Search arXiv⌕ Search

arXiv subjects

Junlin Wang

Publications and source records attributed to Junlin Wang.

At least 19 recordsLinked to original sources

Test-Time Compute Scales Better with Parallel Agent Teams

Test-time scaling improves the performance of a fixed model by spending more computation at inference. There are two common ways to spend it on agents when a verifier can score every attempt. Best-of-N sampling runs many independent agents and keeps the best result but saturates quickly. One large team of collaborating agents can score higher but stops improving as it grows. Can we get the best of both? Our method, parallel agent teams, does so by sampling many small teams in parallel and keeping the best team's result. We test it on 12 long-horizon tasks from GPU kernels and formal proofs to exploit development. With a 27B model and a budget of 100 agents, parallel teams of ten beat parallel agents on 10 of 12 tasks and beat one team of 100 on all four tasks where we ran such large teams. At equal dollar cost the same parallel teams beat a frontier model on half the tasks. A simple statistical model shows that individual agents lose because their many samples have a low mean and that one large team loses because a single sample gets no gain from variance. Sampling many small teams in parallel balances a high mean with the gain from variance. Parallel agent teams thus offer a more effective way to scale test-time compute.

cs.AI↗

Atomic Consistency Preference Optimization for Long-Form Question Answering

Large Language Models (LLMs) often produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated (factual, non-factual) pairs. However, this approach often relies on a stronger model (e.g., GPT-4) or an external knowledge base to assess factual correctness that may not always be accessible. Addressing this, we propose Atomic Consistency Preference Optimization (ACPO), a self-supervised preference-tuning method that enhances factual accuracy without external supervision. ACPO leverages atomic consistency signals (i.e., the agreement of individual facts across multiple stochastic responses) to identify high- and low-quality data pairs for model alignment. Despite being fully self-supervised, ACPO outperforms the strong supervised alignment baseline by 1.95 points averaged across Phi-3 and Llama3 on the LongFact and BioGen datasets, demonstrating its effectiveness in improving factual reliability without relying on external models or knowledge bases.

cs.CL↗

PRO-LONG: Programmatic Memory Enables Long-Horizon Reasoning

Long-horizon tasks require sustained perception, reasoning, and exploration, and are a persistent challenge for large language model (LLM) agents. This gap is reflected in their limited performance on continual learning benchmarks such as ARC-AGI-3 and NetHack, especially when evaluated without specialized harnesses. Various agent harnesses have been proposed to close the gap, and each commits to a strategy for handling long sequences of observations, i.e., what information to save from the environment and how to load it into model context, a choice we argue is particularly consequential. Existing methods for context management have typically been designed around a tradeoff, where preserving more information makes relevant details harder to retrieve. Recent progress in coding agents, however, makes new approaches to memory viable. We propose PRO-LONG, a minimal context management framework built around programmatic memory for LLM agents in long-horizon, exploratory settings. PRO-LONG addresses the tradeoff by keeping a complete, structured interaction log and using programmatic tools to search this history efficiently. On NetHack, we find increasing gains from using PRO-LONG for cross-episodic memory, with scores reaching up to 2.2x those of a base coding agent. On the full ARC-AGI-3 public game set, PRO-LONG improves over this baseline agent by an average of 18.0 percentage points across frontier models, and matches or exceeds state-of-the-art specialized harnesses (up to 97.4% best@2) while using 4.2-5.8x fewer tokens. Relevant code and logs are available at https://github.com/alexisfox7/PRO-LONG.

cs.AI↗

Knowing When to Stop: Efficient Context Processing via Latent Sufficiency Signals

Large language models (LLMs) process entire input contexts indiscriminately, which is inefficient when the information required to answer a query is localized within the context. We present dynamic context cutoff, a novel method enabling LLMs to self-terminate processing upon acquiring sufficient task-relevant information. Through analysis of model internals, we discover that specific attention heads inherently encode "sufficiency signals" -- detectable through lightweight classifiers -- that predict when critical information has been processed. This reveals a new efficiency paradigm: models' internal understanding naturally dictates processing needs rather than external compression heuristics. Comprehensive experiments across six QA datasets (up to 40K tokens) with three model families (LLaMA/Qwen/Mistral, 1B-70B) demonstrate 3.4% accuracy improvement while achieving 1.33x token reduction on average. Furthermore, our method demonstrates superior performance compared to other context efficiency methods at equivalent token reduction rates. Additionally, we observe an emergent scaling phenomenon: while smaller models require probing for sufficiency detection, larger models exhibit intrinsic self-assessment capabilities through prompting.

