Searcharxiv⌕ Search

arXiv subjects

Ce Zhang

Publications and source records attributed to Ce Zhang.

At least 19 recordsLinked to original sources

RAG in a Trenchcoat: When Minimal Memory Is Enough for Agentic Systems, and When It Isn't

Agentic memory research has accumulated extraction, supersession, reflection, and graph synthesis on top of retrieval-augmented generation. However, we found that almost none of those primitives earn their cost in our tested memory benchmarks: we show this with BRIDGE, a deliberately minimal RAG memory with chained observation notes, hybrid BM25 with cosine retrieval, and chronological rendering, which reaches parity within bootstrap CI with the prior SOTA on LongMemEval-S and BEAM, and beats the best memory baseline on EverMemBench by 5.7-6.3 points at matched reader. It runs at ~5-7K reader-prompt tokens per query, using a single LLM call for each ingestion or query. We argue that this parity is itself the result: standard benchmarks score what memory retrieves, not what memory enables when the agent acts. To test the latter, we introduce five adversarial probes (Rule Application +-, Pattern Induction Style / Procedural, Dynamic Multiturn) judged at generation time by a cross vendor verification chain, where each query can only be solved by applying a remembered rule or pattern it never mentions. Across 5 probes x 20 scenarios, no architecture wins universally: atomic fact extractors (LangMem) win rule application, while BRIDGE's block-summary memory wins pattern induction. This result holds across four domains, three graders, and three reader families. Each design fails where the other wins, for reasons a recall benchmark cannot see: atomic-fact extractors discard the form a demonstrated pattern takes, while block-summary memory retrieves the rule but the reader does not act on it, a gap that presenting the rule more prominently largely closes. What separates memory designs is not what they retrieve but what they enable an agent to do, and that is what agentic memory should be evaluated on.

cs.AI↗

Singularity-Bench: Can Agents Build Agents?

Frontier AI agents increasingly succeed on realistic, long-horizon tasks and are beginning to improve the systems through which they operate. Yet existing benchmarks often restrict agents to narrow action spaces, use simplified environments, or rely on verification methods that are unreliable for evaluating arbitrary agent-generated systems. We introduce Singularity-Bench, an open-ended benchmark for autonomous AI development with six end-to-end tasks spanning pre-training, mid-training, post-training, inference, the agent harness, and cluster orchestration. Each task requires agents to implement core systems from a minimal starting point and evaluates them against mature human-engineered baselines on identical hardware. Additionally, through extensive red-teaming covering 9 of the 10 categories in the OWASP Top 10 for Agentic Applications, we expose reward-hacking vulnerabilities in the conventional Harbor framework and develop Breakwater, an extension of Harbor that blocks all attacks tested in our study. Across 11 agents built from 7 frontier models and 2 harnesses, Opus 5.5 with Claude Code matches or outperforms the human baseline on all six tasks with a Singularity-Index of 1.25, suggesting it may have the skills needed to reach the singularity. The remaining agents average only 0.60. Our vision is for Singularity-Bench to track how much of AI development agents can already automate and when they may transition from assisting to taking end-to-end responsibility for building stronger successors.

cs.AI↗

CANTON-$μ$ Proposal: A Next-Generation Muon $g-2$ Measurement at Sub-0.1 ppm Precision

