Bowen Pang

Weifang Zhang, Yuzhou Nie, Bowen Pang et al.

Mixed batching (MB)--interleaving prefill and decode in a single batch--has become the standard scheduling strategy for large language model (LLM) inference due to its efficiency in maximizing compute and memory utilization. However, through controlled experiments, we find that prefill-decode interference inflates MB's per-step marginal cost above that of pure decode. On the high-bandwidth H200 (4.8 TB/s), this occurs only when decode tokens exceed 80% of the batch; however, on the bandwidth-constrained RTX PRO 6000 (1.792 TB/s), this threshold plummets to just 20%. Consequently, the optimal choice between MB and exclusive batching (EB) fundamentally depends on GPU memory bandwidth, model size, and workload composition. We derive a closed-form condition for this EB-MB performance crossover, along with asymptotically optimal phase-switching thresholds and memory-safe batch sizing for EB. Optimized EB achieves up to 41.9% higher throughput on bandwidth-constrained GPUs, while MB retains its advantage on high-bandwidth hardware with larger models. Our hybrid scheduler EB+ applies this condition online to dynamically switch between EB and MB without manual intervention. Under non-stationary traffic with distribution or concurrency shifts, EB+ attains the highest or near-highest throughput in every setting, outperforming MB by up to 36.4%.

Bowen Pang, Huan Zhao, Gaosheng Zhang et al.

This paper describes the TSUP team's submission to the ISCSLP 2022 conversational short-phrase speaker diarization (CSSD) challenge which particularly focuses on short-phrase conversations with a new evaluation metric called conversational diarization error rate (CDER). In this challenge, we explore three kinds of typical speaker diarization systems, which are spectral clustering(SC) based diarization, target-speaker voice activity detection(TS-VAD) and end-to-end neural diarization(EEND) respectively. Our major findings are summarized as follows. First, the SC approach is more favored over the other two approaches under the new CDER metric. Second, tuning on hyperparameters is essential to CDER for all three types of speaker diarization systems. Specifically, CDER becomes smaller when the length of sub-segments setting longer. Finally, multi-system fusion through DOVER-LAP will worsen the CDER metric on the challenge data. Our submitted SC system eventually ranks the third place in the challenge.

Jingcheng Zhou, Wei Wei, Xing Li et al.

Deep learning utilizing deep neural networks (DNNs) has achieved a lot of success recently in many important areas such as computer vision, natural language processing, and recommendation systems. The lack of convexity for DNNs has been seen as a major disadvantage of many optimization methods, such as stochastic gradient descent, which greatly reduces the genelization of neural network applications. We realize that the convexity make sense in the neural network and propose the exponential multilayer neural network (EMLP), a class of parameter convex neural network (PCNN) which is convex with regard to the parameters of the neural network under some conditions that can be realized. Besides, we propose the convexity metric for the two-layer EGCN and test the accuracy when the convexity metric changes. For late experiments, we use the same architecture to make the exponential graph convolutional network (EGCN) and do the experiment on the graph classificaion dataset in which our model EGCN performs better than the graph convolutional network (GCN) and the graph attention network (GAT).

Qiming Yang, Wei Wei, Ruizhi Zhang et al.

Most real-world networks suffer from incompleteness or incorrectness, which is an inherent attribute to real-world datasets. As a consequence, those downstream machine learning tasks in complex network like community detection methods may yield less satisfactory results, i.e., a proper preprocessing measure is required here. To address this issue, in this paper, we design a new community attribute based link prediction strategy HAP and propose a two-step community enhancement algorithm with automatic evolution process based on HAP. This paper aims at providing a community enhancement measure through adding links to clarify ambiguous community structures. The HAP method takes the neighbourhood uncertainty and Shannon entropy to identify boundary nodes, and establishes links by considering the nodes' community attributes and community size at the same time. The experimental results on twelve real-world datasets with ground truth community indicate that the proposed link prediction method outperforms other baseline methods and the enhancement of community follows the expected evolution process.

