arXiv subjects

Yinlong Xu

Publications and source records attributed to Yinlong Xu.

At least 19 recordsLinked to original sources

arXiv2026

A Sparse Glimpse of the Whole: Train-Free Self-Speculative Decoding

Speculative decoding alleviates the memory-bandwidth bottleneck in large language model inference, but its acceleration is jointly constrained by drafting overhead, token acceptance, and speculation length. We present a unified efficiency analysis showing that extending the speculation horizon can reduce rather than improve speedup when the marginal acceptance probability falls below the relative drafting cost. Guided by this analysis, we introduce SparseSpec-L, a training-free self-speculative decoding framework for long-context inference. SparseSpec-L generates lightweight drafts directly from the target model using a dynamically sparsified and recallable KV cache. It recycles per-head attention statistics produced during full-context verification as a no-extra-forward importance signal, allowing critical historical tokens to be recalled without permanently discarding the dense KV cache. An online entropy-based controller further selects the speculation length according to expected step-wise efficiency. Experiments across multiple long-context tasks and model scales show consistent end-to-end acceleration, with up to speedup over autoregressive decoding while preserving the target model's output distribution.

cs.CL↗

arXiv2026

Varuna: Enabling Failure-Type Aware RDMA Failover

RDMA link failures can render connections temporarily unavailable, causing both performance degradation and significant recovery overhead. To tolerate such failures, production datacenters assign each primary link with a standby link and, upon failure, uniformly retransmit all in-flight RDMA request over the backup path. However, we observe that such blanket retransmission is unnecessary. In-flight requests can be split into pre-failure and post-failure categories depending on whether the responder has already executed. Retransmitting post-failure requests is not only redundant (consuming bandwidth), but also incorrect for non-idempotent operations, where duplicate execution can violate application semantics. We present Varuna, a failure-type-aware RDMA recovery mechanism that enables correct retransmission and us-level failover. Varuna piggybacks a lightweight completion log on every RDMA operation; after a link failure, this log deterministically reveals which in-flight requests were executed (post-failure) and which were lost (pre-failure). Varuna then retransmits only the pre-failure subset and fetches/recovers the return values for post-failure requests. Evaluated using synthetic microbenchmarks and end-to-end RDMA TPC-C transactions, Varuna incurs only 0.6-10% steady-state latency overhead in realistic applications, eliminates 65% of recovery retransmission time, preserves transactional consistency, and introduces zero connectivity rebuild overhead and negligible memory overhead during RDMA failover.

cs.DC↗

arXiv2025

Reason from Future: Reverse Thought Chain Enhances LLM Reasoning

It has been demonstrated that carefully designed reasoning paradigms, like Chain-of-Thought (CoT) and Tree-of-Thought (ToT), can enhance the reasoning capabilities of small language models by detailed thinking and extensive thought searching, unbounded branching factors in the searching space create prohibitive reasoning consumption. However these methods fall into the trap of local optimum reasoning, which means the model lacks a global perspective while solving problems. We propose a novel reasoning paradigm called Reason from Future (RFF), which generates reasoning paths by bidirectional reasoning that combines top-down planning with bottom-up reasoning accumulation. The essence of RFF lies in its reverse reasoning mechanism, which prioritizes core logical relationships and imposes goal-oriented constraints on intermediate steps, thereby reducing the searching space and mitigating error accumulation inherent in sequential forward reasoning. Empirical evaluations across diverse experiments demonstrate that RFF outperforms conventional paradigms with higher accuracy and less searching space to solve complex tasks.

cs.AI↗

arXiv2025

FOCUS: Boosting Schema-aware Access for KV Stores via Hierarchical Data Management

Persistent key-value (KV) stores are critical infrastructure for data-intensive applications. Leveraging high-performance Non-Volatile Memory (NVM) to enhance KV stores has gained traction. However, previous work has primarily focused on optimizing KV stores themselves, without adequately addressing their integration into applications. Consequently, existing applications, represented by NewSQL databases, still resort to a flat mapping approach, which simply maps structured records into flat KV pairs to use KV stores. Such semantic mismatch may cause significant I/O amplification and I/O splitting under production workloads, harming the performance. To this end, we propose FOCUS, a log-structured KV store optimized for fine-grained hierarchical data organization and schema-aware access. FOCUS introduces a hierarchical KV model to provide native support for upper-layer structured data. We implemented FOCUS from scratch. Experiments show that FOCUS can increase throughput by 2.1-5.9x compared to mainstream NVM-backed KV stores under YCSB SQL workloads.

