Identify Critical KV Cache in LLM Inference from an Output Perturbation Perspective

Large language models have revolutionized natural language processing but face significant challenges of high storage and runtime costs, due to the transformer architecture's reliance on self-attention, particularly the large Key-Value (KV) cache for long-sequence inference. Recent efforts to reduce KV cache size by pruning less critical entries based on attention weights remain empirical and lack formal grounding. This paper presents a formal study on identifying critical KV cache entries by analyzing attention output perturbation. Our analysis reveals that, beyond attention weights, the value states within KV entries and pretrained parameter matrices are also crucial. Based on this, we propose a perturbation-constrained selection algorithm that optimizes the worst-case output perturbation to identify critical entries. Evaluations on the Needle-in-a-Haystack test and Longbench benchmark show our algorithm enhances state-of-the-art cache eviction methods. Further empirical analysis confirms that our algorithm achieves lower output perturbations in over 92% attention heads in Llama model, thereby providing a significant improvement over existing methods.

Fine-grained Preference Optimization Improves Zero-shot Text-to-Speech

Integrating human feedback to align text-to-speech (TTS) system outputs with human preferences has proven to be an effective approach for enhancing the robustness of language model-based TTS systems. Current approaches primarily focus on using preference data annotated at the utterance level. However, frequent issues that affect the listening experience often only arise in specific segments of audio samples, while other segments are well-generated. In this study, we propose a fine-grained preference optimization approach (FPO) to enhance the robustness of TTS systems. FPO focuses on addressing localized issues in generated samples rather than uniformly optimizing the entire utterance. Specifically, we first analyze the types of issues in generated samples, categorize them into two groups, and propose a selective training loss strategy to optimize preferences based on fine-grained labels for each issue type. Experimental results show that FPO enhances the robustness of zero-shot TTS systems by effectively addressing local issues, significantly reducing the bad case ratio, and improving intelligibility. Furthermore, FPO exhibits superior data efficiency compared with baseline systems, achieving similar performance with fewer training samples.

FRAG: A Flexible Modular Framework for Retrieval-Augmented Generation based on Knowledge Graphs

To mitigate the hallucination and knowledge deficiency in large language models (LLMs), Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) has shown promising potential by utilizing KGs as external resource to enhance LLMs reasoning.However, existing KG-RAG approaches struggle with a trade-off between flexibility and retrieval quality.Modular methods prioritize flexibility by avoiding the use of KG-fine-tuned models during retrieval, leading to fixed retrieval strategies and suboptimal retrieval quality.Conversely, coupled methods embed KG information within models to improve retrieval quality, but at the expense of flexibility.In this paper, we propose a novel flexible modular KG-RAG framework, termed FRAG, which synergizes the advantages of both approaches.FRAG estimates the hop range of reasoning paths based solely on the query and classify it as either simple or complex.To match the complexity of the query, tailored pipelines are applied to ensure efficient and accurate reasoning path retrieval, thus fostering the final reasoning process.By using the query text instead of the KG to infer the structural information of reasoning paths and employing adaptable retrieval strategies, FRAG improves retrieval quality while maintaining flexibility.Moreover, FRAG does not require extra LLMs fine-tuning or calls, significantly boosting efficiency and conserving resources.Extensive experiments show that FRAG achieves state-of-the-art performance with high efficiency and low resource consumption.

3rd Workshop on Maritime Computer Vision (MaCVi) 2025: Challenge Results

The 3rd Workshop on Maritime Computer Vision (MaCVi) 2025 addresses maritime computer vision for Unmanned Surface Vehicles (USV) and underwater. This report offers a comprehensive overview of the findings from the challenges. We provide both statistical and qualitative analyses, evaluating trends from over 700 submissions. All datasets, evaluation code, and the leaderboard are available to the public at https://macvi.org/workshop/macvi25.

PCDreamer: Point Cloud Completion Through Multi-view Diffusion Priors

This paper presents PCDreamer, a novel method for point cloud completion. Traditional methods typically extract features from partial point clouds to predict missing regions, but the large solution space often leads to unsatisfactory results. More recent approaches have started to use images as extra guidance, effectively improving performance, but obtaining paired data of images and partial point clouds is challenging in practice. To overcome these limitations, we harness the relatively view-consistent multi-view diffusion priors within large models, to generate novel views of the desired shape. The resulting image set encodes both global and local shape cues, which is especially beneficial for shape completion. To fully exploit the priors, we have designed a shape fusion module for producing an initial complete shape from multi-modality input (\ie, images and point clouds), and a follow-up shape consolidation module to obtain the final complete shape by discarding unreliable points introduced by the inconsistency from diffusion priors. Extensive experimental results demonstrate our superior performance, especially in recovering fine details.

