arXiv subjects

Yikai Zhao

Publications and source records attributed to Yikai Zhao.

16 recordsLinked to original sources

arXiv2026

A Tri-Agent Framework for Evaluating and Aligning Question Clarification Capabilities of Large Language Models

Large Language Models (LLMs) are increasingly deployed in interactive systems where understanding user intent precisely is paramount. A key capability for such systems is effective question clarification, especially when user queries are ambiguous or underspecified. This paper introduces a novel tri-agent framework for the robust evaluation of an LLM's ability to engage in clarifying dialogue. Our framework comprises three distinct LLM-based agents: (1) a Question Clarifying Agent (QCA), the system under evaluation, tasked with identifying ambiguities and posing clarifying questions; (2) a Respondent Agent (RA), designed to simulate human user responses, potentially including irrelevant or challenging replies; and (3) an Evaluator Agent (EA), an LLM-as-a-judge, which assesses the quality of the dialogue based on a comprehensive set of metrics. We detail a methodology for synthetic data generation in the supply chain domain as an example. We propose metrics evaluating ambiguity handling, question quality, dialogue efficiency, language appropriateness, and final intent alignment. We also briefly discuss the validation of the EA against human judgments. This work provides a structured approach to benchmark, validate, and improve the clarification capabilities of conversational LLM applications.

cs.CL↗

arXiv2026

Context-Aware Cluster Decoding: Semantic Anchor-Driven Coherence in dMLLMs

Diffusion multimodal large language models (dMLLMs) frequently produce long-form outputs marred by semantic drift and repetition, with quality generally degrading as output length increases. We identify two structural deficiencies in existing decoding methods as primary drivers of these failures: confidence-based scoring ignores decoded-neighbor support, and block partitioning prevents access to high-readiness semantic anchors, together causing tokens to be committed before their local context is sufficiently established. We propose \ours{} (\textbf{C}ontext-\textbf{A}ware \textbf{C}luster \textbf{D}ecoding), a training-free decoding method that scores each masked position by a multiplicative composite of softmax confidence and neighbor proximity, promoting contextually ready tokens above isolated candidates while suppressing low-confidence positional noise, operating block-free to keep high-readiness anchors globally accessible. \ours{} further applies architecture-aware calibration to handle confidence heterogeneity induced by diverse visual integration strategies. Experiments on three dMLLMs across four benchmarks demonstrate consistent quality gains and hallucination reduction over Original, with larger gains in several longer generation settings, highlighting the importance of neighbor support and visual integration strategy for future dMLLM decoding method design. Our code is openly available at https://github.com/zhaoyk-sysu/CACD-dMLLM.

cs.CL↗

arXiv2026

TRACE: TRajectory Attribution for Automated Context Engineering

Production AI agents fail when their context sources -- system prompts, knowledge bases, tool descriptions, and procedural skills -- contain errors or gaps. Current maintenance relies on manual log review and ad-hoc debugging, creating a scalability bottleneck as interaction volume grows. We present TRACE (TRajectory Attribution for Automated Context Engineering), an automated feedback loop that mines historical agent trajectories to diagnose and remediate context failures. Our key insight is that trajectories are rich with implicit dissatisfaction signals -- user corrections, rephrasing, abandonment cues -- that reveal precisely where context sources failed, without explicit feedback collection. Unlike model fine-tuning, TRACE operates on the context layer, enabling rapid iteration without retraining. We make four contributions: (1) a trajectory mining framework that systematically extracts diagnostic information from historical agent executions; (2) multi-component causal attribution that extends textual gradients from monolithic prompt optimization to heterogeneous context sources (skills, knowledge bases, tools, prompts); (3) exploratory verification, where agents actively read context sources to distinguish content gaps requiring CREATE from stale content requiring UPDATE, achieving 96% operation accuracy; and (4) a reusable simulation methodology and verifiable benchmark addressing the absence of open datasets for context debugging, with a six-category fault taxonomy, ground truth annotations, and a cross-layer verification protocol. On 60 dissatisfaction traces spanning three complexity tiers (up to 16 execution nodes), TRACE achieves 72.7% root cause attribution and 82% end-to-end fix effectiveness, showing that over 80% of context-layer failures can be automatically diagnosed and remediated by mining historical trajectories, an overlooked resource in production systems.

