arXiv subjects

Wenhong Tian

Publications and source records attributed to Wenhong Tian.

At least 19 recordsLinked to original sources

arXiv2026

ST$^2$U: Stateful Test-Time Unlearning via Restricted Knowledge Boundary Control

Controlling restricted knowledge in large language models is essential for model alignment and safe deployment. Test-time unlearning avoids costly retraining and parameter updates by intervening only during inference. However, existing activation-editing methods apply isolated pointwise corrections, overlooking how autoregressive generation continually reconstructs hidden states from the prompt, cache, and generated prefix. Consequently, later states may return to restricted knowledge regions after a locally successful correction, causing restricted knowledge re-entry. In this work, we propose Stateful Test-Time Unlearning via restricted knowledge boundary control (ST$^2$U), which formulates test-time unlearning as trajectory-wide boundary control. ST$^2$U first models restricted knowledge boundaries in low-dimensional invertible coordinates while leaving orthogonal non-target components unchanged. During inference, ST$^2$U monitors risk along the trajectory, applies minimal boundary corrections with contextual anchoring, and propagates historical correction states across tokens to mitigate knowledge re-entry. This trajectory-wide control enables more persistent forgetting while preserving non-target capabilities and limiting inference overhead. Across three benchmarks and three model families, ST$^2$U delivers the strongest overall balance, combining best or second-best retention with competitive forgetting and substantially less restricted-knowledge re-entry than test-time baselines (13.76%-19.84% versus 46.50%-59.10%).

cs.LG↗

arXiv2026

AdapShot: Adaptive Many-Shot In-Context Learning with Semantic-Aware KV Cache Reuse

Many-Shot In-Context Learning (ICL) has emerged as a promising paradigm, leveraging extensive examples to unlock the reasoning potential of Large Language Models (LLMs). However, existing methods typically rely on a predetermined, fixed number of shots. This static approach often fails to adapt to the varying difficulty of different queries, leading to either insufficient context or interference from noise. Furthermore, the prohibitive computational and memory costs of long contexts severely limit Many-Shot's feasibility. To address the above limitations, we propose AdapShot, which dynamically optimizes shot counts and leverages KV cache reuse for efficient inference. Specifically, we design a probe-based evaluation mechanism that utilizes output entropy to determine the optimal number of shots. To bypass the redundant prefilling computation during both the probing and inference phases, we incorporate a semantics-aware KV cache reuse strategy. Within this reuse strategy, to address positional encoding incompatibilities, we introduce a decoupling and re-encoding method that enables the flexible reordering of cached key-value pairs. Extensive experiments demonstrate that AdapShot achieves an average performance gain of around 10% and a 4.64x speedup compared to state-of-the-art DBSA.

cs.AI↗

arXiv2026

CAP: Controllable Alignment Prompting for Unlearning in LLMs

Large language models (LLMs) trained on unfiltered corpora inherently risk retaining sensitive information, necessitating selective knowledge unlearning for regulatory compliance and ethical safety. However, existing parameter-modifying methods face fundamental limitations: high computational costs, uncontrollable forgetting boundaries, and strict dependency on model weight access. These constraints render them impractical for closed-source models, yet current non-invasive alternatives remain unsystematic and reliant on empirical experience. To address these challenges, we propose the Controllable Alignment Prompting for Unlearning (CAP) framework, an end-to-end prompt-driven unlearning paradigm. CAP decouples unlearning into a learnable prompt optimization process via reinforcement learning, where a prompt generator collaborates with the LLM to suppress target knowledge while preserving general capabilities selectively. This approach enables reversible knowledge restoration through prompt revocation. Extensive experiments demonstrate that CAP achieves precise, controllable unlearning without updating model parameters, establishing a dynamic alignment mechanism that overcomes the transferability limitations of prior methods.

cs.LG↗

arXiv2026

Cloud-native and Distributed Systems for Efficient and Scalable Large Language Models -- A Research Agenda

