arXiv subjects

Yang

Publications and source records attributed to Yang.

At least 19 recordsLinked to original sources

arXiv2026

SAGE: SLO-Aware Adaptive Retrieval for Production RAG Systems

Retrieval-Augmented Generation (RAG) systems in production operate under strict service level objectives (SLOs) on tail latency and infrastructure cost. However, standard retrieval pipelines rely on fixed retrieval budgets that ignore query difficulty, over-retrieving for easy queries and under-serving hard ones, forcing operators to trade answer quality against SLO compliance. This paper proposes SAGE, a learned SLO-aware adaptive retrieval policy that dynamically selects the number of passages k per query. SAGE uses lightweight features derived from initial retrieval (e.g., score distributions, rank gaps, lexical signals) and is trained offline via imitation learning from an oracle that approximates optimal latency-quality trade-offs. At inference, it adds no LLM calls and minimal overhead. On Natural Questions, under a 5s P95 latency SLO, SAGE achieves 95% SLO compliance versus 30% for the best static baseline (k=20), reduces P95 latency by 36% and retrieval cost by 51% with only 2 percentage points Exact Match (EM) loss. A single policy trained on Natural Questions generalizes across HotpotQA, UnSeenTimeQA, and four LLM families (Llama, Qwen, Mistral, Gemma), consistently yielding +45-52 point SLO improvements without quality degradation.

cs.LG↗

arXiv2026

Unleashing the Potential of Large Language Models: A Blueprint for Real-Time, Enterprise-Ready Deployments

Large language models deployed in real-time, regulated settings face knowledge staleness, catastrophic forgetting, hallucination, and weak feedback loops. We present a unified, pattern-driven LLMOps architecture integrating real-time data ingestion, continual learning, retrieval-augmented generation (RAG), and human-in-the-loop feedback into a single operational pipeline. Four contributions map to established software design patterns: an adaptive ingestion pattern orchestrator (AIPO) evaluated with FreshStreamBench; STAR+FAR continual learning with sparse temporal adapter routing and freshness-aware replay; SAGE, an SLO-aware adaptive retrieval policy predicting a per-query passage budget to meet tail-latency targets; and an automated feedback-driven convergence stage with RLHF triggers. The result reduces latency-cost-accuracy trade-offs while supporting auditability and rollback for high-risk sectors such as health care and finance.

cs.LG↗

arXiv2026

ArcLP: A Matlab implementation of an $\mathcal{O}(\sqrt{n}L)$ arc-search infeasible interior-point algorithm for linear programming

This paper presents a Matlab implementation of an arc-search infeasible interior point algorithm for linear programming (LP), which has a proven polynomial bound of $\mathcal{O}(\sqrt{n}L)$, the best among all interior-point algorithms for LP. Software architecture and major functions are discussed. Its ease of use is described by a simple example. Crucial strategies are summarized. Quality of the software is assured because this software has been extensively tested on both PC and Linux for the widely used Netlib benchmark linear programming problems in standard form. Some benchmark test problems involve tens of thousands of constraints and hundreds of thousands of variables. For all tested problems, the code found the optimal solution. The numerical results have been compared to those obtained by the popular Mehrotra's predictor-corrector algorithm. We conclude that the implemented algorithm not only has the best polynomial bound but also is computationally competitive compared to the popular Mehrotra's predictor-corrector algorithm.

math.OC↗

arXiv2026

SLFS: a Flexible, Low-Cost Distributed File System Using Serverless Designs

Large-scale distributed file systems must provision resources for peak demand, yet file access patterns fluctuate significantly, leaving substantial capacity idle during off-peak periods. Existing scaling mechanisms operate at the granularity of entire servers and take minutes to hours, making them unable to track the rapid, fine-grained load variations that file systems commonly experience. Serverless computing, with its millisecond-granularity elasticity and pay-per-use pricing, offers a compelling alternative. We present SLFS, the first distributed file system built with serverless functions for both data and metadata operations. SLFS implements file services on top of key-value stores, keeping function operations simple and short, and introduces a novel multi-threaded, short-lived server design that overcomes the cold-start problem while maintaining low cost. A policy-enforcing coordinator efficiently maps files to function instances, scales the system elastically, and controls function lifetimes to balance performance and cost. SLFS can flexibly run on diverse storage backends -- from cloud-native services like S3 to user-managed key-value stores -- enabling configurable cost-performance trade-offs. Our evaluation shows that SLFS mitigates cold starts by 580$\times$ compared to the base serverless design and outperforms $\lambda$FS, EFS, and Ceph at up to 63%, 68%, and 63% lower cost, respectively.