cs.AI↗

arXiv2026

Seer: Online Context Learning for Fast Synchronous LLM Reinforcement Learning

Reinforcement Learning (RL) has emerged as a critical technique for advancing modern Large Language Models (LLMs), yet existing synchronous RL systems face severe performance bottlenecks. The rollout phase, which dominates end-to-end iteration time, suffers from substantial long-tail latency and poor resource utilization due to inherent workload imbalance. We present Seer, a novel context learning RL system that addresses these challenges through a key observation: requests sharing the same prompt exhibit strong similarities in output lengths and response patterns. Leveraging this insight, Seer introduces three coordinated techniques: (1) divided rollout for dynamic load balancing, (2) context-aware scheduling to mitigate long-tail request delays, and (3) adaptive grouped speculative decoding to accelerate generation. These mechanisms work in concert to markedly reduce long-tail latency and improve resource efficiency during rollout. Evaluations on production-grade RL workloads demonstrate that Seer achieves up to 2.04$\times$ end-to-end rollout throughput improvement compared to the state-of-the-art synchronous RL systems, while notably reducing long-tail latency by 72-94%.

cs.DC↗

arXiv2026

TENT: A Declarative Slice Spraying Engine for Performant and Resilient Data Movement in Disaggregated LLM Serving

Modern GPU clusters rely on complex, heterogeneous interconnects. As large language model (LLM) serving shifts toward agentic reasoning, KVCache becomes a first-class mobile asset, driving frequent migrations and massive elephant flows that dominate the execution critical path. Operating Mooncake Transfer Engine (TE) on thousands of GPUs exposed a fundamental flaw in existing frameworks: imperative, early-binding path selection. This rigidity results in state-blind striping that ignores congestion and grey failures, leading to bandwidth stranding. This also produces operational fragility where routine faults require manual intervention. We present TENT, a declarative orchestration engine that decouples transfer intent from physical execution. By abstracting interconnects into a unified resource pool, TENT shifts path resolution from initialization to slice-time late binding. Applications simply declare transfer intents, while TENT dynamically "sprays" fine-grained slices across rails based on real-time telemetry and predictive cost modeling. This orchestration eliminates head-of-line (HoL) blocking and enables transparent, sub-50 ms self-healing by rerouting slices around failures or degradations without application-level intervention. TENT serves as the production data plane for LLM inference and reinforcement learning (RL) pipelines at multiple industrial clusters. Our evaluation shows that TE+ outperforms state-of-the-art baselines, including Mooncake TE, NIXL, and UCCL. In LLM inference with SGLang HiCache, TE+ achieves up to 1.36x higher throughput and 26% lower P90 time-to-first-token (TTFT) than Mooncake TE. In RL pipelines, TENT accelerates parameter updates in Moonshot Checkpoint Engine by 20-26%.

cs.DC↗

arXiv2025

MergeMoE: Efficient Compression of MoE Models via Expert Output Merging

The Mixture-of-Experts (MoE) technique has proven to be a promising solution to efficiently scale the model size, which has been widely applied in recent LLM advancements. However, the substantial memory overhead of MoE models has made their compression an important research direction. In this work, we provide a theoretical analysis of expert merging, a recently proposed technique for compressing MoE models. Rather than interpreting expert merging from the conventional perspective of parameter aggregation, we approach it from the perspective of merging experts' outputs. Our key insight is that the merging process can be interpreted as inserting additional matrices into the forward computation, which naturally leads to an optimization formulation. Building on this analysis, we introduce MergeMoE, a method that leverages mathematical optimization to construct the compression matrices. We evaluate MergeMoE on multiple MoE models and show that our algorithm consistently outperforms the baselines with the same compression ratios.

cs.LG↗

arXiv2025

Faithful and Interpretable Explanations for Complex Ensemble Time Series Forecasts using Surrogate Models and Forecastability Analysis

Modern time series forecasting increasingly relies on complex ensemble models generated by AutoML systems like AutoGluon, delivering superior accuracy but with significant costs to transparency and interpretability. This paper introduces a comprehensive, dual-approach framework that addresses both the explainability and forecastability challenges in complex time series ensembles. First, we develop a surrogate-based explanation methodology that bridges the accuracy-interpretability gap by training a LightGBM model to faithfully mimic AutoGluon's time series forecasts, enabling stable SHAP-based feature attributions. We rigorously validated this approach through feature injection experiments, demonstrating remarkably high faithfulness between extracted SHAP values and known ground truth effects. Second, we integrated spectral predictability analysis to quantify each series' inherent forecastability. By comparing each time series' spectral predictability to its pure noise benchmarks, we established an objective mechanism to gauge confidence in forecasts and their explanations. Our empirical evaluation on the M5 dataset found that higher spectral predictability strongly correlates not only with improved forecast accuracy but also with higher fidelity between the surrogate and the original forecasting model. These forecastability metrics serve as effective filtering mechanisms and confidence scores, enabling users to calibrate their trust in both the forecasts and their explanations. We further demonstrated that per-item normalization is essential for generating meaningful SHAP explanations across heterogeneous time series with varying scales. The resulting framework delivers interpretable, instance-level explanations for state-of-the-art ensemble forecasts, while equipping users with forecastability metrics that serve as reliability indicators for both predictions and their explanations.