cs.DB↗

arXiv2025

ProtFlow: Fast Protein Sequence Design via Flow Matching on Compressed Protein Language Model Embeddings

The design of protein sequences with desired functionalities is a fundamental task in protein engineering. Deep generative methods, such as autoregressive models and diffusion models, have greatly accelerated the discovery of novel protein sequences. However, these methods mainly focus on local or shallow residual semantics and suffer from low inference efficiency, large modeling space and high training cost. To address these challenges, we introduce ProtFlow, a fast flow matching-based protein sequence design framework that operates on embeddings derived from semantically meaningful latent space of protein language models. By compressing and smoothing the latent space, ProtFlow enhances performance while training on limited computational resources. Leveraging reflow techniques, ProtFlow enables high-quality single-step sequence generation. Additionally, we develop a joint design pipeline for the design scene of multichain proteins. We evaluate ProtFlow across diverse protein design tasks, including general peptides and long-chain proteins, antimicrobial peptides, and antibodies. Experimental results demonstrate that ProtFlow outperforms task-specific methods in these applications, underscoring its potential and broad applicability in computational protein sequence design and analysis.

cs.LG↗

arXiv2024

Unraveling Babel: Exploring Multilingual Activation Patterns of LLMs and Their Applications

Recently, large language models (LLMs) have achieved tremendous breakthroughs in the field of NLP, but still lack understanding of their internal neuron activities when processing different languages. We designed a method to convert dense LLMs into fine-grained MoE architectures, and then visually studied the multilingual activation patterns of LLMs through expert activation frequency heatmaps. Through comprehensive experiments on different model families, different model sizes, and different variants, we analyzed the similarities and differences in the internal neuron activation patterns of LLMs when processing different languages. Specifically, we investigated the distribution of high-frequency activated experts, multilingual shared experts, whether multilingual activation patterns are related to language families, and the impact of instruction tuning on activation patterns. We further explored leveraging the discovered differences in expert activation frequencies to guide sparse activation and pruning. Experimental results demonstrated that our method significantly outperformed random expert pruning and even exceeded the performance of unpruned models in some languages. Additionally, we found that configuring different pruning rates for different layers based on activation level differences could achieve better results. Our findings reveal the multilingual processing mechanisms within LLMs and utilize these insights to offer new perspectives for applications such as sparse activation and model pruning.

cs.CL↗

arXiv2024

MambaCapsule: Towards Transparent Cardiac Disease Diagnosis with Electrocardiography Using Mamba Capsule Network

Cardiac arrhythmia, a condition characterized by irregular heartbeats, often serves as an early indication of various heart ailments. With the advent of deep learning, numerous innovative models have been introduced for diagnosing arrhythmias using Electrocardiogram (ECG) signals. However, recent studies solely focus on the performance of models, neglecting the interpretation of their results. This leads to a considerable lack of transparency, posing a significant risk in the actual diagnostic process. To solve this problem, this paper introduces MambaCapsule, a deep neural networks for ECG arrhythmias classification, which increases the explainability of the model while enhancing the accuracy.Our model utilizes Mamba for feature extraction and Capsule networks for prediction, providing not only a confidence score but also signal features. Akin to the processing mechanism of human brain, the model learns signal features and their relationship between them by reconstructing ECG signals in the predicted selection. The model evaluation was conducted on MIT-BIH and PTB dataset, following the AAMI standard. MambaCapsule has achieved a total accuracy of 99.54% and 99.59% on the test sets respectively. These results demonstrate the promising performance of under the standard test protocol.