cs.DC↗

arXiv2026

Characterizing Driver Interactions with Autonomous Vehicles via Response Maps

Understanding human responses to autonomous vehicle (AV) behaviors is essential for socially aware interaction, which is crucial for socially compatible navigation in shared traffic environments. We characterize human driving responses in interactions with AVs as feedback laws over the coupled state space of the human driven vehicle and the AV. We model the human driver's actions using a response map, a concept based in game theory, and employ a linear representation to capture driver behaviors as a function of AV behaviors, based on empirical data from a driving simulator study. Our results show that 1) human driver acceleration behavior can be captured using response maps, and 2) human driver responses differ significantly with respect to AV behaviors of yielding, non-yielding, and responsive to the human driver.

eess.SY↗

arXiv2026

Wick Renormalized Parabolic Stochastic Quantization Equations on Rough Metric Measure Spaces

On metric measure spaces with sub-Gaussian heat kernel behavior in small time, we obtain a sufficient condition to solve Wick renormalized stochastic quantization equations with polynomial interaction. Given the power of the nonlinearity, the local solution condition depends on the Hausdorff dimension $d_h$, the walk dimension $d_w$, and the maximal spatial H\"older regularity of the heat kernel $\Theta$. A slightly more restrictive condition based on the same parameters is required for a global solution. For all global solutions, we construct an invariant measure for the Markov process defined by the solution. Our results apply to many rough spaces such as Barlow--Kigami type fractals as well as their Cartesian products and open up the possibility of making rigorous various structures in quantum field theory and statistical mechanics in non-integer dimensions. In the process, we build entirely from the short-time heat semigroup the necessary analytic framework that accommodates the issues which come with allowing rough local geometry.

math.PR↗

arXiv2026

Half Strong Ill-Posedness of $2 \frac{1}{2}$D Electron Magnetohydrodynamics with Fractional Resistivity

We study the $2\frac{1}{2}$D electron magnetohydrodynamics (MHD): the electron MHD system that has $3$D magnetic field but is independent of $z$-variable. We establish a "half" strong ill-posedness result in $2\frac{1}{2}$D electron MHD with fractional resistivity $(-\Delta)^\alpha$ in the supercritical Sobolev space $H^{\beta}\times H^{\beta-1}$ for $3<\beta<4-2\alpha$. Specifically, we construct small initial data $(a_0,b_0)$ whose solution develops a norm inflation in $a$ but the norm of $b$ remains small.

math.AP↗

arXiv2026

Mozart: Modularized and Efficient MoE Training on 3.5D Wafer-Scale Chiplet Architectures

Mixture-of-Experts (MoE) architecture offers enhanced efficiency for Large Language Models (LLMs) with modularized computation, yet its inherent sparsity poses significant hardware deployment challenges, including memory locality issues, communication overhead, and inefficient computing resource utilization. Inspired by the modular organization of the human brain, we propose Mozart, a novel algorithm-hardware co-design framework tailored for efficient training of MoE-based LLMs on 3.5D wafer-scale chiplet architectures. On the algorithm side, Mozart exploits the inherent modularity of chiplets and introduces: (1) an expert allocation strategy that enables efficient on-package all-to-all communication, and (2) a fine-grained scheduling mechanism that improves communication-computation overlap through streaming tokens and experts. On the architecture side, Mozart adaptively co-locates heterogeneous modules on specialized chiplets with a 2.5D NoP-Tree topology and hierarchical memory structure. Evaluation across three popular MoE models demonstrates significant efficiency gains, enabling more effective parallelization and resource utilization for large-scale modularized MoE-LLMs.

cs.AR↗

arXiv2025

Who Gets the Reward & Who Gets the Blame? Evaluation-Aligned Training Signals for Multi-LLM Agents

Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent- and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation to agent credit to response-level signals. Unlike prior approaches that rely only on attribution (Shapley) or step-level labels (PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage; in failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement- or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.

cs.MA↗

arXiv2025

Hallucination reduction with CASAL: Contrastive Activation Steering For Amortized Learning

Large Language Models (LLMs) exhibit impressive capabilities but often hallucinate, confidently providing incorrect answers instead of admitting ignorance. Prior work has shown that models encode linear representations of their own knowledge and that activation steering can reduce hallucinations. These approaches, however, require real-time monitoring and intervention during inference. We introduce Contrastive Activation Steering for Amortized Learning (CASAL), an efficient algorithm that connects interpretability with amortized optimization. CASAL directly bakes the benefits of activation steering into model's weights. Once trained, LLMs answer questions they know while abstaining from answering those they do not. CASAL's light-weight design requires training only a submodule of a single transformer layer and yet reduces hallucination by 30%-40% across multiple short-form QA benchmarks. CASAL is 30x more compute-efficient and 20x more data-efficient than strong LoRA-based baselines such as SFT and DPO, boosting its practical applicability in data scarce domains. Importantly, CASAL also generalizes effectively to out-of-distribution (OOD) domains. We showcase CASAL's flexibility in mitigating hallucinations in both text-only and vision-language models. To our knowledge, CASAL is the first steering-based training method that has been shown to be effective for both dense and Mixture-of-Experts (MoE) models. CASAL represents a promising step forward for applying interpretability-inspired method for practical deployment in production systems.