cs.LG↗

arXiv2025

Router Upcycling: Leveraging Mixture-of-Routers in Mixture-of-Experts Upcycling

The Mixture-of-Experts (MoE) models have gained significant attention in deep learning due to their dynamic resource allocation and superior performance across diverse tasks. However, efficiently training these models remains challenging. The MoE upcycling technique has been proposed to reuse and improve existing model components, thereby minimizing training overhead. Despite this, simple routers, such as linear routers, often struggle with complex routing tasks within MoE upcycling. In response, we propose a novel routing technique called Router Upcycling to enhance the performance of MoE upcycling models. Our approach initializes multiple routers from the attention heads of preceding attention layers during upcycling. These routers collaboratively assign tokens to specialized experts in an attention-like manner. Each token is processed into diverse queries and aligned with the experts' features (serving as keys). Experimental results demonstrate that our method achieves state-of-the-art (SOTA) performance, outperforming other upcycling baselines.

cs.CL↗

arXiv2025

Confidence Scoring for LLM-Generated SQL in Supply Chain Data Extraction

Large Language Models (LLMs) have recently enabled natural language interfaces that translate user queries into executable SQL, offering a powerful solution for non-technical stakeholders to access structured data. However, one of the limitation that LLMs do not natively express uncertainty makes it difficult to assess the reliability of their generated queries. This paper presents a case study that evaluates multiple approaches to estimate confidence scores for LLM-generated SQL in supply chain data retrieval. We investigated three strategies: (1) translation-based consistency checks; (2) embedding-based semantic similarity between user questions and generated SQL; and (3) self-reported confidence scores directly produced by the LLM. Our findings reveal that LLMs are often overconfident in their own outputs, which limits the effectiveness of self-reported confidence. In contrast, embedding-based similarity methods demonstrate strong discriminative power in identifying inaccurate SQL.

stat.AP↗

arXiv2024

Carbonyl4: A Sketch for Set-Increment Mixed Updates

In the realm of data stream processing, the advent of SET-INCREMENT Mixed (SIM) data streams necessitates algorithms that efficiently handle both SET and INCREMENT operations. We present Carbonyl4, an innovative algorithm designed specifically for SIM data streams, ensuring accuracy, unbiasedness, and adaptability. Carbonyl4 introduces two pioneering techniques: the Balance Bucket for refined variance optimization, and the Cascading Overflow for maintaining precision amidst overflow scenarios. Our experiments across four diverse datasets establish Carbonyl4's supremacy over existing algorithms, particularly in terms of accuracy for item-level information retrieval and adaptability to fluctuating memory requirements. The versatility of Carbonyl4 is further demonstrated through its dynamic memory shrinking capability, achieved via a re-sampling and a heuristic approach. The source codes of Carbonyl4 are available at GitHub.

cs.DS↗

arXiv2024

Approaching 100% Confidence in Stream Summary through ReliableSketch

To approximate sums of values in key-value data streams, sketches are widely used in databases and networking systems. They offer high-confidence approximations for any given key while ensuring low time and space overhead. While existing sketches are proficient in estimating individual keys, they struggle to maintain this high confidence across all keys collectively, an objective that is critically important in both algorithm theory and its practical applications. We propose ReliableSketch, the first to control the error of all keys to less than $Λ$ with a small failure probability $Δ$, requiring only $O(1 + Δ\ln\ln(\frac{N}Λ))$ amortized time and $O(\frac{N}Λ + \ln(\frac{1}Δ))$ space. Furthermore, its simplicity makes it hardware-friendly, and we implement it on CPU servers, FPGAs, and programmable switches. Our experiments show that under the same small space, ReliableSketch not only keeps all keys' errors below $Λ$ but also achieves near-optimal throughput, outperforming competitors with thousands of uncontrolled estimations. We have made our source code publicly available.

cs.DS↗

arXiv2024

CAFE: Towards Compact, Adaptive, and Fast Embedding for Large-scale Recommendation Models

