Lei Gong

Yixuan Zhu, Bo Zhang, Yinkang Gao et al.

In real-time systems, both individual task execution and data propagation must meet strict timing constraints. Cause-effect (CE) chains are widely used to analyze such behaviors by end-to-end latency. However, timing anomalies (TAs) can distort it, where a local reduction in execution times leads to an increase in the overall end-to-end latency. As a result, precisely analyzing the upper bounds of the latency becomes challenging, and such systems typically exhibit larger upper bounds than TA-eliminated systems. Existing studies either eliminate TAs by completely sacrificing average latency to simplify analysis or, despite adopting complex safe analysis methods, do not eliminate TAs effectively, still having high latencies. To address this issue, we identify two basic causes of TAs in end-to-end latency. Based on these causes, we propose the first treatment that eliminates TAs in the latency with negligible average latency loss using Deterministic Data Flow (DDF). We further formally prove its TA-free property. Therefore, we can get a precise upper bound for latency when all jobs execute with their worst-case execution times. Experimental results show that it effectively reduces the maximum end-to-end latency, the average latency, and latency jitter compared with the state-of-the-art (SOTA) method.

Zihan Wang, Cheng Tang, Lei Gong et al.

Chain-of-Thought (CoT) reasoning in large language models (LLMs) significantly improves accuracy on complex tasks, yet incurs excessive memory overhead due to the long think-stage sequences stored in the Key-Value (KV) cache. Unlike traditional generation tasks where all tokens are uniformly important, CoT emphasizes the final answer, rendering conventional KV compression strategies ineffective. In this paper, we present Crystal-KV, an efficient KV cache management framework tailored for CoT reasoning. Our key insight is the answer-first principle. By mapping answer preferences into think-stage attention map, we distinguish between SlipKV, which mainly maintains the reasoning flow but may occasionally introduce misleading context, and CrystalKV, which truly contributes to the correctness of the final answer. Next, we propose an attention-based Least Recently Frequently Used algorithm. It precisely identifies when a SlipKV entry's utility expires and evicts it, retaining CrystalKV without disrupting reasoning flow. Finally, we introduce an adaptive cache budget allocation algorithm. Based on the dynamic proportion of CrystalKV, it estimates the importance of each layer/head and adjusts the KV cache budget during inference, amplifying critical components to improve budget utilization. Results show that Crystal-KV achieves state-of-the-art KV cache compression, significantly improves throughput, and enables faster response time, while maintaining, or even improving, answer accuracy for CoT reasoning.

Yuntao Dai, Hang Gu, Teng Wang et al.

Vision-Language-Action (VLA) models have emerged as a unified paradigm for robotic perception and control, enabling emergent generalization and long-horizon task execution. However, their deployment in dynamic, real-world environments is severely hin dered by high inference latency. While smooth robotic interaction requires control frequencies of 20 to 30 Hz, current VLA models typi cally operate at only 3-5 Hz on edge devices due to the memory bound nature of autoregressive decoding. Existing optimizations often require extensive retraining or compromise model accuracy. To bridge this gap, we introduce ActionFlow, a system-level inference framework tailored for resource-constrained edge plat forms. At the core of ActionFlow is a Cross-Request Pipelin ing strategy, a novel scheduler that redefines VLA inference as a macro-pipeline of micro-requests. The strategy intelligently batches memory-bound Decode phases with compute-bound Prefill phases across continuous time steps to maximize hardware utilization. Furthermore, to support this scheduling, we propose a Cross Request State Packed Forward operator and a Unified KV Ring Buffer, which fuse fragmented memory operations into efficient dense computations. Experimental results demonstrate that ActionFlow achieves a 2.55x improvement in FPS on the OpenVLA-7B model without retraining, enabling real-time dy namic manipulation on edge hardware. Our work is available at https://anonymous.4open.science/r/ActionFlow-1D47.

Lijun Zhou, Hongcheng Luo, Zhenxin Zhu et al.

This report presents a unified technical system addressing the two core capabilities of world models for autonomous driving: world representation and world generation. For world representation, we propose WorldRec, a feed-forward reconstruction architecture driven by sparse scene queries. WorldRec initializes structured queries in 3D space, leveraging them to aggregate cross-view, cross-temporal features, thereby naturally enforcing spatial consistency across frames and yielding compact yet high-fidelity 3D Gaussian scene representations. For world generation, we propose WorldGen, a two-stage training framework of bidirectional pretraining followed by causal fine-tuning through three progressive stages (Teacher Forcing, ODE distillation, and DMD), enabling high-quality online causal video generation in as few as 4 denoising steps. Building on both modules, we further introduce the JWM, which deeply integrates WorldRec and WorldGen to achieve synergistic gains in generation stability, cross-frame consistency, and visual fidelity, providing a solid foundation for closed-loop simulation, data synthesis, and end-to-end training in autonomous driving.

