arXiv subjects

Qi Luo

Publications and source records attributed to Qi Luo.

At least 19 recordsLinked to original sources

arXiv2026

Parameter-Free First-Order Method for Constrained Variational Inequalities

Constrained variational inequalities (VIs) provide a general framework for modeling and solving complementarity and equilibrium problems with functional constraints. In this paper, we propose a parameter-free accelerated operator sliding (PFAOS) method for solving monotone VIs with convex smooth constraints. For a single constraint, PFAOS achieves an operator evaluation complexity of $\mathcal{O}(1/\varepsilon)$ and a constraint gradient evaluation complexity of $\mathcal{O}(1/\sqrt{\varepsilon})$, without requiring prior knowledge of the Lipschitz constant of the VI operator or the smoothness constant of the constraint function. We further extend PFAOS to multiple constraints by incorporating a primal-dual hybrid gradient update, yielding a multi-constrained PFAOS (MC-PFAOS) method that preserves the same evaluation complexities. Finally, numerical experiments are conducted on large-scale games with coupling constraints to validate the efficiency and effectiveness of the proposed algorithms.

math.OC↗

arXiv2026

Utility-Aware Data Pricing: Token-Level Quality and Empirical Training Gain for LLMs

Traditional ``row-count $\times$ quality coefficient'' approaches fail to capture the nonlinear utility of data for Large Language Model (LLM) capabilities. We present a utility-aware data valuation framework that moves from static accounting toward utility-based pricing. The framework combines token-level information density and data quality, empirical training contribution estimated by influence functions, proxy models, and Data Shapley values, and cryptographic mechanisms based on hash commitments, Merkle trees, and tamper-evident training ledgers. Preliminary evaluation includes a smoke-scale multi-domain study and a synthetic duplication test. Results show that proxy-based empirical gain achieves strong ranking agreement with proxy-defined realized utility and substantially outperforms row-count and token-count baselines in the tested setting, while the prototype provides a controlled sanity check of the valuation pipeline and its sensitivity to duplication-based manipulation.

cs.LG↗

arXiv2026

Where Does the Energy Go? Profiling LLM Agent Inference on Blackwell GPUs

LLM agents that iteratively reason, plan, and invoke tools create workload profiles fundamentally different from single-pass inference, yet how their energy consumption is distributed across hardware components and workload phases remains poorly understood. Characterizing these workloads therefore requires simultaneous visibility into both component-level power and phase-level execution. We conduct a full-stack energy profiling study combining NVML GPU counters, Intel RAPL CPU/DRAM counters, and Intelligent Platform Management Interface (IPMI) system-level sensors on 2x NVIDIA RTX PRO 6000 Blackwell GPUs, and profile three representative workloads with Qwen3.8-27B. For mathematical reasoning, we compare thinking-enabled and thinking-disabled modes. Our measurements reveal that GPU-only telemetry misses 41-45% of system energy across all three workloads, with non-GPU components accounting for the remainder. In our setup, the sequential agent workload consumes 63x more system energy per output token than saturated serving, reflecting the absence of batching, context growth across turns, and tool-induced idle periods. Extended thinking generates 21-75% more tokens per problem, while per-token energy differs by less than 1% between modes within each dataset, indicating that the resulting increase in energy is driven by output volume rather than a change in per-token efficiency. Continuous batching improves system-level energy efficiency by 3.2x from 1 to 16 requests per second, as GPU power plateaus while throughput continues to increase with batching depth. For the memory-bandwidth-bound reasoning workload, throughput remains unchanged under per-GPU power caps of 400-600 W but drops sharply at 300 W. These findings suggest that context-management techniques such as summarization and selective retrieval may help reduce energy consumption in agent deployments.

cs.DC↗

arXiv2026

SpeechGym: An Audio-Native Gym for Training Voice Agents via Reinforcement Learning

