Searcharxiv⌕ Search

arXiv subjects

Tao Feng

Publications and source records attributed to Tao Feng.

At least 19 recordsLinked to original sources
arXiv2026

GUI-AC: Enhancing Continual Learning in GUI Agents

Graphical User Interfaces (GUIs) serve as the dominant medium for human-computer interaction, yet building GUI agents that generalize across the vast diversity of real-world interface environments, with the same flexibility and robustness that humans naturally exhibit, remains unsolved. Notably, GUI data are inherently non-stationary: the continual emergence of previously unseen interface instances (e.g., novel domains and resolutions) induces persistent distribution shifts, significantly impeding the continual learning of existing GUI agents. Reinforcement fine-tuning (RFT) has attracted considerable attention as a promising approach. Nevertheless, RFT exhibits pronounced instability in its grounding capability, where rollout outcomes are highly imbalanced and fixed clipping can suppress valuable exploration. More importantly, policy probability alone does not reveal the semantic role of a grounding prediction. To address these challenges, we propose GUI-AC, a continual RFT framework that jointly characterizes grounding rollouts through advantage and policy probability. This yields a four-region view of grounding behavior and enables targeted optimization of Exploration and Overconfidence. GUI-AC further introduces grounding certainty to support two core mechanisms: (i) Adaptive Advantage, which strengthens the negative correction for overconfident grounding failures; and (ii) Dynamic Clipping, which relaxes the upper clipping bound for exploratory grounding rollouts. Extensive experiments show that GUI-AC consistently outperforms state-of-the-art baselines. Code is available at https://anonymous.4open.science/r/GUI-AC-Code.

cs.CV↗
arXiv2026

Primitive flag-transitive generalized quadrangles

Let $\mathcal{Q}$ be a finite thick generalized quadrangle of order $(s,t)$ with a flag-transitive automorphism group $G$. A well-known conjecture of Kantor from the early 1990s asserts that up to duality, $\mathcal{Q}$ is either of classical type, or it is one of two sporadic examples of order $(3,5)$ or $(15,17)$. Although this fundamental problem in finite geometry remains open, it has been resolved in some special cases. For example, a theorem of Buekenhout and Van Maldeghem from 1994 proves Kantor's conjecture under the stronger assumption that $G$ acts distance-transitively on the set of points of $\mathcal{Q}$. In this paper, we study Kantor's conjecture in the setting where $G$ acts primitively on both the points and lines of $\mathcal{Q}$. Here a theorem of Bamberg et al. implies that $G$ is almost simple with socle $X$ a finite simple group of Lie type, and several special cases where the Lie rank of $X$ is at most $2$ have recently been settled in a sequence of papers. Our main theorem resolves the problem for all almost simple groups of Lie type, thereby completing the proof of Kantor's conjecture for generalized quadrangles with a flag-transitive automorphism group acting primitively on points and lines.

math.CO↗
arXiv2026

LRanker: Scaling LLM Ranking to Millions of Candidates with Test-Time Scaling

Large language models (LLMs) offer strong semantic representations for ranking, but processing candidate texts within limited contexts and repeatedly invoking an LLM for candidate-level comparisons can be costly when candidate pools contain millions of items. We propose LRanker, a framework for scaling LLM ranking to millions of candidates with test-time scaling. The framework combines two complementary components. A candidate aggregation encoder summarizes offline candidate embeddings using K-means clustering and conditions the LLM on compact global candidate information; ranking then uses inner-product scores with cached candidate embeddings. A graph-based test-time scaling mechanism allocates additional inference computation to partition-conditioned query embeddings and ensembles their ranking scores. This separates the ability to score large candidate pools from the use of additional test-time computation to improve ranking quality. We evaluate LRanker on seven tasks across three scenarios in RBench, including a recommendation task with more than 6.8M candidates. Against the strongest baseline reported for each task, the relative MRR improvements in RBench-Large are 27.5%, 3.9%, 0.9%, and 8.9% on Rec-Movie, Rec-Toy, MS MARCO, and ESCI, respectively. On RBench-Small, the reported task scores improve by 26.7% on Rec-Music and 6.1% on Routing-Balance; on the 6.8M-candidate Rec-Clothing task, the improvement over Tiger is 27.0%. Ablations support the complementary contributions of candidate aggregation and test-time ensembling. These results demonstrate a practical approach to million-scale LLM ranking, with test-time computation providing an additional means of improving ranking effectiveness.