We propose a next-generation precision measurement of the muon anomalous magnetic moment (muon g-2) at the High Intensity Heavy-Ion Accelerator Facility (HIAF) in Huizhou, China. We refer to this proposed experimental programme as CANTON-$μ$ (Coherent Anomalous magNetic momenT ObservatioN with muon). HIAF's intense, pulsed GeV-scale muon beams, particularly for negative muons, provide a promising basis for this programme. Building on two previously proposed storage-ring concepts, this work develops HIAF-specific experimental schemes that relax the conventional magic-momentum constraint and allow greater flexibility in the choice of beam momentum. We assess the expected muon intensity at HIAF and the corresponding statistical sensitivity. For each scheme, we identify its distinctive systematic effects, examine feasible control strategies, and propose quantitative systematic-uncertainty targets. Together, the statistical projections and systematic-uncertainty targets indicate prospective total precisions of 0.1 ppm in Phase-I, comparable to the current Fermilab precision, with the potential to reach the 0.05 ppm level in Phase-II following the planned HIAF upgrade. At the ultimate projected precision, the measurement would provide a stringent test of the Standard Model and probe new physics at multi-TeV scales. A focus on negative-muon measurements would enable direct comparison with existing high-precision positive-muon results and strengthen tests of CPT symmetry in the muon sector within the Standard-Model Extension, with a projected sensitivity at the $10^{-24}$ GeV level, an order of magnitude beyond current limits.

hep-ex↗

Searching Videos as Trees: Self-Correcting Agents for Grounded Long Video QA

Grounded long-video question answering (Grounded LVQA) requires answering a question about a long video while localizing the short evidence interval that supports the answer. Recent agentic methods frame this task as multi-turn exploration with a single crop_video(start, end) action, which supports coarse-to-fine narrowing but provides no primitive for fine-to-coarse backtracking. As a result, these agents typically converge prematurely and cannot recover from an early mistake. We propose VideoTreeSearch (VTS), a framework that casts grounded LVQA as iterative self-correcting search over an adaptive temporal tree. VTS constructs a non-uniform tree through adaptive temporal segmentation so that each node corresponds to a semantically coherent segment, and trains an agent to navigate the tree through four discrete operations: zoom_in, zoom_out, shift, and answer. These operations expose backtracking and recovery as explicit, learnable primitives rather than implicit behaviors. To train this navigation, we introduce a trajectory synthesis pipeline that produces multi-step paths through the tree, including deliberate detours into incorrect branches followed by recovery. We use these trajectories for supervised fine-tuning, followed by reinforcement learning with grounding and answer-accuracy rewards. On three Grounded LVQA benchmarks (CG-Bench, Haystack-LVBench, Haystack-Ego4D), VTS outperforms the strongest prior multi-turn cropping agents by +12.5 mIoU on CG-Bench and +7.4 T-F1 on Haystack-Ego4D. The learned policy also transfers to general long-video QA, surpassing all prior multi-turn cropping agents on Video-MME, MLVU, and LVBench by up to +7.1 accuracy points. Ablations confirm that self-correcting hierarchical search is the central mechanism behind these gains: removing either adaptive descent or explicit backtracking substantially degrades performance.

cs.CV↗

VLA-$Δ$: Learning Physical State Transition for Vision-Language-Action Models

Vision-language-action (VLA) models transfer semantic knowledge acquired through vision-language pretraining to robot control, achieving strong performance across various benchmarks. However, most VLAs directly map observations to demonstrated actions without explicit supervision of intermediate physical state changes, making them prone to overfitting observation-action correspondences. Some methods introduce intermediate reasoning or future-state prediction in language, visual, or latent spaces to guide action generation, but do not explicitly supervise the physical state transition itself. We therefore propose VLA-$Δ$, a general framework that uses physical changes between current and future states as explicit learning targets. To support this objective, we construct a large-scale pretraining dataset and derive physical state-transition annotations covering task-relevant semantics and visual motion changes, providing more direct supervision for continuous action generation. During inference, the model generates actions directly from current observations without additional inference overhead. VLA-$Δ$ achieves performance competitive with the state of the art on LIBERO and RoboTwin, and outperforms $π_{0.5}$ in real-robot experiments. Our code is available at https://github.com/ZJU-OmniAI/VLA-Delta