Recently, the growing memory demands of embedding tables in Deep Learning Recommendation Models (DLRMs) pose great challenges for model training and deployment. Existing embedding compression solutions cannot simultaneously meet three key design requirements: memory efficiency, low latency, and adaptability to dynamic data distribution. This paper presents CAFE, a Compact, Adaptive, and Fast Embedding compression framework that addresses the above requirements. The design philosophy of CAFE is to dynamically allocate more memory resources to important features (called hot features), and allocate less memory to unimportant ones. In CAFE, we propose a fast and lightweight sketch data structure, named HotSketch, to capture feature importance and report hot features in real time. For each reported hot feature, we assign it a unique embedding. For the non-hot features, we allow multiple features to share one embedding by using hash embedding technique. Guided by our design philosophy, we further propose a multi-level hash embedding framework to optimize the embedding tables of non-hot features. We theoretically analyze the accuracy of HotSketch, and analyze the model convergence against deviation. Extensive experiments show that CAFE significantly outperforms existing embedding compression methods, yielding 3.92% and 3.68% superior testing AUC on Criteo Kaggle dataset and CriteoTB dataset at a compression ratio of 10000x. The source codes of CAFE are available at GitHub.

cs.LG↗

arXiv2023

ChainedFilter: Combining Membership Filters by Chain Rule

Membership (membership query / membership testing) is a fundamental problem across databases, networks and security. However, previous research has primarily focused on either approximate solutions, such as Bloom Filters, or exact methods, like perfect hashing and dictionaries, without attempting to develop a an integral theory. In this paper, we propose a unified and complete theory, namely chain rule, for general membership problems, which encompasses both approximate and exact membership as extreme cases. Building upon the chain rule, we introduce a straightforward yet versatile algorithm framework, namely ChainedFilter, to combine different elementary filters without losing information. Our evaluation results demonstrate that ChainedFilter performs well in many applications: (1) it requires only 26% additional space over the theoretical lower bound for implicit static dictionary, (2) it requires only 0.22 additional bit per item over the theoretical lower bound for lossless data compression, (3) it reduces up to 31% external memory access than raw Cuckoo Hashing, (4) it reduces up to 36% P99 tail point query latency than Bloom Filter under the same space cost in RocksDB database, and (5) it reduces up to 99.1% filter space than original Learned Bloom Filter.

cs.DB↗

arXiv2023

ChameleMon: Shifting Measurement Attention as Network State Changes

Flow-level network measurement is critical to many network applications. Among various measurement tasks, packet loss detection and heavy-hitter detection are two most important measurement tasks, which we call the two key tasks. In practice, the two key tasks are often required at the same time, but existing works seldom handle both tasks. In this paper, we design ChameleMon to support the two key tasks simultaneously. One key design/novelty of ChameleMon is to shift measurement attention as network state changes, through two dimensions of dynamics: 1) dynamically allocating memory between the two key tasks; 2) dynamically monitoring the flows of importance. To realize the key design, we propose a key technique, leveraging Fermat's little theorem to devise a flexible data structure, namely FermatSketch. FermatSketch is dividable, additive, and subtractive, supporting the two key tasks. We have fully implemented a ChameleMon prototype on a testbed with a Fat-tree topology. We conduct extensive experiments and the results show ChameleMon supports the two key tasks with low memory/bandwidth overhead, and more importantly, it can automatically shift measurement attention as network state changes.

cs.NI↗

arXiv2022

QCluster: Clustering Packets for Flow Scheduling

Flow scheduling is crucial in data centers, as it directly influences user experience of applications. According to different assumptions and design goals, there are four typical flow scheduling problems/solutions: SRPT, LAS, Fair Queueing, and Deadline-Aware scheduling. When implementing these solutions in commodity switches with limited number of queues, they need to set static parameters by measuring traffic in advance, while optimal parameters vary across time and space. This paper proposes a generic framework, namely QCluster, to adapt all scheduling problems for limited number of queues. The key idea of QCluster is to cluster packets with similar weights/properties into the same queue. QCluster is implemented in Tofino switches, and can cluster packets at a speed of 3.2 Tbps. To the best of our knowledge, QCluster is the fastest clustering algorithm. Experimental results in testbed with programmable switches and ns-2 show that QCluster reduces the average flow completion time (FCT) for short flows up to 56.6%, and reduces the overall average FCT up to 21.7% over state-of-the-art. All the source code in ns-2 is available in Github without.

cs.NI↗

arXiv2021

Novel tile segmentation scheme for omnidirectional video

Regular omnidirectional video encoding technics use map projection to flatten a scene from a spherical shape into one or several 2D shapes. Common projection methods including equirectangular and cubic projection have varying levels of interpolation that create a large number of non-information-carrying pixels that lead to wasted bitrate. In this paper, we propose a tile based omnidirectional video segmentation scheme which can save up to 28% of pixel area and 20% of BD-rate averagely compared to the traditional equirectangular projection based approach.

cs.CV↗