The rapid rise of Large Language Models (LLMs) has revolutionized various artificial intelligence (AI) applications, from natural language processing to code generation. However, the computational demands of these models, particularly in training and inference, present significant challenges. Traditional systems are often unable to meet these requirements, necessitating the integration of cloud-native and distributed architectures. This paper explores the role of cloud platforms and distributed systems in supporting the scalability, efficiency, and optimization of LLMs. We discuss the complexities of LLM deployment, including data management, resource optimization, and the need for microservices, autoscaling, and hybrid cloud-edge solutions. Additionally, we examine emerging research trends, such as serverless inference, quantum computing, and federated learning, and their potential to drive the next phase of LLM innovation. The paper concludes with a roadmap for future developments, emphasizing the need for continued research, standardization, and cross-sector collaboration to sustain the growth of LLMs in both research and enterprise applications.

cs.DC↗

arXiv2026

ALTER: Asymmetric LoRA for Token-Entropy-Guided Unlearning of LLMs

Large language models (LLMs) have advanced to encompass extensive knowledge across diverse domains. Yet controlling what a LLMs should not know is important for ensuring alignment and thus safe use. However, effective unlearning in LLMs is difficult due to the fuzzy boundary between knowledge retention and forgetting. This challenge is exacerbated by entangled parameter spaces from continuous multi-domain training, often resulting in collateral damage, especially under aggressive unlearning strategies. Furthermore, the computational overhead required to optimize State-of-the-Art (SOTA) models with billions of parameters poses an additional barrier. In this work, we present ALTER, a lightweight unlearning framework for LLMs to address both the challenges of knowledge entanglement and unlearning efficiency. ALTER operates through two phases: (I) high entropy tokens are captured and learned via the shared A matrix in LoRA, followed by (II) an asymmetric LoRA architecture that achieves a specified forgetting objective by parameter isolation and unlearning tokens within the target subdomains. Serving as a new research direction for achieving unlearning via token-level isolation in the asymmetric framework. ALTER achieves SOTA performance on TOFU, WMDP, and MUSE benchmarks with over 95% forget quality and shows minimal side effects through preserving foundational tokens. By decoupling unlearning from LLMs' billion-scale parameters, this framework delivers excellent efficiency while preserving over 90% of model utility, exceeding baseline preservation rates of 47.8-83.6%.

cs.CL↗

arXiv2025

Noise-Robustness Through Noise: A Framework combining Asymmetric LoRA with Poisoning MoE

Current parameter-efficient fine-tuning methods for adapting pre-trained language models to downstream tasks are susceptible to interference from noisy data. Conventional noise-handling approaches either rely on laborious data pre-processing or employ model architecture modifications prone to error accumulation. In contrast to existing noise-process paradigms, we propose a noise-robust adaptation method via asymmetric LoRA poisoning experts (LoPE), a novel framework that enhances model robustness to noise only with generated noisy data. Drawing inspiration from the mixture-of-experts architecture, LoPE strategically integrates a dedicated poisoning expert in an asymmetric LoRA configuration. Through a two-stage paradigm, LoPE performs noise injection on the poisoning expert during fine-tuning to enhance its noise discrimination and processing ability. During inference, we selectively mask the dedicated poisoning expert to leverage purified knowledge acquired by normal experts for noise-robust output. Extensive experiments demonstrate that LoPE achieves strong performance and robustness purely through the low-cost noise injection, which completely eliminates the requirement of data cleaning.

cs.LG↗

arXiv2025

HASH-RAG: Bridging Deep Hashing with Retriever for Efficient, Fine Retrieval and Augmented Generation

Retrieval-Augmented Generation (RAG) encounters efficiency challenges when scaling to massive knowledge bases while preserving contextual relevance. We propose Hash-RAG, a framework that integrates deep hashing techniques with systematic optimizations to address these limitations. Our queries directly learn binary hash codes from knowledgebase code, eliminating intermediate feature extraction steps, and significantly reducing storage and computational overhead. Building upon this hash-based efficient retrieval framework, we establish the foundation for fine-grained chunking. Consequently, we design a Prompt-Guided Chunk-to-Context (PGCC) module that leverages retrieved hash-indexed propositions and their original document segments through prompt engineering to enhance the LLM's contextual awareness. Experimental evaluations on NQ, TriviaQA, and HotpotQA datasets demonstrate that our approach achieves a 90% reduction in retrieval time compared to conventional methods while maintaining considerate recall performance. Additionally, The proposed system outperforms retrieval/non-retrieval baselines by 1.4-4.3% in EM scores.

cs.IR↗

arXiv2025

Accelerating Adaptive Retrieval Augmented Generation via Instruction-Driven Representation Reduction of Retrieval Overlaps