cs.CV↗

Spatial Memory by Agents, for Agents

Spatial memory enables robots to reuse past observations for localization, navigation, and question answering. Designing such memory involves jointly deciding what to retain and how to retrieve it for diverse semantic, metric, and temporal needs. We investigate whether coding agents can automate this design process through iterative implementation and evaluation. Under a fixed development budget, a Designer Agent writes and revises code for memory construction and retrieval using feedback from a fixed Query Agent. The resulting implementations are frozen before building memories from held-out observation histories, without access to downstream questions, and evaluated across five complementary task tracks using the same Query Agent. Across three independent design searches, Codex / GPT-6 Astra averages 44.23 on the equally weighted five-track score, versus 29.32 for ReMEmbR, the strongest evaluated baseline supporting all five tracks, a 50.9% relative improvement. Qualitative analysis relates performance differences to evidence retention and retrieval, while identifying grounding and evidence-use failures. These results demonstrate the potential of coding agents to automate spatial memory design, while revealing task-dependent trade-offs in performance and computational cost. Project page: https://ba-fa-3344.github.io/.

cs.RO↗

Progress in Muon Cooling and Moderation Techniques

Muon beams are essential tools for a wide range of applications such as the search for rare muon decays and muon spectroscopy. However, conventional muon beams produced from pion decay have large emittance and broad momentum spread. Conventional beam-cooling techniques cannot improve the beam quality significantly owing to the 2.2 $μ$s muon mean lifetime. This has motivated the development of dedicated muon beam cooling and moderation methods. This review surveys the principal approaches, including ionisation cooling for high-brightness muon beams, low-energy muon moderation and frictional cooling for slow muon production, and laser ionisation of thermal muonium for ultra-low-emittance muon sources. The challenges associated with producing slow negative muons and emerging concepts based on cyclotron trapping and muon-catalysed fusion are also discussed. These complementary techniques offer promising pathways towards next-generation muon facilities and precision experiments.

physics.acc-ph↗

Payment Design under Belief Disagreement: Rent Placement, Bid Resolution, and Auction Choice

When auction bidders misperceive their rivals, a seller can change not only the amount of information rent but also the states in which it is delivered. We ask when this payment-design opportunity survives a finite bidding interface, nonnegative-payment restrictions, and comparison with conventional auctions under the same beliefs. We formulate payment design as a linear program in nonnegative bidder rents. A single state can safely carry a report's rent when it is both cheapest for its intended type and least attractive to potential mimics. We characterize this alignment, bound the cost of a restricted payment architecture when alignment is imperfect, and use shadow prices when the rankings conflict. Own-type projection yields a unique optimal expected-payment rule for a fixed allocation. Contiguous bid bands implement truthful reporting at every value, but create pooling losses and within-band surplus. Under a separate smooth-belief model, feasible truthful rules and revenue upper bounds support comparisons with a first-price equilibrium under the same beliefs. A calibration using 1,000 eBay proof-set auction records illustrates a partition-dependent reversal: first price dominates all exactly truthful, no-subsidy rules on ten equal-probability bands in one scenario, while a richer contingent rule on ten equal-dollar bands outperforms it. An operator should choose interface resolution, payment dependence, and integrity tolerances jointly. Validate the probabilities that support incentives, compare formats under the same beliefs, and distinguish redesign from deployment. Any receipt advantage must cover implementation costs before it can imply a profit advantage.

cs.GT↗

SEIS: Self-Evolving Inference Systems

Inference systems determine how fast and how cheaply language models can be served, so making them faster has direct practical value. However, prior work focuses mostly on optimizing certain parts such as kernels or memory within the large system. In this work, we take a holistic approach and apply agentic self-evolution to optimize the whole system end-to-end. Our SEIS (Self-Evolving Inference Systems) autonomously optimizes the entire mini-sglang engine without human intervention through iterative sessions with inherited experiences and code changes. Serving Qwen3-0.6B on H100, the resulting engine reaches 3.27X the throughput of the original mini-sglang implementation and beats SOTA engines like vLLM, TensorRT-LLM, and SGLang in the single-request workload. The correctness of the optimized inference engine by SEIS is tested in terms of numerical difference and downstream accuracy on math and long-context retrieval tasks. The code and session histories show that the speedup comes from redesigning the whole engine and that building on earlier sessions beats independent attempts. These results suggest that agentic self-evolution can optimize a complex system end-to-end. The evaluation also has to evolve with the engine, and letting agents evolve it is a natural next step.