Hang Zhang, Yang Xu, Lei Gong et al.

This paper introduces a new framework for clustering in a distributed network called Distributed Clustering based on Distributional Kernel (K) or KDC that produces the final clusters based on the similarity with respect to the distributions of initial clusters, as measured by K. It is the only framework that satisfies all three of the following properties. First, KDC guarantees that the combined clustering outcome from all sites is equivalent to the clustering outcome of its centralized counterpart from the combined dataset from all sites. Second, the maximum runtime cost of any site in distributed mode is smaller than the runtime cost in centralized mode. Third, it is designed to discover clusters of arbitrary shapes, sizes and densities. To the best of our knowledge, this is the first distributed clustering framework that employs a distributional kernel. The distribution-based clustering leads directly to significantly better clustering outcomes than existing methods of distributed clustering. In addition, we introduce a new clustering algorithm called Kernel Bounded Cluster Cores, which is the best clustering algorithm applied to KDC among existing clustering algorithms. We also show that KDC is a generic framework that enables a quadratic time clustering algorithm to deal with large datasets that would otherwise be impossible.

Kefan Song, Amir Moeini, Peng Wang et al.

Reinforcement learning (RL) is a human-designed framework for solving sequential decision making problems. In this work, we demonstrate that, surprisingly, RL emerges in LLM's (Large Language Model) inference time -- a phenomenon known as in-context RL (ICRL). Specifically, we propose a novel multi-round prompting framework called ICRL prompting. The goal is to prompt the LLM to complete a task. After the LLM generates a response at the current round, we give numerical scalar feedbacks for the response, called the rewards. At the next round, we prompt the LLM again with the same task and a context consisting of all previous responses and rewards. We observe that the quality of the LLM's response increases as the context grows. In other words, the LLM is able to maximize the scalar reward signal in the inference time, just like an RL algorithm. We evaluate ICRL prompting in three benchmarks (Game of 24, creative writing, and ScienceWorld) and demonstrate significant performance improvements over baseline methods such as Self-Refine and Reflexion. Surprisingly, in some experiments the reward signals are generated by the LLM itself, yet performance improvements are still observed from ICRL prompting, offering a promising paradigm for scaling test-time compute.

Yutong Chen, Zhang Wen, Chao Wang et al.

Hazy images degrade visual quality, and dehazing is a crucial prerequisite for subsequent processing tasks. Most current dehazing methods rely on neural networks and face challenges such as high computational parameter pressure and weak generalization capabilities. This paper introduces PriorNet--a novel, lightweight, and highly applicable dehazing network designed to significantly improve the clarity and visual quality of hazy images while avoiding excessive detail extraction issues. The core of PriorNet is the original Multi-Dimensional Interactive Attention (MIA) mechanism, which effectively captures a wide range of haze characteristics, substantially reducing the computational load and generalization difficulties associated with complex systems. By utilizing a uniform convolutional kernel size and incorporating skip connections, we have streamlined the feature extraction process. Simplifying the number of layers and architecture not only enhances dehazing efficiency but also facilitates easier deployment on edge devices. Extensive testing across multiple datasets has demonstrated PriorNet's exceptional performance in dehazing and clarity restoration, maintaining image detail and color fidelity in single-image dehazing tasks. Notably, with a model size of just 18Kb, PriorNet showcases superior dehazing generalization capabilities compared to other methods. Our research makes a significant contribution to advancing image dehazing technology, providing new perspectives and tools for the field and related domains, particularly emphasizing the importance of improving universality and deployability.

Chao Wang, Qi Yu, Lei Gong et al.

As the emerging field of machine learning, deep learning shows excellent ability in solving complex learning problems. However, the size of the networks becomes increasingly large scale due to the demands of the practical applications, which poses significant challenge to construct a high performance implementations of deep learning neural networks. In order to improve the performance as well to maintain the low power cost, in this paper we design DLAU, which is a scalable accelerator architecture for large-scale deep learning networks using FPGA as the hardware prototype. The DLAU accelerator employs three pipelined processing units to improve the throughput and utilizes tile techniques to explore locality for deep learning applications. Experimental results on the state-of-the-art Xilinx FPGA board demonstrate that the DLAU accelerator is able to achieve up to 36.1x speedup comparing to the Intel Core2 processors, with the power consumption at 234mW.