Zilin Zhu, Longteng Guo, Yanghong Mei et al.

Long-horizon household tasks demand robust high-level planning and sustained reasoning capabilities, which are largely overlooked by existing embodied AI benchmarks that emphasize short-horizon navigation or manipulation and rely on fixed task categories. We introduce LongAct, a benchmark designed to evaluate planning-level autonomy in long-horizon household tasks specified through free-form instructions. By abstracting away embodiment-specific low-level control, LongAct isolates high-level cognitive capabilities such as instruction understanding, dependency management, memory maintenance, and adaptive planning. We further propose HoloMind, a VLM-driven agent with a DAG-based long-horizon hierarchical planner, a Multimodal Spatial Memory for persistent world modeling, an Episodic Memory for experience reuse, and a global Critic for reflective supervision. Experiments with GPT-5 and Qwen3-VL models show that HoloMind substantially improves long-horizon performance while reducing reliance on model scale. Even top models achieve only 59% goal completion and 16% full-task success, underscoring the difficulty of LongAct and the need for stronger long-horizon planning in embodied agents.

Shaohong Wang, Bin Lu, Xinyu Xiao et al.

Collaborative visual perception methods have gained widespread attention in the autonomous driving community in recent years due to their ability to address sensor limitation problems. However, the absence of explicit depth information often makes it difficult for camera-based perception systems, e.g., 3D object detection, to generate accurate predictions. To alleviate the ambiguity in depth estimation, we propose RayFusion, a ray-based fusion method for collaborative visual perception. Using ray occupancy information from collaborators, RayFusion reduces redundancy and false positive predictions along camera rays, enhancing the detection performance of purely camera-based collaborative perception systems. Comprehensive experiments show that our method consistently outperforms existing state-of-the-art models, substantially advancing the performance of collaborative visual perception. The code is available at https://github.com/wangsh0111/RayFusion.

Zhu Yu, Bowen Pang, Lizhe Liu et al.

We introduce LOcc, an effective and generalizable framework for open-vocabulary occupancy (OVO) prediction. Previous approaches typically supervise the networks through coarse voxel-to-text correspondences via image features as intermediates or noisy and sparse correspondences from voxel-based model-view projections. To alleviate the inaccurate supervision, we propose a semantic transitive labeling pipeline to generate dense and fine-grained 3D language occupancy ground truth. Our pipeline presents a feasible way to dig into the valuable semantic information of images, transferring text labels from images to LiDAR point clouds and ultimately to voxels, to establish precise voxel-to-text correspondences. By replacing the original prediction head of supervised occupancy models with a geometry head for binary occupancy states and a language head for language features, LOcc effectively uses the generated language ground truth to guide the learning of 3D language volume. Through extensive experiments, we demonstrate that our transitive semantic labeling pipeline can produce more accurate pseudo-labeled ground truth, diminishing labor-intensive human annotations. Additionally, we validate LOcc across various architectures, where all models consistently outperform state-of-the-art zero-shot occupancy prediction approaches on the Occ3D-nuScenes dataset.

Bowen Pang, Kai Li, Ruifeng She et al.

With the development of large language models (LLMs), it has become increasingly important to optimize hardware usage and improve throughput. In this paper, we study the inference optimization of the serving system that deploys LLMs. To optimize system throughput and maximize hardware utilization, we formulate the inference optimization problem as a mixed-integer programming (MIP) model and propose a hybrid offline-online method as solution. The offline method improves large-scale inference systems by introducing a Minimizing Makespan Bin Packing Problem. We further provide a theoretical lower bound computation method. Then, we propose an online sorting and preemptive scheduling method to better utilize hardware. In the online iteration scheduling process, a Lagrangian method is applied to evaluate the cost efficiency of inserting prefill stages versus decode stages at each iteration and dynamically determine when to preempt decoding tasks and insert prefill tasks. Experiments using real-world data from the LLaMA-65B model and the GSM8K dataset demonstrate that system utilization improves from 80.2% to 89.1%, and the total inference time decreases from 201.00 to 190.58 seconds. A 100-cases study shows that our method consistently outperforms the baseline method and improves the utilization rate by 8.0% on average. Finally, we discuss potential future extensions, including stochastic modeling, reinforcement learning-based schedulers, and dynamic decision-making strategies for system throughput and hardware utilization.