cs.AI↗

LEAP: Learning Efficient Action Proposals For LLM Agents

LLM agents are known to be slow in rollouts. An agent completes a task one step at a time. At each step, it reasons and then chooses an action to execute. The next step and action cannot start until the previous one has finished. Speculative decoding accelerates the rollouts at the reason phase by drafting and verifying the inference tokens. Recent works have also started to apply similar ideas at the action phase. These works use off-the-shelf models, usually large, to draft action proposals for target model to verify. Large drafters match the target more often but take longer to propose, while small off-the-shelf models are fast but rarely make the same decision as the target. We ask a more general question: what determines the end-to-end speedup of action speculation? To answer it, we develop a latency framework for the speculative round. The framework compares what a round gains with what it costs. The gain depends on how well the drafter predicts the target and on how many steps the task can take before it ends. The cost comes from drafting, from waiting for target verification and from executing tools. Guided by the framework, we introduce LEAP (Learning Efficient Action Proposals) which keeps the drafter small and makes it accurate by training it on the target actions sequences. With a small 0.6B model, LEAP agrees with the target on most decisions and makes agents up to 60% faster in end-to-end wall clock time, with no systematic change in task success. Across various datasets, target models and draft models, the framework accounts for most of the measured speedups. We also show the draft model can be online trained with no prior trace collection and match the performance of offline training, making LEAP practical to deploy in the real world.

cs.LG↗

RSIGame: Autonomous Agentic Game Development with Recursive Self-improvement

Recent advances in large language models have made automatic game generation increasingly feasible, yet reliably improving generated games beyond a playable version remains challenging. Naive iterative refinement can easily overfit a small set of test cases, producing fragile games with unresolved bugs, missing behaviors, and poor generalization to broader player interactions. We introduce RSIGame, an autonomous agentic game development framework with recursive self-improvement. RSIGame organizes development into complementary local and global loops. Concretely, a local explore-diagnose-improve loop broadly explores the executable game, diagnoses and prioritizes discovered issues, and performs evidence-grounded revision, where an evolving checklist continually accumulates new testing and improvement guidance. A global loop tracks overall quality, preserves the best checkpoint, and detects saturation or regression over long-horizon development. Beyond test-time improvement, RSIGame further internalizes successful development experience into the generator through training. Across 140 GameCraft-Bench tasks, two game engines, and five generators, RSIGame consistently improves game quality under matched development budgets. Notably, experience internalization enables Qwen3.8-27B to reach 61.38 on Godot and 58.53 on Phaser, exceeding GPT-5.5 one-shot scores while reducing Qwen's generation tokens by 11 times.

cs.CL↗

ProgressCompass: Embodied Progress Reward Models Are Lost Without the Right Context

Embodied agents now take on ever longer tasks. For long tasks, knowing only whether a task finally succeeds or fails says little; the steps along the way matter. Progress Reward Models (PRMs) score how far a task has come at every step, and serve as dense rewards, verifiers and monitors. Yet in long tasks the current frame alone often cannot tell how far the task has come, because progress depends on what happened before. We call this problem context-dependent progress estimation. Existing benchmarks on progress estimation mostly focus on short tasks whose progress can be read from the current observation, and whether PRMs can estimate progress when context is needed remains underexplored. We therefore build ContextProgress-Bench, with 24 manipulation tasks for 120 episodes. The benchmark covers three settings: (i) State Recall, where information needed for progress appeared earlier but is not in the current frame; (ii) Sequence Tracking, where steps follow a fixed order, so progress requires knowing which steps are done and which comes next; and (iii) Recurrence Disambiguation, where look-alike frames sit at very different progress. We then run a paired diagnosis: each PRM keeps the same input format in both runs, and in one run its instruction integrates the right context. Even PRMs that read the entire history get lost in estimating progress, yet with the right context the same five models cut their progress error by 77-82%. Embodied PRMs are thus not incapable of progress estimation, but lost without the right context. We therefore propose ProgressCompass, an autonomous agentic loop that reorients an existing PRM and uses current general-purpose VLMs to supply the context the PRM needs. Wrapped in the loop, the same frozen PRM cuts its progress error by 63% and raises its rank agreement by 76%. With such a compass, PRMs estimate progress far better on longer, more complex tasks.