cs.CL↗

Vision2Code: A Multi-Domain Benchmark for Evaluating Image-to-Code Generation

We study whether vision-language models (VLMs) can reconstruct structured, graphics-based images by writing executable code to generate them. Existing benchmarks on image-to-code generation are generally limited in the domains they cover, rely on reference source code for evaluation, and do not provide metrics to quantify domain-specific issues. To address this, we introduce Vision2Code, a multi-domain, reference-code-free benchmark containing 539 test examples from 15 different datasets, spanning charts and plots, geometry, graphs, scientific diagrams, documents and 3D spatial scenes. Evaluation is done by executing the generated programs and comparing their rendered outputs to the source images using dataset- and domain-specific rubrics for a VLM judge. We find that the final scores obtained in this manner have substantially higher correlation with human ratings than those obtained via generic VLM rubrics or standard pixel-based, embedding-based or perceptual similarity metrics. We further demonstrate the reliability of our evaluation under repeated sampling from the VLM judges, varying rubric weights, and in the presence of controlled errors. Our results highlight the variability across domains of existing image-to-code generation models, with accuracy on charts and graphs being significantly higher than other domains. Lastly, we demonstrate the utility of our scoring method for rejection sampling of generated programs for fine-tuning, improving Qwen3.5-9B by 16% and Phi-4-MM by 28% compared to their base checkpoints. Our code and data are publicly available at https://image2code.github.io/vision2code/.

cs.CV↗

How Much Backtracking is Enough? Exploring the Interplay of SFT and RL in Enhancing LLM Reasoning

Recent advancements in large language models (LLMs) suggest that reinforcement learning (RL) effectively internalizes search strategies, yielding significant improvements on challenging reasoning tasks through extended chains of thought. While backtracking is widely viewed as the core mechanism enabling this improvement, its precise dynamics and how SFT and RL interact to optimize for it, remain poorly understood. In this work, we argue that backtracking, the ability to explicitly revisit and revise earlier reasoning steps, is a key operator that enables the transition from linear generation to non-linear tree search for LLMs, unlocking strong performance on combinatorially complex tasks. We motivate this claim by first studying the interplay between supervised fine-tuning (SFT) warm-up and RL across eight reasoning tasks: Countdown, Sudoku, Arc 1D, Advanced Geometry, Color Cube Rotation, List Functions, Zebra Puzzles, and Self Reference. We find that a regular SFT warm-up using self-sampled CoT sequences provides a modest boost to subsequent RL compared to training without any SFT warm-up; however, such gains saturate as tasks become more difficult. Motivated by this failure mode, we introduce a backtracking-centric training recipe. By synthetically varying the number of explicit backtracking steps in the SFT warm-up, we show that (i) longer CoTs containing backtracks stabilize and amplify RL, and (ii) the optimal backtrack depth scales with task difficulty: zero for Arc 1D, one for Countdown, and five or more for Sudoku. Our findings establish that principled backtracking is essential for improving the model's reasoning capability.

cs.AI↗

Rethinking Inference-Time Scaling: Efficiency Limits and Linguistic Signals

There is intense interest in investigating how inference time compute (ITC) (e.g. repeated sampling, refinements, etc) can improve large language model (LLM) capabilities. While breakthroughs like DeepSeek-R1 highlight the power of reinforcement learning for reasoning, the interaction between ITC and reasoning-optimized weights remains poorly understood. This work conducts a comprehensive analysis of inference-time scaling methods for both reasoning and non-reasoning models on challenging reasoning tasks. While prior work suggests that scaling test-time compute can optimally substitute for model parameter scaling, we identify a fundamental limit to this compute-equivalence, the reasoning floor, a performance plateau that non-reasoning models cannot escape, no matter how much inference compute is spent. We demonstrate that general-purpose models fail to match the accuracy of reasoning-optimized models even with an order of magnitude more inference compute, suggesting that internalizing reasoning protocols is a prerequisite for effective test-time scaling. Within reasoning models, we find that the complexity of the scaling method often yields diminishing returns; simple majority voting consistently outperforms sophisticated sequential revision and mixture-of-agents frameworks. Crucially, we identify a Linguistic Signal of Correctness - correct responses are significantly more concise and exhibit a lower density of hedging and thinking markers. We demonstrate that these intrinsic linguistic features can serve as zero-compute proxies for response quality, providing a pathway to more efficient, self-diagnostic reasoning agents.

cs.AI↗

Quantum many-body mixed phase space revealed by hybrid feedback control

Understanding how complex systems transition between order and chaos is a central challenge of nonequilibrium physics. While weak perturbations of classical integrable systems give rise to a mixed phase space of coexisting regular and chaotic trajectories, analogous behavior in interacting quantum many-body systems has remained elusive. Here we develop and experimentally implement a hybrid quantum-classical feedback protocol that autonomously discovers and stabilizes long-lived regular trajectories in a superconducting quantum processor. Each iteration combines short-time quantum evolution with classical optimization that projects the dynamics back onto a low-entanglement variational manifold, effectively distilling coherence from chaotic evolution. The stabilized trajectories reveal a quantum many-body mixed phase space emerging from nonlinear variational dynamics, without a direct analogue in classical or few-body quantum systems. Our results establish a versatile framework for algorithmic discovery and control of coherent dynamics previously inaccessible to experiment.

