time prediction errors below 5% across diverse heterogeneous configurations, with pipeline communication modeling errors around 2%, while also providing key metrics such as pipeline bubble duration and straggler waiting time.

0 citationsRead paper

Simulating LLM training workloads for heterogeneous compute and network infrastructure

Aug 07, 2025

Existing LLM training simulators commonly assume hardware homogeneity, failing to capture real-world performance deviations arising from resource sharing, device-generation heterogeneity, and on-chip interconnect diversity in cloud environments. To address this, we propose the first distributed LLM training simulator supporting fine-grained device heterogeneity. Our approach systematically models non-uniform computation and communication capabilities, enabling customizable hardware topologies and parallelism strategy abstractions. We further design a non-uniform workload partitioning mechanism that integrates computation-communication overlap modeling, fine-grained pipeline scheduling, and heterogeneous resource mapping. Experimental evaluation demonstrates that our simulator faithfully reproduces actual training time trends—achieving significantly higher fidelity than homogeneous baselines. This enables reliable performance assessment for training optimization and architecture co-design in heterogeneous systems.

0 citationsRead paper

The 2nd Place Solution from the 3D Semantic Segmentation Track in the 2024 Waymo Open Dataset Challenge

Jan 06, 2025

To address poor generalization on long-tailed classes and insufficient training data diversity in LiDAR point cloud 3D semantic segmentation, this paper proposes MixSeg3D—a novel framework built upon MinkUNet. It is the first to jointly integrate LaserMix (mixing along the elevation angle dimension) and PolarMix (mixing along the azimuth angle dimension) in the vehicle coordinate system at the scene level. The framework further incorporates multi-scale feature fusion and fine-grained preprocessing tailored for the Waymo Open Dataset. This cross-dimensional mixing strategy significantly improves model robustness and generalization, particularly for sparse categories and complex urban scenes. Evaluated on the 2024 Waymo Open Dataset 3D semantic segmentation benchmark, MixSeg3D achieves second place, with a substantial improvement in mean Intersection-over-Union (mIoU) over both the baseline and prior state-of-the-art methods—empirically validating the efficacy of the proposed hybrid augmentation paradigm.

0 citationsRead paper