cs.CL↗

Video2Skill: From Streaming Experience to Reusable Embodied Skills

Manipulation behaviors vary widely across objects and scenes, but they share a small set of reusable skills, and planning with these skills helps embodied agents generalize to new tasks. Yet an agent can only plan with skills it knows. Recovering skills from observed experience, the inverse of planning, builds this knowledge over time and yields skill data for training future agents. Vision-Language Models (VLMs) describe individual manipulation events well, but can they organize a stream of events into reusable skills? We formulate this problem as Streaming Embodied Skill Discovery (SESD): a model watches videos in sequence and maintains a persistent skill library that shapes its later decisions. To systematically measure this ability, we introduce Video2Skill, a benchmark that covers robot tabletop manipulation and human kitchen activity and tests three core capabilities: (i) locating manipulation events in time, (ii) grouping events of the same transformation, and (iii) deciding when to reuse an existing skill or create a new one. Across 19 open-source VLMs, many models group events at near-chance level, and scale does not consistently help. Their errors depend on how perception and library updates are coupled: joint models merge distinct transformations into one skill, while models that update the library from text descriptions duplicate recurring ones. Supervised fine-tuning, including our counterfactual library-state rebalancing (CLaRe), improves grouping but exposes a deeper bottleneck: trained models consolidate familiar skills yet rarely expand the library. Their libraries stall below half the reference size, and transformations unseen in training are located in time but almost never given a new skill. Recognizing when existing skills are insufficient thus emerges as the central challenge.

cs.CL↗

BiFE: Search-Efficient Discovery of CPU-Only Branching Policies via LLM-based Bi-Fidelity Evolution

In branch-and-bound (B&B) for mixed-integer linear programming (MILP), branching variable selection critically impacts efficiency. Existing neural branching policies often require GPU inference, while CPU-efficient symbolic expressions lack the representational capacity for complex logic. Large Language Model (LLM)-generated code provides a flexible search space for designing lightweight branching rules with diverse algorithmic logic. To discover effective rules within LLM-based evolutionary frameworks, a core challenge arises: full B&B evaluation on real instances is prohibitively expensive, whereas offline imitation learning suffers from distribution shift. To address this, we introduce a Bi-Fidelity Evolutionary framework (BiFE). It employs low-fidelity imitation scores as a rapid pre-screener and selectively applies high-fidelity on-instance evaluation only to elite candidates, effectively balancing search efficiency with performance reliability. Experiments validate both the search efficiency of BiFE and the competitiveness of its discovered rules, which outperform the SCIP solver and other baselines on CPUs, and even surpass certain GPU-based neural policies.

cs.AI↗

Flow induced two-dimensional $ζ$ potential on chemically uniform solid-liquid interface: A possible electrokinetic origin of streamwise vortices