Sicong Liu, Weiye Wu, Xiangrui Xu et al.

Foundation models have reshaped AI by unifying fragmented architectures into scalable backbones with multimodal reasoning and contextual adaptation. In parallel, the long-standing notion of AI agents, defined by the sensing-decision-action loop, is entering a new paradigm: with FMs as their cognitive core, agents transcend rule-based behaviors to achieve autonomy, generalization, and self-reflection. This dual shift is reinforced by real-world demands such as autonomous driving, robotics, virtual assistants, and GUI agents, as well as ecosystem advances in embedded hardware, edge computing, mobile deployment platforms, and communication protocols that together enable large-scale deployment. Yet this convergence collides with reality: while applications demand long-term adaptability and real-time interaction, mobile and edge deployments remain constrained by memory, energy, bandwidth, and latency. This creates a fundamental tension between the growing complexity of FMs and the limited resources of deployment environments. This survey provides the first systematic characterization of adaptive, resource-efficient agentic AI systems. We summarize enabling techniques into elastic inference, test-time adaptation, dynamic multimodal integration, and agentic AI applications, and identify open challenges in balancing accuracy-latency-communication trade-offs and sustaining robustness under distribution shifts. We further highlight future opportunities in algorithm-system co-design, cognitive adaptation, and collaborative edge deployment. By mapping FM structures, cognition, and hardware resources, this work establishes a unified perspective toward scalable, adaptive, and resource-efficient agentic AI. We believe this survey can help readers to understand the connections between enabling technologies while promoting further discussions on the fusion of agentic intelligence and intelligent agents.

Ruifeng She, Bowen Pang, Kai Li et al.

As the artificial intelligence community advances into the era of large models with billions of parameters, distributed training and inference have become essential. While various parallelism strategies-data, model, sequence, and pipeline-have been successfully implemented for popular neural networks on main-stream hardware, optimizing the distributed deployment schedule requires extensive expertise and manual effort. Further more, while existing frameworks with most simple chain-like structures, they struggle with complex non-linear architectures. Mixture-of-experts and multi-modal models feature intricate MIMO and branch-rich topologies that require fine-grained operator-level parallelization beyond the capabilities of existing frameworks. We propose formulating parallelism planning as a scheduling optimization problem using mixed-integer programming. We propose a bi-level solution framework balancing optimality with computational efficiency, automatically generating effective distributed plans that capture both the heterogeneous structure of modern neural networks and the underlying hardware constraints. In experiments comparing against expert-designed strategies like DeepSeek's DualPipe, our framework achieves comparable or superior performance, reducing computational bubbles by half under the same memory constraints. The framework's versatility extends beyond throughput optimization to incorporate hardware utilization maximization, memory capacity constraints, and other considerations or potential strategies. Such capabilities position our solution as both a valuable research tool for exploring optimal parallelization strategies and a practical industrial solution for large-scale AI deployment.

Zichao Dong, Bowen Pang, Xufeng Huang et al.

Multi-modality fusion is proven an effective method for 3d perception for autonomous driving. However, most current multi-modality fusion pipelines for LiDAR semantic segmentation have complicated fusion mechanisms. Point painting is a quite straight forward method which directly bind LiDAR points with visual information. Unfortunately, previous point painting like methods suffer from projection error between camera and LiDAR. In our experiments, we find that this projection error is the devil in point painting. As a result of that, we propose a depth aware point painting mechanism, which significantly boosts the multi-modality fusion. Apart from that, we take a deeper look at the desired visual feature for LiDAR to operate semantic segmentation. By Lifting Visual Information as Cue, LVIC ranks 1st on nuScenes LiDAR semantic segmentation benchmark. Our experiments show the robustness and effectiveness. Codes would be make publicly available soon.