Voice agents must call tools and hold multi-turn dialogue entirely through speech, yet the dominant paradigm trains them in text. Existing frameworks either cascade TTS and ASR around a proprietary voice API, where gradients cannot flow and per-call cost makes on-policy reinforcement learning prohibitive, or stay in text: they measure voice agents but cannot improve them. We present SpeechGym, an audio-native agentic environment in which two omni-modal models converse in native audio, with no external ASR or TTS and no API boundary, over the unmodified tasks, tools and success check of an established text agentic benchmark, so that the interaction modality is the only variable and the loop stays local and trainable end to end. Audio agentic capability does not follow from audio understanding. The failures speech introduces are perceptual rather than reasoning deficits: the agent picks the right tool and the right argument slot but fills it with a value misheard from the waveform, and that single error cascades into a failed call, a retry of the same call, and a wasted step budget. A second failure is behavioural: under an insistent caller the agent performs an unauthorised write and ends the episode believing it helped. Both are trainable, because the environment labels them for free: a call with a misheard argument fails against the database while a correct one succeeds. The obstacle is sparsity, not signal. Outcome-only GRPO is gradient-starved here, since almost every rollout group fails identically, while a per-turn process reward crediting each successful tool call restores variance to nearly every group. Trained this way, the agent transfers with no further tuning to an independently implemented voice benchmark, more than doubling task success and carrying an open-weights model from last place to second on that leaderboard, while using fewer turns and tokens than before training.

cs.SD↗

arXiv2026

Can We Read the Mind of an Audio LLM? A Verbalizable, Multilingual Middle-Layer Workspace

An audio language model is a black box in a specific way: we see what it says, never what it works out on the way there, and chain-of-thought monitoring helps only if the model writes its reasoning down. Reading a base Qwen3-Omni with a logit lens at the audio-token positions, we find that the answer to a spoken question becomes legible - in words - in the model's middle layers, before it emits any token. Five findings follow. (1) The readout carries concepts in neither the question, the options, nor the model's own transcription: on a clip whose verbatim transcription is empty garbling, it reconstructs Watergate and scandal, passes through the role president, and resolves to Nixon - a hidden multi-hop chain, read with no chain-of-thought. (2) The content is language-agnostic: one audio-inferred concept surfaces in several scripts at once, and 38% of top-1 readouts are Chinese on English inputs. (3) It is paralinguistic: given the same clip as audio and as the model's own emotion-free caption, the audio mind forms the sound source, speaker role, or affect that the caption discards, and answers correctly more often. (4) The audio-driven signal is absent at the input, turns on about a tenth of the way into the network, separates most cleanly from the text prior in the middle band (35-80% of depth), and activation patching shows it is causally used and committed before the last fifth of the layers. (5) Deleting single layers maps the pipeline: reading the sound in is localized to the entry layers and answer delivery to the output layer, while retrieval is distributed across the interior. Throughout, a waveform-swap control - identical text, only the sound changed - isolates the audio-driven signal from a prior over the printed options. This is a qualitative account of what an audio model works out before it speaks: the quantities are controls, not benchmark scores.

cs.SD↗

arXiv2026

CoWAM: Coordination Contracts for Selective Policy Intervention with WAMs

World Action Models (WAMs) augment robot policies with action-conditioned predicted futures, but a plausible future alone does not justify changing the action that a bimanual policy would execute. We present CoWAM, a selective intervention layer that expresses synchronization, role compatibility, and collision convergence as coordination contracts. Each contract combines typed admissibility checks with event-conditioned verification and calibrated intervention gates. CoWAM preserves the nominal action unless an alternative satisfies every active obligation and provides a clear, low-risk improvement; when the nominal action is also inadmissible, it invokes a predefined abstention fallback. To separate selector quality from proposal quality, all methods operate on identical candidate pools and commit their decisions before shared oracle labeling. Across eight simulated bimanual tasks, CoWAM improves coordination-valid selection by 16.7 percentage points over the contract-only variant and raises closed-loop success by 9.6 percentage points over the strongest selective baseline, while keeping harmful interventions below 1%. Together, these results establish coordination contracts as an effective interface for conservative policy intervention with predicted world-action evidence across coordination-rich bimanual tasks.