The $ζ$ potential at solid-liquid interfaces governs electrokinetic phenomena across chemistry, biology, and engineering. Despite its centrality, local measurement of this potential has remained a formidable challenge, forcing the universal assumption of spatial uniformity on chemically homogeneous surfaces. Here we introduce Fluorescence Photobleaching Electrochemistry Analysis (FLEA)-- a non-invasive technique that achieves 180 nm lateral resolution -- to map the $ζ$ potential at unprecedented spatial detail. We discover that a perfectly uniform glass surface, under laminar flow, develops an intrinsic two-dimensional $ζ$ potential landscape: approximately symmetrical V-shaped distribution across the channel width that varies systematically with flow velocity, pH, and ionic strength. This two-dimensional electrochemical heterogeneity generates a rotational electric body force that spontaneously produces streamwise vortices-structures commonly observed in boundary layers. Our findings establish FLEA as a transformative tool for surface characterization, reveal a previously unrecognized origin of streamwise vorticity at interfaces from the aspect of electrokinetic.

cond-mat.mes-hall↗

ISS Policy : Scalable Diffusion Policy with Implicit Scene Supervision

Vision-based imitation learning has enabled impressive robotic manipulation skills, but action imitation alone provides limited supervision of the geometric consequences of robot behavior. To address this limitation, we introduce **Implicit Scene Supervision (ISS) Policy**, a 3D visuomotor diffusion policy with a DiT backbone that predicts continuous action sequences from point-cloud observations. ISS augments action diffusion with a supervised robot motion predictor that maps generated actions and robot-state context to end-effector motion, and then uses the predicted motion together with gripper intent to forecast future point-cloud representations. By explicitly modeling the intermediate transition from action to robot motion, ISS encourages the policy to capture how its actions affect the surrounding 3D scene. We further introduce asymmetric gradient routing to separate direct motion regression from scene-level policy supervision, together with a change-balanced objective that accounts for variations in scene-change magnitude. These auxiliary objectives provide dynamics-aware geometric supervision using only expert demonstrations, without requiring additional annotations or auxiliary modules at inference time. ISS Policy achieves state-of-the-art performance on single-arm manipulation tasks in MetaWorld and dexterous manipulation tasks in Adroit, while real-world dual-arm experiments further demonstrate its effectiveness on physical robotic manipulation. The resulting framework preserves the scalable DiT backbone and standard diffusion-policy control interface. Code and videos will be released.

cs.RO↗

TimeBlind: A Spatio-Temporal Compositionality Benchmark for Video LLMs

Fine-grained spatio-temporal understanding is essential for video reasoning and embodied AI. Yet, while Multimodal Large Language Models (MLLMs) master static semantics, their grasp of temporal dynamics remains brittle. We present TimeBlind, a diagnostic benchmark for compositional spatio-temporal understanding. Inspired by cognitive science, TimeBlind categorizes fine-grained temporal understanding into three levels: recognizing atomic events, characterizing event properties, and reasoning about event interdependencies. Unlike benchmarks that conflate recognition with temporal reasoning, TimeBlind leverages a minimal-pairs paradigm: video pairs share identical static visual content but differ solely in temporal structure, utilizing complementary questions to neutralize language priors. Evaluating over 20 state-of-the-art MLLMs (e.g., GPT-5, Gemini 3 Pro) on 600 curated instances (2400 video-question pairs), reveals that the Instance Accuracy (correctly distinguishing both videos in a pair) of the best performing MLLM is only 48.2%, far below the human performance (98.2%). These results demonstrate that even frontier models rely heavily on static visual shortcuts rather than genuine temporal logic, positioning TimeBlind as a vital diagnostic tool for next-generation video understanding. Dataset and code are available at https://baiqi-li.github.io/timeblind_project/ .

cs.CV↗

Muonium Spectroscopy as a Quantum Sensor for Ultralight Axion Dark Matter

High-intensity muon beams could enable a Muonium-based Axion Search through resonant quantum transitions between Hyperfine states (MASH). Combining theoretical calculations with simulation results, we demonstrate that such a muonium-based experimental approach could complement and tighten constraints on the axion-muon coupling beyond existing limits from the muon $g\!-\!2$ measurement, over the axion mass range of 18--130 $μ$eV. These results establish a new spectroscopic channel in muonium that enables searches for axion and axion-like particle dark matter.

hep-ph↗