cs.LG↗

arXiv2024

TWIN-GPT: Digital Twins for Clinical Trials via Large Language Model

Clinical trials are indispensable for medical research and the development of new treatments. However, clinical trials often involve thousands of participants and can span several years to complete, with a high probability of failure during the process. Recently, there has been a burgeoning interest in virtual clinical trials, which simulate real-world scenarios and hold the potential to significantly enhance patient safety, expedite development, reduce costs, and contribute to the broader scientific knowledge in healthcare. Existing research often focuses on leveraging electronic health records (EHRs) to support clinical trial outcome prediction. Yet, trained with limited clinical trial outcome data, existing approaches frequently struggle to perform accurate predictions. Some research has attempted to generate EHRs to augment model development but has fallen short in personalizing the generation for individual patient profiles. Recently, the emergence of large language models has illuminated new possibilities, as their embedded comprehensive clinical knowledge has proven beneficial in addressing medical issues. In this paper, we propose a large language model-based digital twin creation approach, called TWIN-GPT. TWIN-GPT can establish cross-dataset associations of medical information given limited data, generating unique personalized digital twins for different patients, thereby preserving individual patient characteristics. Comprehensive experiments show that using digital twins created by TWIN-GPT can boost the clinical trial outcome prediction, exceeding various previous prediction approaches.

cs.LG↗

arXiv2020

Deterministic Data Distribution for Efficient Recovery in Erasure-Coded Storage Systems

Due to individual unreliable commodity components, failures are common in large-scale distributed storage systems. Erasure codes are widely deployed in practical storage systems to provide fault tolerance with low storage overhead. However, random data distribution (RDD), commonly used in erasure-coded storage systems, induces heavy cross-rack traffic, load imbalance, and random access, which adversely affects failure recovery. In this paper, with orthogonal arrays, we define a Deterministic Data Distribution ($D^3$) to uniformly distribute data/parity blocks among nodes, and propose an efficient failure recovery approach based on $D^3$, which minimizes the cross-rack repair traffic against a single node failure. Thanks to the uniformity of $D^3$, the proposed recovery approach balances the repair traffic not only among nodes within a rack but also among racks. We implement $D^3$ over Reed-Solomon codes and Locally Repairable Codes in Hadoop Distributed File System (HDFS) with a cluster of 28 machines. Compared with RDD, our experiments show that $D^3$ significantly speeds up the failure recovery up to 2.49 times for RS codes and 1.38 times for LRCs. Moreover, $D^3$ supports front-end applications better than RDD in both of normal and recovery states.

cs.DC↗

arXiv2019

Enabling Efficient Updates in KV Storage via Hashing: Design and Performance Evaluation

Persistent key-value (KV) stores mostly build on the Log-Structured Merge (LSM) tree for high write performance, yet the LSM-tree suffers from the inherently high I/O amplification. KV separation mitigates I/O amplification by storing only keys in the LSM-tree and values in separate storage. However, the current KV separation design remains inefficient under update-intensive workloads due to its high garbage collection (GC) overhead in value storage. We propose HashKV, which aims for high update performance atop KV separation under update-intensive workloads. HashKV uses hash-based data grouping, which deterministically maps values to storage space so as to make both updates and GC efficient. We further relax the restriction of such deterministic mappings via simple but useful design extensions. We extensively evaluate various design aspects of HashKV. We show that HashKV achieves 4.6x update throughput and 53.4% less write traffic compared to the current KV separation design. In addition, we demonstrate that we can integrate the design of HashKV with state-of-the-art KV stores and improve their respective performance.

cs.DB↗

arXiv2017

Paid Peering, Settlement-Free Peering, or Both?

With the rapid growth of congestion-sensitive and data-intensive applications, traditional settlement-free peering agreements with best-effort delivery often do not meet the QoS requirements of content providers (CPs). Meanwhile, Internet access providers (IAPs) feel that revenues from end-users are not sufficient to recoup the upgrade costs of network infrastructures. Consequently, some IAPs have begun to offer CPs a new type of peering agreement, called paid peering, under which they provide CPs with better data delivery quality for a fee. In this paper, we model a network platform where an IAP makes decisions on the peering types offered to CPs and the prices charged to CPs and end-users. We study the optimal peering schemes for the IAP, i.e., to offer CPs both the paid and settlement-free peering to choose from or only one of them, as the objective is profit or welfare maximization. Our results show that 1) the IAP should always offer the paid and settlement-free peering under the profit-optimal and welfare-optimal schemes, respectively, 2) whether to simultaneously offer the other peering type is largely driven by the type of data traffic, e.g., text or video, and 3) regulators might want to encourage the IAP to allocate more network capacity to the settlement-free peering for increasing user welfare.