cs.RO↗

arXiv2026

The Gate, Not the Cache: Gate Provenance Bounds the Closed-Loop Reliability of Training-Free VLA Token Skipping

Token skipping is a widely used training-free way to accelerate vision--language--action (VLA) models by bypassing computation for most visual tokens at each control step according to a gate. When the next gate is harvested from the previous accelerated forward, however, the tokens skipped at one step are also the ones least visible to the next gate, and the damage can compound across control steps until the task fails. We study the two mechanisms this class is built on, reuse and deletion, crossing each against where its gate signal comes from on identical episodes. At a skip ratio of 0.9 on LIBERO-Object, both collapse when the gate comes from the model's own accelerated forwards, to 0.68 under reuse and to 0.31 under deletion against a dense 1.00, and the collapse is invisible to the action-level detectors we evaluate. What separates collapse from dense-level operation is not the mechanism but whether the gate is clean, computed by a forward that skipped nothing. We therefore propose actuation-slack refresh, one dense pass run while the robot executes its current action chunk, off the critical path, that hands the next step a clean gate and a fresh KV base. Since the measured detectors do not reliably reveal the failure, the refresh is unconditional rather than triggered. Both mechanisms then recover to 0.98, keeping the speed of skipping and the information of a dense pass. We then integrate the refresh into state-of-the-art caching and pruning methods across two VLA policies, 4 LIBERO suites, and 4 SIMPLER tasks, where it repairs every collapse caused by using a self-harvested gate. Serve latency drops 18--22\% below dense, measured both in simulation and on a physical robot. Where the gate signal comes from, not how tokens are skipped, decides closed-loop reliability for accelerated VLAs.

cs.RO↗

arXiv2026

OmniVAE: An Audio-Video VAE with Cross-Modal Alignment for Joint Generation

Recent generative models are moving beyond silent video or standalone audio synthesis toward the joint generation of synchronized audio and video. Despite this progress, jointly generating audio and video with fine-grained cross-modal correspondence remains challenging due to their fundamental structural differences. Most existing methods use audio and video VAEs trained separately. As a result, the two latent spaces lack cross-modal alignment, leaving the downstream generative model to learn cross-modal synchronization from scratch. We present OmniVAE, a jointly trained audio-video VAE that learns fine-grained semantic alignment between audio and video latent representations. Beyond reconstruction, OmniVAE uses a segment-level audio-video contrastive objective to capture temporal-semantic correspondence and align the two latent spaces. In parallel, it distills features from pretrained modality-specific semantic encoders into each modality, improving the downstream learnability of both latent spaces. Extensive experiments show that both objectives consistently improve the learnability of the latent spaces, translating into higher generation quality and more accurate cross-modal synchronization in downstream text-to-audio-video generation. These findings underscore the importance of learning unified representations as a foundation for omnimodal modeling.1

cs.SD↗

arXiv2026

Request-Level Energy Attribution for Batched LLM Serving

Batched LLM serving improves throughput but complicates energy accounting. GPU power telemetry is aggregate, whereas sustainability reporting, chargeback, and workload analysis often require request-level energy charges. Existing inference-energy benchmarks report model-, phase-, or token-level energy, and recent carbon-accounting work motivates Shapley fairness conceptually. Neither provides measured request-level ground truth, so how far the accounting rules used in practice deviate from a fair allocation has remained unknown. We present JouleShare, an attribution framework with two components. An offline harness establishes this ground truth by replaying request subsets under vLLM with a reproducible protocol, integrating GPU power telemetry, and computing exact Shapley energy for each request. A lightweight calibration model, JCalib, then learns to predict Shapley shares from cheap request features for use at serving time. Across 16 model/workload runs, token-proportional attribution differs from exact Shapley by 0.440 normalized L1 on average under static batching and by 0.458 under continuous batching, a gap that reproduces across three data-center GPUs. JCalib reduces this error to 0.116 under static batching and 0.177 under continuous batching, below even a standalone-measurement baseline that is unavailable online, while preserving exact batch-energy efficiency. Sampled Shapley extends the measured reference to larger group sizes, where the gap persists and a single offline calibration remains the most accurate deployable rule. The results show that token attribution is not a reliable proxy for marginal energy under batched execution, and that measured Shapley ground truth can calibrate low-cost request features toward fairer attribution.

