Searcharxiv⌕ Search

arXiv subjects

Quanming Yao

Publications and source records attributed to Quanming Yao.

At least 19 recordsLinked to original sources
arXiv2026

QRAE: Query-Relevant Attention Entropy for Context Compression in DeepSearch Agents

DeepSearch LLM agents repeatedly search and carry retrieved sources into subsequent reasoning steps. This expanding context increases token consumption and latency, although many results do not contribute to later reasoning. We propose Query-Relevant Attention Entropy (QRAE), an online context compression method that uses sparse evidence-assessment heads to decide which intact retrieved chunks should persist across reasoning steps. It scores bounded candidate batches through paired prefills: a query-free control suppresses shared attention responses, while normalized entropy measures the breadth of the positive response within each chunk. High-scoring chunks enter a persistent evidence whitelist and are included in the reasoning context. Uncertain chunks remain outside the reasoning context but are retained in a candidate cache, where they can be retrieved for later reevaluation. QRAE requires neither an additional generative compression pass nor modification of the attention operator. Across multiple DeepSearch benchmarks and reasoning models, QRAE improves task success by an average of 2.35 percentage points and reduces end-to-end latency and reported token usage by an average of 30% and 45%, respectively. These benefits generalize to a model with nearly 100B parameters.

cs.LG↗
arXiv2026

TAGR: Temporally Adaptive Generative Recommendation for Industrial Live-Streaming Advertising

Live-streaming advertising is an important monetization channel on short-video and e-commerce platforms, where rapidly changing live content, promoted products, and user feedback impose strong freshness requirements on recommendation models. Existing generative recommenders designed for static domains fail at three levels: static semantic IDs (SID) cannot track evolving live ads; single-scale behavior modeling misses shifting intent; preference optimization conflicts between fresh on-policy feedback and training stability. We propose TAGR, a generative recommendation framework with temporal adaptation at three levels: live-ad tokenization, user intent modeling, and preference alignment. At the token level, Live Semantic-Collaborative ID (LSID) periodically refreshes each active ad's SID based on its current live scene and promoted products, while retaining a stable hierarchical token vocabulary for autoregressive generation. At the intent level, Intent-Aware Generation (IAG) models live-room entry histories at multiple temporal granularities as the primary intent sequence, keeps auxiliary behaviors as separate inputs, and weights next-token prediction (NTP) using post-request intent evidence and business value. At the alignment level, Intermittent On-Policy Preference Optimization (IOPO) periodically samples fresh candidate groups from the current policy and performs behavior- and value-aligned preference updates interleaved with supervised NTP maintenance to preserve learned behavior distribution. Deployed on a large-scale e-commerce live-stream advertising platform, TAGR improves live-room entry and shopping-cart click rates by 8.5% and 7.4%, respectively, and achieves a 16.1% revenue lift over the production baseline. These results demonstrate the effectiveness and industrial viability of temporally adaptive generative recommendation for live-stream advertising.

cs.IR↗
arXiv2026

Attention Sinks Induce Gradient Sinks: Massive Activations as Gradient Regulators in Transformers

Attention sinks and massive activations are recurring and closely related phenomena in Transformer models. Existing explanations have largely focused on the forward pass, yet in pre-norm Transformers, large residual-stream norms play only an indirect forward role because sublayers operate on normalized inputs. We study this relationship from the perspective of backpropagation. Empirically and theoretically, we show that under causal masking, attention sinks can induce pronounced gradient concentration, which we term gradient sinks. Since the RMSNorm Jacobian attenuates gradients roughly in inverse proportion to input norm, massive activations can be understood as adaptive regulators of this localized gradient pressure during training. This interpretation predicts that attenuating sink-induced gradients should weaken massive activations. We test this prediction with V-scale, a modification that adjusts backpropagated gradients on the value paths. In V-scale models and a backward-only control, attention sinks are preserved, whereas massive activations are suppressed. These results identify gradient sinks as a backward-pass counterpart of attention sinks, and massive activations as an adaptive RMSNorm-mediated response that attenuates the resulting localized training pressure. Our code is available at https://github.com/LARS-research/GradientSink.

cs.LG↗
arXiv2026

Agentic In-Context Learning: What Algorithms Can Transformers Learn in Closed-Loop Interaction?