cs.NI↗

arXiv2017

The Role of Data Cap in Optimal Two-part Network Pricing

Internet services are traditionally priced at flat rates; however, many Internet service providers (ISPs) have recently shifted towards two-part tariffs where a data cap is imposed to restrain data demand from heavy users. Although the two-part tariff could generally increase the revenue for ISPs and has been supported by the US FCC, the role of data cap and its optimal pricing structures are not well understood. In this article, we study the impact of data cap on the optimal two-part pricing schemes for congestion-prone service markets. We model users' demand and preferences over pricing and congestion alternatives and derive the market share and congestion of service providers under a market equilibrium. Based on the equilibrium model, we characterize the two-part structures of the revenue- and welfare-optimal pricing schemes. Our results reveal that 1) the data cap provides a mechanism for ISPs to transition from the flat-rate to pay-as-you-go type of schemes, 2) both the revenue and welfare objectives of the ISP will drive the optimal pricing towards usage-based schemes with diminishing data caps, and 3) the welfare-optimal tariff comprises lower fees than the revenue-optimal counterpart, suggesting that regulators might want to promote usage-based pricing but regulate the lump-sum and per-unit fees.

cs.NI↗

arXiv2017

On Optimal Two-Sided Pricing of Congested Networks

Traditionally, Internet Access Providers (APs) only charge end-users for Internet access services; however, to recoup infrastructure costs and increase revenues, some APs have recently adopted two-sided pricing schemes under which both end-users and content providers are charged. Meanwhile, with the rapid growth of traffic, network congestion could seriously degrade user experiences and influence providers' utility. To optimize profit and social welfare, APs and regulators need to design appropriate pricing strategies and regulatory policies that take the effects of network congestion into consideration. In this paper, we model two-sided networks under which users' traffic demands are influenced by exogenous pricing and endogenous congestion parameters and derive the system congestion under an equilibrium. We characterize the structures and sensitivities of profit- and welfare-optimal two-sided pricing schemes and reveal that 1) the elasticity of system throughput plays a crucial role in determining the structures of optimal pricing, 2) the changes of optimal pricing under varying AP's capacity and users' congestion sensitivity are largely driven by the type of data traffic, e.g., text or video, and 3) APs and regulators will be incentivized to shift from one-sided to two-sided pricing when APs' capacities and user demand for video traffic grow. Our results can help APs design optimal two-sided pricing and guide regulators to legislate desirable policies.

cs.NI↗

arXiv2017

On Optimal Service Differentiation in Congested Network Markets

As Internet applications have become more diverse in recent years, users having heavy demand for online video services are more willing to pay higher prices for better services than light users that mainly use e-mails and instant messages. This encourages the Internet Service Providers (ISPs) to explore service differentiations so as to optimize their profits and allocation of network resources. Much prior work has focused on the viability of network service differentiation by comparing with the case of a single-class service. However, the optimal service differentiation for an ISP subject to resource constraints has remained unsolved. In this work, we establish an optimal control framework to derive the analytical solution to an ISP's optimal service differentiation, i.e. the optimal service qualities and associated prices. By analyzing the structures of the solution, we reveal how an ISP should adjust the service qualities and prices in order to meet varying capacity constraints and users' characteristics. We also obtain the conditions under which ISPs have strong incentives to implement service differentiation and whether regulators should encourage such practices.

cs.NI↗

arXiv2017

Measuring and Maximizing Influence via Random Walk in Social Activity Networks

With the popularity of OSNs, finding a set of most influential users (or nodes) so as to trigger the largest influence cascade is of significance. For example, companies may take advantage of the "word-of-mouth" effect to trigger a large cascade of purchases by offering free samples/discounts to those most influential users. This task is usually modeled as an influence maximization problem, and it has been widely studied in the past decade. However, considering that users in OSNs may participate in various kinds of online activities, e.g., giving ratings to products, joining discussion groups, etc., influence diffusion through online activities becomes even more significant. In this paper, we study the impact of online activities by formulating the influence maximization problem for social-activity networks (SANs) containing both users and online activities. To address the computation challenge, we define an influence centrality via random walks to measure influence, then use the Monte Carlo framework to efficiently estimate the centrality in SANs. Furthermore, we develop a greedy-based algorithm with two novel optimization techniques to find the most influential users. By conducting extensive experiments with real-world datasets, we show our approach is more efficient than the state-of-the-art algorithm IMM[17] when we needs to handle large amount of online activities.