cs.IR↗
arXiv2026

MemComposer: Learning to Assemble Memories for LLM Agents via Reinforcement Learning

Agent memory enables large language model (LLM) agents to reuse past experiences without updating their parameters. However, semantic similarity alone does not guarantee memory usefulness: downstream performance depends on how memories are selected and presented to the agent. We propose MemComposer, a memory composer that jointly examines the current query and retrieved candidates to decide which experiences to include, how many to select, and how to order them. We train the composer using Experience Policy Optimization (ExPO), a GRPO-based framework that evaluates assembled memory contexts through complete agent interactions and uses task-success feedback to optimize the assembly policy. Across three $τ^2$-bench domains with two frozen executors, MemComposer outperforms the strongest baseline for each executor while reducing average memory-token usage by 16.9% and 4.9%, respectively. The learned policy also transfers to unseen domains, executors, and memory banks, and benefits from memories generated by stronger models without further training. These results demonstrate the value of learning query-specific memory assembly using downstream task feedback.

cs.AI↗
arXiv2026

Asymptotic behavior of solutions to critical Hartree equations with isolated singularities

We investigate the qualitative properties of positive classical solutions $u$ to critical Hartree equations on punctured domains, assuming that $u(x)\to+\infty$ as $x\to0$. For the global problem set in $\mathbb R^n\setminus\{0\}$, we prove that such solutions are radial and nonincreasing in $|x|$. For the local problem in $B_R\setminus\{0\}$, we establish the upper bound $u(x)\leqslant C|x|^{-(n-2)/2}$ near the origin and the asymptotic radial symmetry estimate \begin{equation*} u(x)=\bar u(|x|)\bigl(1+\mathcal O(|x|)\bigr)\qquad\text{as }x\to0, \end{equation*} where $\bar u$ is the spherical average of $u$. Our approach combines integral representations, blow-up analysis, and the method of moving spheres. A main difficulty is controlling the nonlocal Hartree potential under Kelvin inversion and in the blow-up limit. We also give an alternative proof of the classification of positive entire classical solutions in $\mathbb{R}^{n}$, which is used to identify the blow-up profile.

math.AP↗
arXiv2026

AURAL: Adaptive Latent Reasoning with Joint Chunk for Speech Language Models

Model intelligence and fast response jointly shape the quality of interaction with speech language models, yet remain difficult to achieve together. Explicit chain-of-thought (CoT) improves reasoning and audio understanding, but generating intermediate reasoning tokens delays responses. Describing fine-grained acoustic cues further lengthens CoT and increases latency. Latent reasoning can reduce this overhead, yet existing methods often trail CoT and remain limited by single-path supervision and reasoning budgets that do not adapt to problem difficulty. We introduce AURAL, which models a distribution over multiple plausible reasoning continuations in latent space and jointly predicts chunks of future states to reduce sequential forward passes and reasoning latency. To provide initial supervision for latent reasoning, we construct AuralReason-683K: 683K bilingual speech utterances (about 1,000 hours) with concise CoT for emotion recognition, empathetic dialogue, and general reasoning. AURAL-RL then explores beyond these traces, rewarding concise reasoning that yields high-quality answers and adapting reasoning effort to each problem. Across two backbones, AURAL-RL achieves performance comparable to CoT-RL, with larger gains over the respective supervised checkpoints on most metrics. Analysis further shows that harder questions elicit more latent reasoning steps. On Qwen2.5-Omni, it reduces time to the first answer token by 11.8x, from 1.22 to 0.10 s, versus 0.05 s for direct answering.