Takin: A Cohort of Superior Quality Zero-shot Speech Generation Models

With the advent of the big data and large language model era, zero-shot personalized rapid customization has emerged as a significant trend. In this report, we introduce Takin AudioLLM, a series of techniques and models, mainly including Takin TTS, Takin VC, and Takin Morphing, specifically designed for audiobook production. These models are capable of zero-shot speech production, generating high-quality speech that is nearly indistinguishable from real human speech and facilitating individuals to customize the speech content according to their own needs. Specifically, we first introduce Takin TTS, a neural codec language model that builds upon an enhanced neural speech codec and a multi-task training framework, capable of generating high-fidelity natural speech in a zero-shot way. For Takin VC, we advocate an effective content and timbre joint modeling approach to improve the speaker similarity, while advocating for a conditional flow matching based decoder to further enhance its naturalness and expressiveness. Last, we propose the Takin Morphing system with highly decoupled and advanced timbre and prosody modeling approaches, which enables individuals to customize speech production with their preferred timbre and prosody in a precise and controllable manner. Extensive experiments validate the effectiveness and robustness of our Takin AudioLLM series models. For detailed demos, please refer to https://takinaudiollm.github.io.

Innovative RIS Prototyping Enhancing Wireless Communication with Real-Time Spot Beam Tracking and OAM Wavefront Manipulation

This paper presents a sophisticated reconfigurable metasurface architecture that introduces an advanced concept of flexible full-array space-time wavefront manipulation with enhanced dynamic capabilities. The practical 2-bit phase-shifting unit cell on the RIS is distinguished by its ability to maintain four stable phase states, each with ${90^ \circ }$ differences, and features an insertion loss of less than 0.6 dB across a bandwidth of 200 MHz. All reconfigurable units are equipped with meticulously designed control circuits, governed by an intelligent core composed of multiple Micro-Controller Units (MCUs), enabling rapid control response across the entire RIS array. Owing to the capability of each unit cell on the metasurface to independently switch states, the entire RIS is not limited to controlling general beams with specific directional patterns, but also generates beams with more complex structures, including multi-focus 3D spot beams and vortex beams. This development substantially broadens its applicability across various industrial wireless transmission scenarios. Moreover, by leveraging the rapid-respond space-time coding and the full-array independent programmability of the RIS prototyping operating at 10.7 GHz, we have demonstrated that: 1) The implementation of 3D spot beams scanning facilitates dynamic beam tracking and real-time communication under the indoor near-field scenario; 2) The rapid wavefront rotation of vortex beams enables precise modulation of signals within the Doppler domain, showcasing an innovative approach to wireless signal manipulation; 3) The beam steering experiments for blocking users under outdoor far-field scenarios, verifying the beamforming capability of the RIS board.

Optimizing KV Cache Eviction in LLMs: Adaptive Allocation for Enhanced Budget Utilization

Large Language Models have excelled in various fields but encounter efficiency limitations due to the extensive KV cache required for long sequences inference. Many efforts try to evict non-critical cache elements during runtime, thereby reducing cache size within a given memory budget while preserving generation quality. Our reexamination of their underlying principles discerns that prevailing strategies essentially aim to minimize an upper bound of eviction loss within a specific budget allocation. However, we observe that the current practice of uniformly allocating budgets across different attention heads during the eviction procedure tends to degrade the quality of generation posten-eviction. In light of these findings, we propose a simple yet effective adaptive allocation algorithm that not only theoretically ensures its loss upper bound does not exceed that of previous uniform allocation methods, but also effectively aligns with the characteristics of the self-attention mechanism, thus practically reducing the upper bound. Further, integrating this algorithm with two of the most advanced methods yields Ada-SnapKV and Ada-Pyramid. Extensive experimental validation across 16 datasets and the Needle-in-a-Haystack test confirm that Ada-SnapKV and Ada-Pyramid achieve further enhancements, establishing new benchmarks in state-of-the-art performance.

Environment Sensing-aided Beam Prediction with Transfer Learning for Smart Factory

In this paper, we propose an environment sensing-aided beam prediction model for smart factory that can be transferred from given environments to a new environment. In particular, we first design a pre-training model that predicts the optimal beam by sensing the present environmental information. When encountering a new environment, it generally requires collecting a large amount of new training data to retrain the model, whose cost severely impedes the application of the designed pre-training model. Therefore, we next design a transfer learning strategy that fine-tunes the pre-trained model by limited labeled data of the new environment. Simulation results show that when the pre-trained model is fine-tuned by 30\% of labeled data from the new environment, the Top-10 beam prediction accuracy reaches 94\%. Moreover, compared with the way to completely re-training the prediction model, the amount of training data and the time cost of the proposed transfer learning strategy reduce 70\% and 75\% respectively.

Driving Intelligent IoT Monitoring and Control through Cloud Computing and Machine Learning

This article explores how to drive intelligent iot monitoring and control through cloud computing and machine learning. As iot and the cloud continue to generate large and diverse amounts of data as sensor devices in the network, the collected data is sent to the cloud for statistical analysis, prediction, and data analysis to achieve business objectives. However, because the cloud computing model is limited by distance, it can be problematic in environments where the quality of the Internet connection is not ideal for critical operations. Therefore, edge computing, as a distributed computing architecture, moves the location of processing applications, data and services from the central node of the network to the logical edge node of the network to reduce the dependence on cloud processing and analysis of data, and achieve near-end data processing and analysis. The combination of iot and edge computing can reduce latency, improve efficiency, and enhance security, thereby driving the development of intelligent systems. The paper also introduces the development of iot monitoring and control technology, the application of edge computing in iot monitoring and control, and the role of machine learning in data analysis and fault detection. Finally, the application and effect of intelligent Internet of Things monitoring and control system in industry, agriculture, medical and other fields are demonstrated through practical cases and experimental studies.