cs.AI↗

arXiv2026

Universal and Parameter-free Gradient Sliding for Composite Optimization

We propose a Parameter-Free Universal Gradient Sliding (PFUGS) algorithm for computing an approximate solution to the convex composite optimization $\min_{x\in X} \{f(x) + g(x)\}$, where $f$ has $(M_ν,ν)$-Hölder continuous subgradient and $g$ has $L$-Lipschitz continuous gradient. PFUGS computes an $\varepsilon$-approximate solution with $\mathcal{O}((M_ν/\varepsilon)^{{2}/{(1+3ν)}})$ evaluations of (sub)gradients of $f$ and $\mathcal{O}((L/\varepsilon)^{1/2})$ evaluations of gradients of $g$, without prior knowledge of problem constants. To the best of our knowledge, PFUGS is the first gradient sliding algorithm for problems involving two functions whose distinct problem constants are both unknown a priori.

math.OC↗

arXiv2026

Graph-Aware Stealthy Poison-Text Backdoors for Text-Attributed Graphs

Modern graph learning systems often combine links with text, as in citation networks with abstracts or social graphs with user posts. In such systems, text is usually easier to edit than graph structure, which creates a practical security risk: an attacker may hide a small malicious cue in training text and later use it to trigger incorrect predictions. This paper studies that risk in a realistic setting where the attacker edits only node text and leaves the graph unchanged. We propose \textbf{TAGBD}, a graph-aware backdoor attack that first selects training nodes that are easier to manipulate, then generates stealthy poison text with a shadow graph model, and finally injects the text by replacing the original content or appending a short phrase. Experiments on three benchmark datasets show that TAGBD achieves very high attack success rates, transfers across different graph models, and remains effective under common defenses. These results show that inconspicuous poison text alone can serve as a reliable attack channel in text-attributed graphs, highlighting the need for defenses that inspect both node content and graph structure.

cs.LG↗

arXiv2026

DualSentinel: A Lightweight Framework for Detecting Targeted Attacks in Black-box LLM via Dual Entropy Lull Pattern

Recent intelligent systems integrate powerful Large Language Models (LLMs) through APIs, but their trustworthiness may be critically undermined by targeted attacks like backdoor and prompt injection attacks, which secretly force LLMs to generate specific malicious sequences. Existing defensive approaches for such threats typically rely on high access rights, impose prohibitive costs, and hinder normal inference, rendering them impractical for real-world scenarios. To solve these limitations, we introduce DualSentinel, a lightweight and unified defense framework that can accurately and promptly detect the activation of targeted attacks alongside the LLM generation process. We first identify a characteristic of compromised LLMs, termed Entropy Lull: when a targeted attack successfully hijacks the generation process, the LLM exhibits a distinct period of abnormally low and stable token probability entropy, indicating it is following a fixed path rather than making creative choices. DualSentinel leverages this pattern by developing an innovative dual-check approach. It first employs a magnitude and trend-aware monitoring method to proactively and sensitively flag an entropy lull pattern at runtime. Upon such flagging, it triggers a lightweight yet powerful secondary verification based on task-flipping. An attack is confirmed only if the entropy lull pattern persists across both the original and the flipped task, proving that the LLM's output is coercively controlled. Extensive evaluations show that DualSentinel is both highly effective (superior detection accuracy with near-zero false positives) and remarkably efficient (negligible additional cost), offering a truly practical path toward securing deployed LLMs. The source code can be accessed at https://doi.org/10.5281/zenodo.18479273.