cs.CL↗
arXiv2026

RouteWeaver: Weaving Mode Selection and Execution into Unified LLM Routing

LLM routing dynamically selects and coordinates heterogeneous language models across diverse tasks. However, existing approaches are largely built around a predefined execution paradigm, such as single-round, multi-round, or agentic routing, which fixes how model calls are organized and limits adaptation to different query needs. We propose RouteWeaver, a unified routing framework that integrates these routing modes within a single LLM policy, jointly learning which mode to use and how to select models within that mode. To jointly optimize mode selection and within-mode execution, we develop COMET-GRPO (COordinated Mode-Execution Training), which decouples the trajectory-level advantage into separate signals for mode selection and within-mode execution, enabling more targeted credit assignment to each decision level. We further introduce progressive exploration, which gradually shifts training from externally assigned modes to autonomous selection, so that the router first learns how to execute each mode before deciding when to use it. We evaluate RouteWeaver across mathematics, coding, knowledge, logical reasoning, and factual recall. RouteWeaver achieves an average accuracy of 61.4%, an 11.03% relative improvement over the strongest baseline, and outperforms all compared baselines in each of the five domains. Ablations show that both COMET-GRPO and progressive exploration are important for adaptive routing; without progressive exploration, the routing policy collapses entirely to single-round execution.

cs.CL↗
arXiv2026

The existence of symmetric cyclic Steiner quadruple systems

A Steiner quadruple system of order $v$, denoted by $\operatorname{SQS}(v)$, is called cyclic if it admits $\mathbb Z_v$ as a group of automorphisms acting regularly on its point set. A cyclic system is called symmetric cyclic, or S-cyclic, if the multiplier $-1$ preserves each block orbit individually. Grannell and Griggs conjectured in 1983 that such a system of order $v$ exists if and only if $v\equiv2,4,10,20\pmod{24}$ and every odd prime divisor $p$ of $v$ satisfies $p\equiv1$ or $5\pmod{12}$. This paper proves the conjecture. The main step is to construct an $S$-cyclic Steiner quadruple system of order $2p$ for every prime $p\equiv1\pmod4$. Together with existing reduction theorems, these constructions determine which finite abelian groups admit a symmetric group-invariant Steiner quadruple system.

math.CO↗
arXiv2026

Design Anything, Design at Scale

As multimodal resources are transformed into media outputs, agentic AI is expected to stay aligned with design intent over long horizons. Designers follow briefs specifying goals, constraints, and deliverables, with later work building on earlier decisions. Recent work reuses design experience across tasks, yet alignment among artifacts within one project remains challenging. Agents therefore need to preserve and apply this intent across domains and as projects grow. We introduce Design Anything, a graph-based harness that preserves design intent outside the model and binds deliverables to the workflow steps that produce them through an intent-execution graph. A task contract retains constraints and decisions, selected designs become explicit inputs to later generations, and a deterministic controller uses bindings and execution records to check output types and counts, identify outstanding deliverables, and guide subsequent turns. We also introduce DesignAnything-Bench, with 228 tasks across 19 design domains, including 38 project families at scales of 6, 14, and 24 artifacts. With the LLM held fixed, Design Anything satisfies 85.0% and 92.5% of the brief's factual constraints with DeepSeek V4.1 Flash High and GPT-5.6 Sol High, respectively, exceeding the strongest same-model baselines by 15.4% and 14.4%. These gains reach 25.1% and 23.0% on 24-artifact tasks. Under each model, Design Anything leads or ties in constraint satisfaction in 18 of 19 domains. Professional designers also prefer its complete deliverable sets over those of same-model baselines in a blind evaluation.