The learning algorithm view of in-context learning (ICL) studies how fixed-parameter Transformers learn from supplied examples. Language model agents extend this setting through interaction: their actions determine what evidence enters context next. We call this regime Agentic In-Context Learning (Agentic ICL). Existing studies largely depend on externally supplied contexts or reward-bearing interaction histories, while work on general tool-using agents lacks a learning algorithm view. We represent an interactive learner by its state, update, and decision rule, covering active learning, bandits, and online reinforcement learning (RL). For expressivity, we construct causal Transformers that approximate these algorithms over a finite horizon, and bound the resulting loss in return. For training, we characterize when optimizing process rewards preserves behavior that maximizes the final outcome. Distillation experiments show that Transformers can learn to execute three representative recursions. We then train models from outcome or process rewards without oracle action or state targets, as in RL post-training. Different rewards yield distinct decision policies. Improving the alignment of process supervision with outcome-optimal action preferences shifts learned decisions accordingly. Such behavioral differences persist in a pretrained language model across textual paraphrases.

cs.AI↗
arXiv2026

FloodVLM-Pix: Language-Guided Pixel-Level Flood Understanding

As climate change intensifies, floods are becoming more frequent and destructive, making accurate flood understanding an urgent priority. However, existing vision-language models perform poorly on flood-related tasks, suffering from attentional bias and inaccurate descriptions of flood magnitude. To bridge this gap, we construct the first large-scale remote sensing flood multi-modal benchmark that supports language-level flood understanding and pixel-level flood segmentation. The benchmark contains 131,200 flood image patches and corresponding 443,100 textual descriptions for flood severity and evolution. Based on this benchmark, we propose FloodVLM-Pix, a language-guided pixel-level flood understanding model that includes a flood prior module and a multi-temporal flood pixel decoder to achieve accurate flood understanding. The flood prior module automatically distills sub-region flood spatial priors and inputs priors into FloodVLM-Pix, solving the attentional bias problem in flood language understanding. The flood pixel decoder introduces a novel general-specific feature fusion mechanism to address performance degradation caused by interference (e.g., clouds or mountain shadows) during flood segmentation. Experiments on FloodVL-Pix demonstrate that FloodVLM-Pix improves both flood description quality and pixel-level flood segmentation performance compared with state-of-the-art VLMs and flood segmentation models. Our code and data will be released at https://github.com/Haomiaoyanmeng/FloodPix.

cs.CV↗
arXiv2026

DeltaReplay: Task-Relative Memory Reuse for Mobile GUI Agents

Memory-augmented mobile GUI agents store successful execution trajectories and reuse them in later tasks, but a stored trajectory rarely matches a new task exactly. The new task may use different parameters, share only some of its steps with a stored trajectory, or have no relevant record in memory. Forcing the agent to use irrelevant memory can mislead it, whereas discarding memory that may still be useful deprives it of guidance from past experience. To address this dilemma, we propose DeltaReplay, a step-level memory reuse framework that decides how to use existing memory without modifying it. We observe that the reusable part of a stored record is determined not by the record itself but by its relation to the new task, mainly through two factors: page-level consistency and action-level generality. We therefore store execution trajectories as paths in a transition graph, whose nodes (pages) and edges (actions between pages) capture these two factors. At reuse time, the action on each edge is split into a task-independent operation and task-specific parameters. DeltaReplay then compares each recorded step with the new task and the current screen, and decides whether to follow it, execute it after replacing its parameters, or leave it to the base agent. On AndroidWorld and SPA-Bench, DeltaReplay improves the task success rate over a base agent with the same backbone by up to 10.3 and 25.0 percentage points, respectively. These results indicate that deciding at each step how to use retrieved memory lets agents benefit even from partially matching trajectories.