cs.CL↗

arXiv2025

Smart Fast Finish: Preventing Overdelivery via Daily Budget Pacing at DoorDash

We present a budget pacing feature called Smart Fast Finish (SFF). SFF builds upon the industry standard Fast Finish (FF) feature in budget pacing systems that depletes remaining advertising budget as quickly as possible towards the end of some fixed time period. SFF dynamically updates system parameters such as start time and throttle rate depending on historical ad-campaign data. SFF is currently in use at DoorDash, one of the largest delivery platforms in the US, and is part of its budget pacing system. We show via online budget-split experimentation data and offline simulations that SFF is a robust solution for overdelivery mitigation when pacing budget.

cs.GT↗

arXiv2025

MAHL: Multi-Agent LLM-Guided Hierarchical Chiplet Design with Adaptive Debugging

As program workloads (e.g., AI) increase in size and algorithmic complexity, the primary challenge lies in their high dimensionality, encompassing computing cores, array sizes, and memory hierarchies. To overcome these obstacles, innovative approaches are required. Agile chip design has already benefited from machine learning integration at various stages, including logic synthesis, placement, and routing. With Large Language Models (LLMs) recently demonstrating impressive proficiency in Hardware Description Language (HDL) generation, it is promising to extend their abilities to 2.5D integration, an advanced technique that saves area overhead and development costs. However, LLM-driven chiplet design faces challenges such as flatten design, high validation cost and imprecise parameter optimization, which limit its chiplet design capability. To address this, we propose MAHL, a hierarchical LLM-based chiplet design generation framework that features six agents which collaboratively enable AI algorithm-hardware mapping, including hierarchical description generation, retrieval-augmented code generation, diverseflow-based validation, and multi-granularity design space exploration. These components together enhance the efficient generation of chiplet design with optimized Power, Performance and Area (PPA). Experiments show that MAHL not only significantly improves the generation accuracy of simple RTL design, but also increases the generation accuracy of real-world chiplet design, evaluated by Pass@5, from 0 to 0.72 compared to conventional LLMs under the best-case scenario. Compared to state-of-the-art CLARIE (expert-based), MAHL achieves comparable or even superior PPA results under certain optimization objectives.

cs.AR↗

arXiv2025

Optimal Corpus Aware Training for Neural Machine Translation

Corpus Aware Training (CAT) leverages valuable corpus metadata during training by injecting corpus information into each training example, and has been found effective in the literature, commonly known as the "tagging" approach. Models trained with CAT inherently learn the quality, domain and nuance between corpora directly from data, and can easily switch to different inference behavior. To achieve the best evaluation, CAT models pre-define a group of high quality data before training starts which can be error-prone and inefficient. In this work, we propose Optimal Corpus Aware Training (OCAT), which fine-tunes a CAT pre-trained model by freezing most of the model parameters and only tuning small set of corpus-related parameters. We show that OCAT is lightweight, resilient to overfitting, and effective in boosting model accuracy. We use WMT23 English to Chinese and English to German translation tasks as our test ground and show +3.6 and +1.8 chrF improvement, respectively, over vanilla training. Furthermore, our approach is on-par or slightly better than other state-of-the-art fine-tuning techniques while being less sensitive to hyperparameter settings.

cs.LG↗

arXiv2025

My Advisor, Her AI and Me: Evidence from a Field Experiment on Human-AI Collaboration and Investment Decisions

Amid ongoing policy and managerial debates on keeping humans in the loop of AI decision-making, we investigate whether human involvement in AI-based service production benefits downstream consumers. Partnering with a large savings bank in Europe, we produced pure AI and human-AI collaborative investment advice, passed it to customers, and examined their advice-taking in a field experiment. On the production side, contrary to concerns that humans might inefficiently override AI output, we find that giving a human banker the final say over AI-generated financial advice does not compromise its quality. More importantly, on the consumption side, customers are more likely to follow investment advice from the human-AI collaboration compared to pure AI, especially when facing riskier decisions. In our setting, this increased reliance leads to higher material welfare for consumers. Additional analyses from the field experiment and an online experiment show that the persuasive power of human-AI advice cannot be explained by consumers' beliefs about enhanced advice quality due to human-AI complementarities. Instead, the benefit stems from human involvement acting as a peripheral cue that increases the advice's affective appeal. Our findings suggest that regulations and guidelines should adopt a consumer-centered approach by fostering service environments in which humans and AI systems can collaborate to improve consumer outcomes. These insights are relevant for managers designing AI-based services and for policymakers advocating for human oversight in AI systems.