cs.CV↗
arXiv2026

JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs

Large language model (LLM) agents now fix software, operate computers and serve customers, but their harnesses, the programs that run them, call a large LLM at every step, which makes long tasks slow and costly. Most of these calls end in a multiple-choice decision, such as which tool to call next. Decision models such as Jev answer these questions directly, at a fraction of an LLM call's time and price. We ask how a harness can delegate these decisions to a decision model with little loss of success. To evaluate decision models in agentic tasks, we build AgentDecisionBench, which tests them from isolated questions up to control of a running agent, on public benchmarks and our DecisionGym environment. The decision model most accurate on isolated questions solved only 69.5% of agent tasks, and without a fitted confidence threshold, delegating tool choices lowered success by 19.0 points. Building on these findings, we then design JevAgentHarness, which fits this threshold for each choice on the agent's own trajectories and lets the LLM decide below it. Because existing decision models never see the states an agent reaches after its own mistakes, we train AgentDecision-4B, an open 4B model, on such states with Jev's answers as labels. On tasks unlike its training data, the agent solved 91.5% of them with AgentDecision-4B and 82.0% with Jev. On tasks like their training data, 4B models trained this way made agents 81.1-89.0% cheaper and up to 2.5 times faster, without losing 3 or more points of task success. The models, data, and harness will be released at https://github.com/JevAgentHarness/JevAgentHarness.

cs.AI↗
arXiv2026

T2J-BENCH: Benchmarking Coding Agents on Execution-Grounded PyTorch-to-JAX Repository Migration

Migrating a machine-learning codebase across frameworks--for example from PyTorch to JAX--is a routine but costly engineering task that is fundamentally cross-paradigm: eager, stateful, object-oriented modules must be rewritten as pure, jittable functions over an explicit parameter pytree. We ask whether coding agents can perform such whole-repository migrations correctly, and argue that existing code benchmarks, which grade unit-test pass rates or surface-level similarity to a reference translation, cannot answer this: they neither require the migrated system to \emph{behave} the same nor prevent a JAX reimplementation from faking equivalence by secretly importing PyTorch itself. We introduce T2J-BENCH, a benchmark of 63 whole-repository PyTorch-to-JAX migration tasks spanning 18 model families across both training and serving, where torch-free checkpoint conversion is itself part of the deliverable. T2J-BENCH grades execution-grounded functional equivalence: each candidate is run inside a separate, torch-free, offline verifier and scored against references frozen from the PyTorch original, under a hard independence gate and a continuous, threshold-free reward. Across twenty off-the-shelf agent--model configurations, the strongest recovers only 0.676 of the reward against a 0.980 oracle, with substantial numeric and behavioral divergence despite high spec pass rates. Finally, we contribute TWINHARNESS, a reference-grounded agent harness that turns the agent's executable PyTorch source into legitimate self-verification feedback; applied to an opus-5 agent it raises the reward from 0.649 to 0.723--the highest of any configuration--using only task-visible resources.

cs.SE↗
arXiv2026

Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent Memory

Multimodal models and agents have emerged to enable AI systems to learn from and interact with the real world more like humans by integrating information across multiple modalities. However, enabling multimodal agents to continually acquire useful information from complex, heterogeneous multimodal data remains a significant challenge as they move toward long-horizon tasks and lifelong settings. Addressing this challenge requires effective memory mechanisms that can maintain continuity across sessions, adapt to changing users and environments, and refine memory-management policies through accumulated experience. We introduce a two-axis analytical framework for multimodal agent memory structured around two key questions: how multimodal information is formed and represented (Axis I, Formation and Representation), and how its management adapts with accumulated experience (Axis II, Memory Management). Axis I, Formation and Representation, characterizes what information is retained and how it is encoded (e.g., modality-specific embeddings, textual and structured records, latent and parametric representations) to support subsequent reasoning and action. Axis II, Memory Management, distinguishes three tiers: T1 Static Memory reuses fixed memory; T2 Dynamic Memory updates memory through a stable policy; and T3 Self-Evolving Memory updates memory through an evolving policy. Based on this framework, we synthesize recurring design patterns and derive principles for balancing source fidelity with abstraction, maintaining temporal validity and provenance, grounding memory use in recoverable evidence, and evolving memory policies over time. Furthermore, we discuss the challenges and potential future directions for the memory mechanism of multimodal agents.