cs.AI↗
arXiv2026

RuleMem: Active Rule Memory for Long-Term Conversational Agents

Question answering agents in long-term conversations must reason over massive, temporally dispersed dialogue histories. However, existing memory mechanisms primarily treat past information as \textit{passively} stored facts, leading to semantic gaps and unreliable reasoning. To address this limitation, we propose RuleMem, a rule-based memory framework that induces reusable logical rules from historical interactions to \textit{actively} guide both evidence retrieval and reasoning. Specifically, RuleMem constructs natural-language Horn clauses from conversations and validates them via a Rule Perplexity Consistency (RPC) mechanism. These induced rules enable the retrieval of semantically distant evidence while providing an explicit logical structure for answer generation. We conducted a comprehensive evaluation of RuleMem on two long-term conversational benchmarks, LoCoMo and LongMemEval_s*. In a rigorous comparison against 14 baselines on LoCoMo, RuleMem achieved the highest accuracy, exceeding the baseline average by 27.47 points (a 54.3% relative improvement).

cs.CL↗
arXiv2026

MAS-on-the-Fly: In-Context Structural Adaptation of LLM-Based Multi-Agent Systems

Large Language Model (LLM)-based multi-agent systems (MAS) have emerged as a promising paradigm for complex tasks. However, existing works often rely on manual designs or "one-size-fits-all" automation and lack adaptability after deployment. We study in-context structural adaptation, where structured experience conditions both query-dependent system generation and execution-time reconfiguration without updating LLM parameters. We introduce MASFly, which realizes this adaptation through two complementary mechanisms. First, a retrieval-augmented SOP instantiation mechanism retrieves and adapts successful collaboration patterns to construct a query-specific MAS. Second, an experience-enhanced process supervision mechanism uses a dedicated Watcher agent to monitor execution against prior failure experience and reconfigure the system upon abnormal behavior. Experiments demonstrate that MASFly achieves state-ofthe-art performance, including a 61.7% success rate on TravelPlanner, with strong task adaptability and robustness.

cs.MA↗
arXiv2026

Fine-Tuning Low-Bit Models with Gradient in Quantized Code Space

Fine-tuning Low-bit models aims to adapt a quantized model while keeping the final deployed checkpoint in the same low-bit form. This setting is practically important as it reduces memory and inference cost for storage and deployment. Under this constraint, adaptation becomes an optimization problem over quantization codes and scales. Existing continuous low-bit training is efficient, but it can be distorted by straight through estimation error or by post-quantize gap; discrete search is deployment-faithful, but it is often too inefficient under a finite training budget. We propose code surrogate gradient as the first order signal in deployable code space to acceleate optimization, and performing guided search to preserve deployment faithfulness. Experiments across arithmetic reasoning, instruction following, and structured language understanding show that GradCodes consistently improves fine-tuning low-bit models across different quantization datatypes. Code is provided at https://github.com/ovo67/GradCodes.

cs.LG↗
arXiv2026

Falsifiable Commitment Planning for Self-Correcting Web Agents

Long-horizon web agents often go off track before final failure: a trajectory can remain locally plausible even after the current state, reused skill, or plan assumption no longer supports the user instruction. Existing agents can plan, reflect, or reuse experience, but their plans rarely specify the evidence under which an active step should still be trusted. We propose FCPAgent, a falsifiable commitment planning framework for robust long-horizon web agents. FCPAgent represents each plan step as a Falsifiable Commitment Unit (FCU): a subgoal grounded in a reusable skill, together with confirming evidence, falsifying evidence, and a confidence score. Execution is organized as a plan-test-repair loop. The hybrid commitment testing module checks candidate actions before they modify the browser and checks observations after execution; for efficiency, it combines lightweight evidence matching with LLM-based diagnostic verification. When evidence falsifies a commitment, scope-aware repair localizes the contradiction to the execution, skill, or planning level and revises the smallest adequate part. On WebArena, FCPAgent achieves a 13.8% relative improvement in average success over the strongest baseline, with especially large gains on long-horizon tasks.