cs.SI↗

arXiv2016

Stochastic Modeling of Hybrid Cache Systems

In recent years, there is an increasing demand of big memory systems so to perform large scale data analytics. Since DRAM memories are expensive, some researchers are suggesting to use other memory systems such as non-volatile memory (NVM) technology to build large-memory computing systems. However, whether the NVM technology can be a viable alternative (either economically and technically) to DRAM remains an open question. To answer this question, it is important to consider how to design a memory system from a "system perspective", that is, incorporating different performance characteristics and price ratios from hybrid memory devices. This paper presents an analytical model of a "hybrid page cache system" so to understand the diverse design space and performance impact of a hybrid cache system. We consider (1) various architectural choices, (2) design strategies, and (3) configuration of different memory devices. Using this model, we provide guidelines on how to design hybrid page cache to reach a good trade-off between high system throughput (in I/O per sec or IOPS) and fast cache reactivity which is defined by the time to fill the cache. We also show how one can configure the DRAM capacity and NVM capacity under a fixed budget. We pick PCM as an example for NVM and conduct numerical analysis. Our analysis indicates that incorporating PCM in a page cache system significantly improves the system performance, and it also shows larger benefit to allocate more PCM in page cache in some cases. Besides, for the common setting of performance-price ratio of PCM, "flat architecture" offers as a better choice, but "layered architecture" outperforms if PCM write performance can be significantly improved in the future.

cs.PF↗

arXiv2015

Sampling Online Social Networks via Heterogeneous Statistics

Most sampling techniques for online social networks (OSNs) are based on a particular sampling method on a single graph, which is referred to as a statistics. However, various realizing methods on different graphs could possibly be used in the same OSN, and they may lead to different sampling efficiencies, i.e., asymptotic variances. To utilize multiple statistics for accurate measurements, we formulate a mixture sampling problem, through which we construct a mixture unbiased estimator which minimizes asymptotic variance. Given fixed sampling budgets for different statistics, we derive the optimal weights to combine the individual estimators; given fixed total budget, we show that a greedy allocation towards the most efficient statistics is optimal. In practice, the sampling efficiencies of statistics can be quite different for various targets and are unknown before sampling. To solve this problem, we design a two-stage framework which adaptively spends a partial budget to test different statistics and allocates the remaining budget to the inferred best statistics. We show that our two-stage framework is a generalization of 1) randomly choosing a statistics and 2) evenly allocating the total budget among all available statistics, and our adaptive algorithm achieves higher efficiency than these benchmark strategies in theory and experiment.

cs.SI↗

arXiv2013

Coding based Data Broadcasting for Time Critical Applications with Rate Adaptation

In this paper, we dynamically select the transmission rate and design wireless network coding to improve the quality of services such as delay for time critical applications. In a network coded system, with low transmission rate and hence longer transmission range, more packets may be encoded, which increases the coding opportunity. However, low transmission rate may incur extra transmission delay, which is intolerable for time critical applications. We design a novel joint rate selection and wireless network coding (RSNC) scheme with delay constraint, so as to maximize the total benefit (where we can define the benefit based on the priority or importance of a packet for example) of the packets that are successfully received at the destinations without missing their deadlines. We prove that the proposed problem is NP-hard, and propose a novel graph model to mathematically formulate the problem. For the general case, we propose a transmission metric and design an efficient algorithm to determine the transmission rate and coding strategy for each transmission. For a special case when all delay constraints are the same, we study the pairwise coding and present a polynomial time pairwise coding algorithm that achieves an approximation ratio of 1 - 1/e to the optimal pairwise coding solution, where e is the base of the natural logarithm. Finally, simulation results demonstrate the superiority of the proposed RSNC scheme.

cs.IT↗