cs.AI↗
arXiv2026

Complete mappings stabilizing Sylow $2$-subgroups in central quotients of special linear groups

A complete mapping of a finite group $G$ is a permutation $θ$ of $G$ such that $x\mapsto xθ(x)$ is also a permutation of $G$. Complete mappings have applications in cryptography, particularly in the design of round functions for cryptographic systems. The study of their parity has led to the question of which finite groups admit complete mappings of both parities. A useful approach to this question is to construct complete mappings stabilizing suitable subgroups, in particular Sylow $2$-subgroups. We show that, for every $n\geq2$, every prime power $q$, and every $Z\leq Z(\mathrm{SL}_n(q))$, the group $\mathrm{SL}_n(q)/Z$ admits a complete mapping stabilizing a Sylow $2$-subgroup if and only if $(n,q)\neq(2,2)$.

math.GR↗
arXiv2026

$\mathrm{S}^3$-PnP: Plug-and-Play Denoising for Scientific Imaging with Self-Supervised Score Priors

Modern image restoration methods increasingly rely on large-scale supervised training, yet clean targets are often prohibitively difficult to obtain in scientific imaging modalities such as hyperspectral imaging (HSI) and magnetic resonance imaging (MRI). Plug-and-play ADMM (PnP-ADMM) with pretrained diffusion priors offers a promising alternative, but the learned prior distribution can mismatch both the target scientific image distribution and the distribution of ADMM denoiser inputs induced by dual feedback. To mitigate these mismatches, we propose $\mathrm{S}^3$-PnP, a self-supervised plug-and-play restoration framework that learns a score prior directly from observation-derived ADMM intermediate states, without clean training targets or externally pretrained weights. Specifically, the learned score corrects the input to the analytical prior update, combining observation-driven corrections with analytical regularization. We further analyze the convergence and reconstruction error of the corrected ADMM algorithm assuming an ideal score and quadratic data fidelity. Experiments on reference-guided HSI restoration, single-image HSI denoising, and MRI denoising demonstrate consistent improvements over competing methods.

cs.CV↗
arXiv2026

SERL-SQL: Selective Hindsight Distillation for Text-to-SQL Reinforcement Agentic Learning

Recent Text-to-SQL systems increasingly rely on multi-turn interaction, execution feedback, and reinforcement learning. However, most existing methods use execution correctness only as a trajectory-level reward, which provides limited guidance for identifying the SQL decisions responsible for success or failure. We propose SERL-SQL, a selective execution-grounded reinforcement learning framework for multi-turn Text-to-SQL agents. SERL-SQL samples on-policy SQL interaction trajectories and uses a training-only teacher to re-score student actions with execution feedback. The resulting teacher--student likelihood gap is converted into bounded, masked weights that reweight GRPO advantages only on SQL and tool-action tokens. In this way, task rewards preserve the optimization direction, while execution hindsight provides localized credit assignment. Experiments on BIRD, Spider, and cross-domain benchmarks show that SERL-SQL achieves competitive performance, reaching 76.56% execution accuracy on BIRD-Dev and 89.92% on Spider-Test. Moreover, our reward-based selection strategy closely approaches the oracle Best-of-N upper bound and consistently outperforms consistency-based selection, showing that SERL-SQL produces high-quality candidates that can be reliably identified by lightweight execution-grounded rewards. Our code will be released at https://github.com/Ffunkytao/SERL-SQL.