cs.AI↗
arXiv2026

Mechanistic Attention Guidance for Agent Memory Refinement

Existing self-evolving memory systems mainly improve agent memory based on textual outputs, such as task trajectories and reflections. However, this text-based paradigm rarely incorporates internal mechanistic signals, leaving how retrieved memory is actually utilized during task execution underexplored. This limitation can lead to unreliable error attribution and hallucinated memory modifications. In this work, we show that retrieval-head attention provides a mechanistic signal for revealing segment-level memory utilization. By aggregating attention over memory segments and decision steps, we construct a context utilization matrix that exposes recurring memory-use patterns and indicates corresponding refinement strategies. Building on this observation, we propose Attention-Guided Memory Refinement (AGMR), a framework that uses utilization patterns revealed by attention to guide targeted segment-level memory updates. AGMR corrects or enhances memory for failed executions, simplifies memory for successful executions, and verifies each update through re-execution. Experiments on interactive decision-making benchmarks show that AGMR improves both task performance and memory efficiency over text-only memory refinement baselines. Code is available at https://anonymous.4open.science/r/AGMR_code-3262/

cs.AI↗
arXiv2026

Graph Unitary Message Passing

Unitarity is a useful principle for stabilizing deep neural networks, but in graph neural networks (GNNs) instability is induced not only by learnable parameters but also by the graph propagation operator. Motivated by this distinction, we propose Graph Unitary Message Passing (GUMP), a message-passing framework that uses a unitary propagation operator on a transformed graph to avoid graph-induced exponential decay under repeated propagation. GUMP combines (i) a graph transformation that maps an input graph to an Eulerian line-graph construction admitting unitary adjacency matrices, and (ii) a practical unitary projection procedure based on Newton-Schulz iteration. Theoretical analysis clarifies that, under standard analysis assumptions, unitary propagation keeps the graph-propagation term depth-stable, while vanilla normalized propagation exhibits exponential decay in its non-trivial spectral components. Across synthetic long-range tasks, TUDataset benchmarks, and LRGB datasets, GUMP improves over vanilla message passing and achieves competitive or superior performance against strong baselines. Code is available at https://github.com/ucker/gump_code.

cs.LG↗
arXiv2026

DDIAgents: Mechanism-Conditioned Context Flow for Drug-Drug Interaction Prediction

Drug-drug interaction (DDI) prediction is essential for medication safety, yet it requires reasoning over heterogeneous biomedical evidence whose relevance changes across interaction mechanisms. We propose DDIAgents, a mechanism-conditioned multi-agent framework that performs DDI prediction through dynamic knowledge orchestration. Given a drug pair, a planner agent instantiates specialized expert agents, routes mechanism-relevant knowledge sources to each agent, and aggregates their analyses through a conclusion agent. By adapting context flow to the inferred interaction mechanism, DDIAgents reduces irrelevant information, supports complementary expert reasoning, and produces interpretable agent-level rationales. Extensive experiments on realistic DDI prediction benchmarks show that DDIAgents consistently outperforms existing feature-based, graph-based, LLM-based, and agent-based baselines. Beyond prediction performance, DDIAgents demonstrates how multi-agent systems can organize heterogeneous scientific knowledge for adaptive and interpretable AI4Science reasoning.

cs.AI↗
arXiv2026

Why Low-Precision Transformer Training Fails: An Analysis on Flash Attention

The pursuit of computational efficiency has driven the adoption of low-precision formats for training transformer models. However, this progress is often hindered by notorious training instabilities. This paper provides the first mechanistic explanation for a long-standing and unresolved failure case where training with flash attention in low-precision settings leads to catastrophic loss explosion. Our in-depth analysis reveals that the failure is not a random artifact but caused by two intertwined phenomena: the emergence of similar low-rank representations within the attention mechanism and the compounding effect of biased rounding errors inherent in low-precision arithmetic. We demonstrate how these factors create a vicious cycle of error accumulation that corrupts weight updates, ultimately derailing the training dynamics. To validate our findings, we introduce a minimal modification to the flash attention that mitigates the bias in rounding errors. This simple change stabilizes the training process, confirming our analysis and offering a practical solution to this persistent problem. Code is available at https://github.com/ucker/why-low-precision-training-fails.