Retrieval-augmented generation (RAG) has emerged as a pivotal method for expanding the knowledge of large language models. To handle complex queries more effectively, researchers developed Adaptive-RAG (A-RAG) to enhance the generated quality through multiple interactions with external knowledge bases. Despite its effectiveness, A-RAG exacerbates the pre-existing efficiency challenges inherent in RAG, which are attributable to its reliance on multiple iterations of generation. Existing A-RAG approaches process all retrieved contents from scratch. However, they ignore the situation where there is a significant overlap in the content of the retrieval results across rounds. The overlapping content is redundantly represented, which leads to a large proportion of repeated computations, thus affecting the overall efficiency. To address this issue, this paper introduces a model-agnostic approach that can be generally applied to A-RAG methods, which is dedicated to reducing the redundant representation process caused by the overlapping of retrieval results. Specifically, we use cache access and parallel generation to speed up the prefilling and decoding stages respectively. Additionally, we also propose an instruction-driven module to further guide the model to more effectively attend to each part of the content in a more suitable way for LLMs. Experiments show that our approach achieves 2.79 and 2.33 times significant acceleration on average for prefilling and decoding respectively while maintaining equal generation quality.

cs.AI↗

arXiv2024

Deep Reinforcement Learning-based Methods for Resource Scheduling in Cloud Computing: A Review and Future Directions

As the quantity and complexity of information processed by software systems increase, large-scale software systems have an increasing requirement for high-performance distributed computing systems. With the acceleration of the Internet in Web 2.0, Cloud computing as a paradigm to provide dynamic, uncertain and elastic services has shown superiorities to meet the computing needs dynamically. Without an appropriate scheduling approach, extensive Cloud computing may cause high energy consumptions and high cost, in addition that high energy consumption will cause massive carbon dioxide emissions. Moreover, inappropriate scheduling will reduce the service life of physical devices as well as increase response time to users' request. Hence, efficient scheduling of resource or optimal allocation of request, that usually a NP-hard problem, is one of the prominent issues in emerging trends of Cloud computing. Focusing on improving quality of service (QoS), reducing cost and abating contamination, researchers have conducted extensive work on resource scheduling problems of Cloud computing over years. Nevertheless, growing complexity of Cloud computing, that the super-massive distributed system, is limiting the application of scheduling approaches. Machine learning, a utility method to tackle problems in complex scenes, is used to resolve the resource scheduling of Cloud computing as an innovative idea in recent years. Deep reinforcement learning (DRL), a combination of deep learning (DL) and reinforcement learning (RL), is one branch of the machine learning and has a considerable prospect in resource scheduling of Cloud computing. This paper surveys the methods of resource scheduling with focus on DRL-based scheduling approaches in Cloud computing, also reviews the application of DRL as well as discusses challenges and future directions of DRL in scheduling of Cloud computing.

cs.DC↗

arXiv2024

Lossless Acceleration of Large Language Model via Adaptive N-gram Parallel Decoding

While Large Language Models (LLMs) have shown remarkable abilities, they are hindered by significant resource consumption and considerable latency due to autoregressive processing. In this study, we introduce Adaptive N-gram Parallel Decoding (ANPD), an innovative and lossless approach that accelerates inference by allowing the simultaneous generation of multiple tokens. ANPD incorporates a two-stage approach: it begins with a rapid drafting phase that employs an N-gram module, which adapts based on the current interactive context, followed by a verification phase, during which the original LLM assesses and confirms the proposed tokens. Consequently, ANPD preserves the integrity of the LLM's original output while enhancing processing speed. We further leverage a multi-level architecture for the N-gram module to enhance the precision of the initial draft, consequently reducing inference latency. ANPD eliminates the need for retraining or extra GPU memory, making it an efficient and plug-and-play enhancement. In our experiments, models such as LLaMA and its fine-tuned variants have shown speed improvements up to 3.67x, validating the effectiveness of our proposed ANPD.

cs.CL↗

arXiv2024

PMFSNet: Polarized Multi-scale Feature Self-attention Network For Lightweight Medical Image Segmentation