econ.GN↗

arXiv2025

HDLxGraph: Bridging Large Language Models and HDL Repositories via HDL Graph Databases

Retrieval Augmented Generation (RAG) is an essential agent for Large Language Model (LLM) aided Description Language (HDL) tasks, addressing the challenges of limited training data and prohibitively long prompts. However, its performance in handling ambiguous queries and real-world, repository-level HDL projects containing thousands or even tens of thousands of code lines remains limited. Our analysis demonstrates two fundamental mismatches, structural and vocabulary, between conventional semantic similarity-based RAGs and HDL codes. To this end, we propose HDLxGraph, the first framework that integrates the inherent graph characteristics of HDLs with RAGs for LLM-assisted tasks. Specifically, HDLxGraph incorporates Abstract Syntax Trees (ASTs) to capture HDLs' hierarchical structures and Data Flow Graphs (DFGs) to address the vocabulary mismatch. In addition, to overcome the lack of comprehensive HDL search benchmarks, we introduce HDLSearch, an LLM generated dataset derived from real-world, repository-level HDL projects. Evaluations show that HDLxGraph improves search, debugging, and completion accuracy by 12.04%/12.22%/5.04% and by 11.59%/8.18%/4.07% over state-of-the-art similarity-based RAG and software-code Graph RAG baselines, respectively. The code of HDLxGraph and HDLSearch benchmark are available at https://github.com/UMN-ZhaoLab/HDLxGraph.

cs.AR↗

arXiv2025

Occult: Optimizing Collaborative Communication across Experts for Accelerated Parallel MoE Training and Inference

Mixture-of-experts (MoE) architectures could achieve impressive computational efficiency with expert parallelism, which relies heavily on all-to-all communication across devices. Unfortunately, such communication overhead typically constitutes a significant portion of the total runtime, hampering the scalability of distributed training and inference for modern MoE models (consuming over $40\%$ runtime in large-scale training). In this paper, we first define collaborative communication to illustrate this intrinsic limitation, and then propose system- and algorithm-level innovations to reduce communication costs. Specifically, given a pair of experts co-activated by one token, we call them "collaborated", which comprises $2$ cases as intra- and inter-collaboration, depending on whether they are kept on the same device. Our pilot investigations reveal that augmenting the proportion of intra-collaboration can accelerate expert parallelism at scale. It motivates us to strategically optimize collaborative communication for accelerated MoE training and inference, dubbed Occult. Our designs are capable of either delivering exact results with reduced communication cost or controllably minimizing the cost with collaboration pruning, materialized by modified fine-tuning. Comprehensive experiments on various MoE-LLMs demonstrate that Occult can be faster than popular state-of-the-art inference or training frameworks (more than $1.5\times$ speed up across multiple tasks and models) with comparable or superior quality compared to the standard fine-tuning. Code is available at $\href{https://github.com/UNITES-Lab/Occult}{https://github.com/UNITES-Lab/Occult}$.

cs.LG↗

arXiv2025

Mobile Distributed MIMO (MD-MIMO) for NextG: Mobility Meets Cooperation in Distributed Arrays

Distributed multiple-input multiple-output (D\mbox{-}MIMO) is a promising technology to realize the promise of massive MIMO gains by fiber-connecting the distributed antenna arrays, thereby overcoming the form factor limitations of co-located MIMO. In this paper, we introduce the concept of mobile D-MIMO (MD-MIMO) network, a further extension of the D-MIMO technology where distributed antenna arrays are connected to the base station with a wireless link allowing all radio network nodes to be mobile. This approach significantly improves deployment flexibility and reduces operating costs, enabling the network to adapt to the highly dynamic nature of next-generation (NextG) networks. We discuss use cases, system design, network architecture, and the key enabling technologies for MD-MIMO. Furthermore, we investigate a case study of MD-MIMO for vehicular networks, presenting detailed performance evaluations for both downlink and uplink. The results show that an MD-MIMO network can provide substantial improvements in network throughput and reliability.

eess.SP↗

arXiv2024

HiVeGen -- Hierarchical LLM-based Verilog Generation for Scalable Chip Design

With Large Language Models (LLMs) recently demonstrating impressive proficiency in code generation, it is promising to extend their abilities to Hardware Description Language (HDL). However, LLMs tend to generate single HDL code blocks rather than hierarchical structures for hardware designs, leading to hallucinations, particularly in complex designs like Domain-Specific Accelerators (DSAs). To address this, we propose HiVeGen, a hierarchical LLM-based Verilog generation framework that decomposes generation tasks into LLM-manageable hierarchical submodules. HiVeGen further harnesses the advantages of such hierarchical structures by integrating automatic Design Space Exploration (DSE) into hierarchy-aware prompt generation, introducing weight-based retrieval to enhance code reuse, and enabling real-time human-computer interaction to lower error-correction cost, significantly improving the quality of generated designs.

cs.LG↗