cs.CL↗
arXiv2026

SpanVLA: Learning from Negative-Recovery Samples with Fast Action Bridging for Vision-Language-Action Model

Vision-Language-Action (VLA) models bring rich world knowledge and reasoning capabilities to autonomous driving. However, most driving VLAs are trained only on positive expert demonstrations, which teach models what to do but provide limited supervision on which behaviors to avoid and how to recover from these failures, leading to limited robustness. We introduce SpanVLA, a robust and efficient driving VLA, together with nuReasoning-NR, the first real-world driving dataset and benchmark for learning from negative and recovery samples. An extension of nuReasoning, it comprises 36K reasoning-intensive driving scenarios, including 3K suboptimal trajectories and 3K expert recovery trajectories. To exploit their asymmetric supervision, we introduce Negative--Recovery Reinforcement Fine-Tuning (NR-RFT), which penalizes undesirable behaviors and rewards expert corrective actions. Because challenging scenarios produce rollout groups dominated by similar suboptimal or even infeasible trajectories, it further incorporates an advantage correction to preserve informative optimization signals and encourage exploration. For efficient action generation, SpanVLA retains autoregressive reasoning and introduces a lightweight flow-matching action bridge conditioned on sparse-layer KV caches and initialized from the historical trajectory. Extensive experiments on NAVSIM v1/v2, nuReasoning, and the nuReasoning-NR benchmark demonstrate strong and robust planning performance and efficient action generation. Qualitative results across diverse challenging scenarios further highlight its planning robustness and recovery capability. The dataset and benchmark code will be released to support future research.

cs.CV↗
arXiv2026

UP-MOPD: Update Projection in Multi-Teacher On-Policy Distillation

On-policy distillation from multiple teachers combines expertise from different domains in a single student, but conflicting gradients can hinder this integration. Gradient corrections directly constrain parameter updates under plain SGD. With optimizers such as AdamW, however, momentum, adaptive scaling, and weight decay can turn a corrected gradient into an update that increases a domain loss to first order. To address this gap, we propose Update Projection for Multi-Teacher On-Policy Distillation (UP-MOPD). UP-MOPD lets the original mixed gradient update the optimizer state and generate a candidate displacement, then projects only violating candidates before they are committed to the parameters. The projection gives the unique feasible update closest to the candidate in Euclidean distance. In experiments combining medical and general domains, UP-MOPD improves IFEval-loose accuracy late in training by 2.96 points over vanilla M-OPD. It achieves an average score of 60.03 across eight metrics, compared with 59.00 for gradient projection and 59.15 for update rejection. On a public benchmark covering mathematics, code, and instruction following, it achieves the best average across six tasks (32.67), leads on LiveCodeBench v5, and ties for the best IFEval result.These results support projecting optimizer updates to reduce interference between domains.

cs.CV↗
arXiv2026

MambaVF: State Space Model for Efficient Video Fusion

Video fusion aims to integrate complementary information from multiple source videos while preserving temporal consistency. Effective modeling of temporal dynamics is essential to this goal, yet existing methods incur substantial computational overhead from optical flow estimation and feature warping. In this paper, we present MambaVF, an efficient video fusion framework that uses state space model (SSM) to achieve temporal modeling without explicit motion estimation. First, by formulating video fusion as a sequential state update process, MambaVF captures long-range temporal dependencies with linear complexity, significantly reducing computation and memory costs. Second, the lightweight SSM-based fusion module eliminates conventional flow-guided alignment. Instead, it introduces a mutual state fusion module and a spatio-temporal bidirectional scanning mechanism to enable information aggregation across video streams. Experiments on multiple benchmarks confirm that MambaVF reaches state-of-the-art performance in different video fusion applications (multi-exposure, multi-focus, infrared-visible, medical), while reducing parameters by >90% and FLOPs by >80%, resulting in >50% shorter runtime. Project page: https://mambavf.github.io

cs.CV↗