cs.CR↗

arXiv2026

MOVA: Towards Scalable and Synchronized Video-Audio Generation

Audio is indispensable for real-world video, yet generation models have largely overlooked audio components. Current approaches to producing audio-visual content often rely on cascaded pipelines, which increase cost, accumulate errors, and degrade overall quality. While systems such as Veo 3 and Sora 2 emphasize the value of simultaneous generation, joint multimodal modeling introduces unique challenges in architecture, data, and training. Moreover, the closed-source nature of existing systems limits progress in the field. In this work, we introduce MOVA (MOSS Video and Audio), an open-source model capable of generating high-quality, synchronized audio-visual content, including realistic lip-synced speech, environment-aware sound effects, and content-aligned music. MOVA employs a Mixture-of-Experts (MoE) architecture, with a total of 32B parameters, of which 18B are active during inference. It supports IT2VA (Image-Text to Video-Audio) generation task. By releasing the model weights and code, we aim to advance research and foster a vibrant community of creators. The released codebase features comprehensive support for efficient inference, LoRA fine-tuning, and prompt enhancement.

cs.CV↗

arXiv2026

Low Rank Comes with Low Security: Gradient Assembly Poisoning Attacks against Distributed LoRA-based LLM Systems

Low-Rank Adaptation (LoRA) has become a popular solution for fine-tuning large language models (LLMs) in federated settings, dramatically reducing update costs by introducing trainable low-rank matrices. However, when integrated with frameworks like FedIT, LoRA introduces a critical vulnerability: clients submit $A$ and $B$ matrices separately, while only their product $AB$ determines the model update, yet this composite is never directly verified. We propose Gradient Assembly Poisoning (GAP), a novel attack that exploits this blind spot by crafting individually benign $A$ and $B$ matrices whose product yields malicious updates. GAP operates without access to training data or inter-client coordination and remains undetected by standard anomaly detectors. We identify four systemic vulnerabilities in LoRA-based federated systems and validate GAP across LLaMA, ChatGLM, and GPT-2. GAP consistently induces degraded or biased outputs while preserving surface fluency, reducing BLEU by up to 14.5\%, increasing factual and grammatical errors by over 800\%, and maintaining 92.6\% long-form response length. These results reveal a new class of stealthy, persistent threats in distributed LoRA fine-tuning.

cs.CR↗

arXiv2026

Multi-hop Reasoning via Early Knowledge Alignment

Retrieval-Augmented Generation (RAG) has emerged as a powerful paradigm for Large Language Models (LLMs) to address knowledge-intensive queries requiring domain-specific or up-to-date information. To handle complex multi-hop questions that are challenging for single-step retrieval, iterative RAG approaches incorporating reinforcement learning have been proposed. However, existing iterative RAG systems typically plan to decompose questions without leveraging information about the available retrieval corpus, leading to inefficient retrieval and reasoning chains that cascade into suboptimal performance. In this paper, we introduce Early Knowledge Alignment (EKA), a simple but effective module that aligns LLMs with retrieval set before planning in iterative RAG systems with contextually relevant retrieved knowledge. Extensive experiments on six standard RAG datasets demonstrate that by establishing a stronger reasoning foundation, EKA significantly improves retrieval precision, reduces cascading errors, and enhances both performance and efficiency. Our analysis from an entropy perspective demonstrate that incorporating early knowledge reduces unnecessary exploration during the reasoning process, enabling the model to focus more effectively on relevant information subsets. Moreover, EKA proves effective as a versatile, training-free inference strategy that scales seamlessly to large models. Generalization tests across diverse datasets and retrieval corpora confirm the robustness of our approach. Overall, EKA advances the state-of-the-art in iterative RAG systems while illuminating the critical interplay between structured reasoning and efficient exploration in reinforcement learning-augmented frameworks. The code is released at \href{https://github.com/yxzwang/EarlyKnowledgeAlignment}{Github}.