Current state-of-the-art medical image segmentation methods prioritize accuracy but often at the expense of increased computational demands and larger model sizes. Applying these large-scale models to the relatively limited scale of medical image datasets tends to induce redundant computation, complicating the process without the necessary benefits. This approach not only adds complexity but also presents challenges for the integration and deployment of lightweight models on edge devices. For instance, recent transformer-based models have excelled in 2D and 3D medical image segmentation due to their extensive receptive fields and high parameter count. However, their effectiveness comes with a risk of overfitting when applied to small datasets and often neglects the vital inductive biases of Convolutional Neural Networks (CNNs), essential for local feature representation. In this work, we propose PMFSNet, a novel medical imaging segmentation model that effectively balances global and local feature processing while avoiding the computational redundancy typical in larger models. PMFSNet streamlines the UNet-based hierarchical structure and simplifies the self-attention mechanism's computational complexity, making it suitable for lightweight applications. It incorporates a plug-and-play PMFS block, a multi-scale feature enhancement module based on attention mechanisms, to capture long-term dependencies. Extensive comprehensive results demonstrate that even with a model (less than 1 million parameters), our method achieves superior performance in various segmentation tasks across different data scales. It achieves (IoU) metrics of 84.68%, 82.02%, and 78.82% on public datasets of teeth CT (CBCT), ovarian tumors ultrasound(MMOTU), and skin lesions dermoscopy images (ISIC 2018), respectively. The source code is available at https://github.com/yykzjh/PMFSNet.

cs.CV↗

arXiv2021

Prepartition: Load Balancing Approach for Virtual Machine Reservations in a Cloud Data Center

Load balancing is vital for the efficient and long-term operation of cloud data centers. With virtualization, post (reactive) migration of virtual machines after allocation is the traditional way for load balancing and consolidation. However, reactive migration is not easy to obtain predefined load balance objectives and may interrupt services and bring instability. Therefore, we provide a new approach, called Prepartition, for load balancing. It partitions a VM request into a few sub-requests sequentially with start time, end time and capacity demands, and treats each sub-request as a regular VM request. In this way, it can proactively set a bound for each VM request on each physical machine and makes the scheduler get ready before VM migration to obtain the predefined load balancing goal, which supports the resource allocation in a fine-grained manner. Simulations with real-world trace and synthetic data show that Prepartition for offline (PrepartitionOff) scheduling has 10%-20% better performance than the existing load balancing algorithms under several metrics, including average utilization, imbalance degree, makespan and Capacity_makespan. We also extend Prepartition to online load balancing. Evaluation results show that our proposed approach also outperforms existing online algorithms.

cs.DC↗

arXiv2021

Uncover Common Facial Expressions in Terracotta Warriors: A Deep Learning Approach

Can advanced deep learning technologies be applied to analyze some ancient humanistic arts? Can deep learning technologies be directly applied to special scenes such as facial expression analysis of Terracotta Warriors? The big challenging is that the facial features of the Terracotta Warriors are very different from today's people. We found that it is very poor to directly use the models that have been trained on other classic facial expression datasets to analyze the facial expressions of the Terracotta Warriors. At the same time, the lack of public high-quality facial expression data of the Terracotta Warriors also limits the use of deep learning technologies. Therefore, we firstly use Generative Adversarial Networks (GANs) to generate enough high-quality facial expression data for subsequent training and recognition. We also verify the effectiveness of this approach. For the first time, this paper uses deep learning technologies to find common facial expressions of general and postured Terracotta Warriors. These results will provide an updated technical means for the research of art of the Terracotta Warriors and shine lights on the research of other ancient arts.

cs.CV↗

arXiv2021

Machine Learning (ML)-Centric Resource Management in Cloud Computing: A Review and Future Directions

Cloud computing has rapidly emerged as model for delivering Internet-based utility computing services. In cloud computing, Infrastructure as a Service (IaaS) is one of the most important and rapidly growing fields. Cloud providers provide users/machines resources such as virtual machines, raw (block) storage, firewalls, load balancers, and network devices in this service model. One of the most important aspects of cloud computing for IaaS is resource management. Scalability, quality of service, optimum utility, reduced overheads, increased throughput, reduced latency, specialised environment, cost effectiveness, and a streamlined interface are some of the advantages of resource management for IaaS in cloud computing. Traditionally, resource management has been done through static policies, which impose certain limitations in various dynamic scenarios, prompting cloud service providers to adopt data-driven, machine-learning-based approaches. Machine learning is being used to handle a variety of resource management tasks, including workload estimation, task scheduling, VM consolidation, resource optimization, and energy optimization, among others. This paper provides a detailed review of challenges in ML-based resource management in current research, as well as current approaches to resolve these challenges, as well as their advantages and limitations. Finally, we propose potential future research directions based on identified challenges and limitations in current research.