cs.LG↗
arXiv2026

Language Model Networks: Supervision-Efficient Learning through Dense Communication

Language models are increasingly used not only as standalone predictors but also as components in larger inference systems, from test-time scaling to multi-agent collaboration. We study language model networks, where pre-trained language models serve as reusable nodes and intelligence emerges from their topology, communication, and optimization. Existing systems mostly communicate through natural language: easy to deploy, but discrete, inefficient, and hard to optimize from end-task supervision. We propose LMNet, a dense and differentiable realization of this paradigm. LMNet uses stripped LLMs as vertex modules and trainable seq2seq modules as communication edges, enabling intermediate nodes to exchange dense vectors while preserving natural-language input and output at the system boundary. By bypassing intermediate embedding and de-embedding, LMNet enables efficient information transfer, end-to-end gradient optimization, and learned communication beyond hand-designed protocols. Experiments show performance with small additional training cost and effective adaptation under limited supervision.

cs.AI↗
arXiv2026

Scaling GraphLLM with Bilevel-Optimized Sparse Querying

LLMs have recently shown strong potential in enhancing node-level tasks on text-attributed graphs (TAGs) by providing explanation features. However, their practical use is severely limited by the high computational and monetary cost of repeated LLM queries. To illustrate, naively generating explanations for all nodes on a medium-sized benchmark like Photo (48k nodes) using a representative method (e.g., TAPE) would consume days of processing time. In this paper, we propose Bilevel-Optimized Sparse Querying (BOSQ), a general framework that selectively leverages LLM-derived explanation features to enhance performance on node-level tasks on TAGs. We design an adaptive sparse querying strategy that selectively decides when to invoke LLMs, avoiding redundant or low-gain queries and significantly reducing computation overhead. Extensive experiments on six real-world TAG datasets involving two types of node-level tasks demonstrate that BOSQ runs substantially faster than existing GraphLLM methods while consistently delivering on-par or superior performance.

cs.DB↗
arXiv2026

Asking LLMs to Verify First is Almost Free Lunch

To enhance the reasoning capabilities of Large Language Models (LLMs) without high costs of training, nor extensive test-time sampling, we introduce Verification-First (VF), a strategy that prompts models to verify a provided candidate answer, even a trivial or random one, before generating a solution. This approach triggers a "reverse reasoning" process complementary to standard forward Chain-of-Thought (CoT), which restricts the logical search space of the answer by pruning the LLM's output distribution. We further generalize VF prompting to Iter-VF, a sequential test-time scaling (TTS) method that iteratively cycles the verification-generation process using the model's previous answer. Extensive experiments across various benchmarks and various LLMs confirm that VF prompting with random answer consistently outperforms standard CoT with minimal computational overhead, and Iter-VF outperforms existing TTS strategies. VF is also effective on SOTA thinking models. For example, by using the simple VF prompting, we obtain a new SOTA 94.9% accuracy on GPQA-Diamond with Gemini-3-Pro-Preview where VF reduces its errors by ~30% relatively.

cs.CL↗
arXiv2026

ContextFlow: Hierarchical Task-State Alignment for Long-Horizon Embodied Agents

Long-horizon embodied agents increasingly delegate navigation, search, approach, and manipulation to specialist executors. As these executors become stronger, the main bottleneck shifts from local skill execution to maintaining a coherent task frontier across planning, monitoring, memory, and execution. We study task-state misalignment, a task-level consistency failure in which the planner's active stage, runtime evidence, remembered context, and delegated executor no longer justify the same next-step decision. This failure can lead to unsupported handoffs, stage lock, executor-context mismatch, and unnecessary replanning. We propose ContextFlow, an inspectable alignment framework that represents stages as explicit contracts, converts runtime observations into evidence packets, and applies scoped updates including continue, refine, transfer, promote, and repair. ContextFlow keeps specialist executors responsible for local closed-loop control while making task-frontier alignment explicit and auditable. Experiments and demonstration traces on long-horizon embodied tasks illustrate how evidence-grounded scoped updates diagnose and mitigate recurring task-state failures.

cs.RO↗