cs.CL↗

arXiv2025

Terahertz Signal Coverage Enhancement in Hall Scenarios Based on Single-Hop and Dual-Hop Reconfigurable Intelligent Surfaces

Terahertz (THz) communication offers ultra-high data rates and has emerged as a promising technology for future wireless networks. However, the inherently high free-space path loss of THz waves significantly limits the coverage range of THz communication systems. Therefore, extending the effective coverage area is a key challenge for the practical deployment of THz networks. Reconfigurable intelligent surfaces (RIS), which can dynamically manipulate electromagnetic wave propagation, provide a solution to enhance THz coverage. To investigate multi-RIS deployment scenarios, this work integrates an antenna array-based RIS model into the ray-tracing simulation platform. Using an indoor hall as a representative case study, the enhancement effects of single-hop and dual-hop RIS configurations on indoor signal coverage are evaluated under various deployment schemes. The developed framework offers valuable insights and design references for optimizing RIS-assisted indoor THz communication and coverage estimation.

eess.SP↗

arXiv2025

Condor: A Code Discriminator Integrating General Semantics with Code Details

LLMs demonstrate significant potential across various software engineering tasks. However, they still face challenges in generating correct code on the first attempt when addressing complex requirements. Introducing a discriminator to select reliable outputs from multiple generated results is an effective way to enhance their reliability and stability. Currently, these discriminators fall into two categories: execution-based discriminators and non-execution-based discriminators. Execution-based discriminators face flexibility challenges due to difficulties in obtaining test cases and security concerns, while non-execution-based discriminators, although more flexible, struggle to capture subtle differences in code details. To maintain flexibility while improving the model's ability to capture fine-grained code details, this paper proposes Condor. We first design contrastive learning to optimize the code representations of the base model, enabling it to reflect differences in code details. Then, we leverage intermediate data from the code modification process to further enrich the discriminator's training data, enhancing its ability to discern code details. Experimental results indicate that on the subtle code difference dataset (i.e., CodeNanoFix), Condor significantly outperforms other discriminators in discriminative performance: Condor (1.3B) improves the discriminative F1 score of DeepSeek-Coder (1.3B) from 67% to 73%. In discriminating LLM-generated outputs, Condor (1.3B) and Condor (110M) raise the Pass@1 score of Meta-Llama-3.1-Instruct (70B) on the CodeNanoFix dataset from 52.64% to 62.63% and 59.64%, respectively. Moreover, Condor demonstrates strong generalization capabilities on the APPS, MBPP, and LiveCodeBench datasets. For example, Condor (1.3B) improves the Pass@1 of Meta-Llama-3.1-Instruct (70B) on the APPS dataset by 147.05%.

cs.SE↗

arXiv2025

Grammar-Based Code Representation: Is It a Worthy Pursuit for LLMs?

Grammar serves as a cornerstone in programming languages and software engineering, providing frameworks to define the syntactic space and program structure. Existing research demonstrates the effectiveness of grammar-based code representations in small-scale models, showing their ability to reduce syntax errors and enhance performance. However, as language models scale to the billion level or beyond, syntax-level errors become rare, making it unclear whether grammar information still provides performance benefits. To explore this, we develop a series of billion-scale GrammarCoder models, incorporating grammar rules in the code generation process. Experiments on HumanEval (+) and MBPP (+) demonstrate a notable improvement in code generation accuracy. Further analysis shows that grammar-based representations enhance LLMs' ability to discern subtle code differences, reducing semantic errors caused by minor variations. These findings suggest that grammar-based code representations remain valuable even in billion-scale models, not only by maintaining syntax correctness but also by improving semantic differentiation.

cs.PL↗