cs.DC↗

arXiv2020

Energy Efficient Algorithms based on VM Consolidation for Cloud Computing: Comparisons and Evaluations

Cloud Computing paradigm has revolutionized IT industry and be able to offer computing as the fifth utility. With the pay-as-you-go model, cloud computing enables to offer the resources dynamically for customers anytime. Drawing the attention from both academia and industry, cloud computing is viewed as one of the backbones of the modern economy. However, the high energy consumption of cloud data centers contributes to high operational costs and carbon emission to the environment. Therefore, Green cloud computing is required to ensure energy efficiency and sustainability, which can be achieved via energy efficient techniques. One of the dominant approaches is to apply energy efficient algorithms to optimize resource usage and energy consumption. Currently, various virtual machine consolidation-based energy efficient algorithms have been proposed to reduce the energy of cloud computing environment. However, most of them are not compared comprehensively under the same scenario, and their performance is not evaluated with the same experimental settings. This makes users hard to select the appropriate algorithm for their objectives. To provide insights for existing energy efficient algorithms and help researchers to choose the most suitable algorithm, in this paper, we compare several state-of-the-art energy efficient algorithms in depth from multiple perspectives, including architecture, modelling and metrics. In addition, we also implement and evaluate these algorithms with the same experimental settings in CloudSim toolkit. The experimental results show the performance comparison of these algorithms with comprehensive results. Finally, detailed discussions of these algorithms are provided.

cs.DC↗

arXiv2018

A New Approach for Resource Scheduling with Deep Reinforcement Learning

With the rapid development of deep learning, deep reinforcement learning (DRL) began to appear in the field of resource scheduling in recent years. Based on the previous research on DRL in the literature, we introduce online resource scheduling algorithm DeepRM2 and the offline resource scheduling algorithm DeepRM_Off. Compared with the state-of-the-art DRL algorithm DeepRM and heuristic algorithms, our proposed algorithms have faster convergence speed and better scheduling efficiency with regarding to average slowdown time, job completion time and rewards.

cs.AI↗

arXiv2018

An Efficient Method to Transform SAT problems to Binary Integer Linear Programming Problem

In computational complexity theory, a decision problem is NP-complete when it is both in NP and NP-hard. Although a solution to a NP-complete can be verified quickly, there is no known algorithm to solve it in polynomial time. There exists a method to reduce a SAT (Satifiability) problem to Subset Sum Problem (SSP) in the literature, however, it can only be applied to small or medium size problems. Our study is to find an efficient method to transform a SAT problem to a mixed integer linear programming problem in larger size. Observing the feature of variable-clauses constraints in SAT, we apply linear inequality model (LIM) to the problem and propose a method called LIMSAT. The new method can work efficiently for very large size problem with thousands of variables and clauses in SAT tested using up-to-date benchmarks.

cs.CC↗

arXiv2017

A Survey on Load Balancing Algorithms for VM Placement in Cloud Computing

The emergence of cloud computing based on virtualization technologies brings huge opportunities to host virtual resource at low cost without the need of owning any infrastructure. Virtualization technologies enable users to acquire, configure and be charged on pay-per-use basis. However, Cloud data centers mostly comprise heterogeneous commodity servers hosting multiple virtual machines (VMs) with potential various specifications and fluctuating resource usages, which may cause imbalanced resource utilization within servers that may lead to performance degradation and service level agreements (SLAs) violations. To achieve efficient scheduling, these challenges should be addressed and solved by using load balancing strategies, which have been proved to be NP-hard problem. From multiple perspectives, this work identifies the challenges and analyzes existing algorithms for allocating VMs to PMs in infrastructure Clouds, especially focuses on load balancing. A detailed classification targeting load balancing algorithms for VM placement in cloud data centers is investigated and the surveyed algorithms are classified according to the classification. The goal of this paper is to provide a comprehensive and comparative understanding of existing literature and aid researchers by providing an insight for potential future enhancements.

cs.DC↗