quant-ph↗

Iterative Diffusion-Refined Neural Attenuation Fields for Multi-Source Stationary CT Reconstruction: NAF Meets Diffusion Model

Multi-source stationary computed tomography (CT) has recently attracted attention for its ability to achieve rapid image reconstruction, making it suitable for time-sensitive clinical and industrial applications. However, practical systems are often constrained by ultra-sparse-view sampling, which significantly degrades reconstruction quality. Traditional methods struggle under ultra-sparse-view settings, where interpolation becomes inaccurate and the resulting reconstructions are unsatisfactory. To address this challenge, this study proposes Diffusion-Refined Neural Attenuation Fields (Diff-NAF), an iterative framework tailored for multi-source stationary CT under ultra-sparse-view conditions. Diff-NAF combines a Neural Attenuation Field representation with a dual-branch conditional diffusion model. The process begins by training an initial NAF using ultra-sparse-view projections. New projections are then generated through an Angle-Prior Guided Projection Synthesis strategy that exploits inter view priors, and are subsequently refined by a Diffusion-driven Reuse Projection Refinement Module. The refined projections are incorporated as pseudo-labels into the training set for the next iteration. Through iterative refinement, Diff-NAF progressively enhances projection completeness and reconstruction fidelity under ultra-sparse-view conditions, ultimately yielding high-quality CT reconstructions. Experimental results on multiple simulated 3D CT volumes and real projection data demonstrate that Diff-NAF achieves the best performance under ultra-sparse-view conditions.

cs.CV↗

Frequency-Guided Action Diffusion via Sub-Frequency Manifold Traversal

Learning visuomotor policies via behavior cloning typically involves mimicking expert demonstrations collected by human operators. However, natural human demonstrations inherently contain high-frequency noise, such as intermittent jerks, pauses, and action jitter. Training policies to directly imitate these raw trajectories inevitably causes the model to inherit these suboptimal behaviors. This pathology is particularly pronounced in diffusion-based policies, where iterative denoising steps can inadvertently amplify high-frequency artifacts at the expense of meaningful fine-grained details. To address these limitations, we present a novel frequency-based algorithm that enables implicit spectral maneuvering and smooth action generation. Our method, Frequency Guidance Operator (FGO), steers the generation process of diffusion polices by progressively driving the noisy samples through intermediate sub-frequency manifolds with expanding spectral bands. Validated on 15 robotic manipulation tasks from 5 benchmarks, FGO achieves superior performance in enhancing action smoothness and temporal consistency while preserving the details necessary for successful task execution. Project website: https://henrywjl.github.io/frequency-guidance-operator/

cs.RO↗

Automated Benchmark Auditing for AI Agents and Large Language Models

Modern AI benchmarks operate at a complexity that outpaces traditional verification methods. Tasks authored by domain experts often contain implicit assumptions, incomplete environment specifications, and brittle evaluation logic that human annotation cannot reliably catch. We introduce Auto Benchmark Audit (ABA), an agentic framework that systematically audits individual benchmark tasks, uncovering issues such as hidden environment dependencies, specification gaps, and limited grading logic. We run ABA on a collection of frontier LLM benchmarks and previous NeurIPS publications, totaling 168 benchmarks across nine domains. Across this corpus, ABA identifies critical issues including ambiguous task design, execution environment conflicts, and incorrect ground truths in over 25.7% of the evaluated tasks. The precision of these automated audits is validated by expert review and independent third-party reports such as upstream PRs. Crucially, we demonstrate that these problematic tasks severely distorts capability assessments for agents and LLMs: filtering out these tasks with issues shifts model rankings and increases average performance on SWE-bench Verified and Terminal-Bench 2 by 9.9% and 9.6%, respectively. We release the agentic tool and all task annotations to support the future development of frontier benchmarks.

cs.CL↗

When Does Divide and Conquer Work for Long Context LLM? A Noise Decomposition Framework

We investigate the challenge of applying Large Language Models (LLMs) to long texts. We propose a theoretical framework that distinguishes the failure modes of long context tasks into three categories: cross-chunk dependence (task noise), confusion that grows with context size (model noise), and the imperfect integration of partial results (aggregator noise). Under this view, we analyze when it is effective to use multi-agent chunking, i.e., dividing a lengthy sequence into smaller chunks and aggregating the processed results of each chunk. Our experiments on tasks such as retrieval, question answering, and summarization confirm both the theoretical analysis and the conditions that favor multi-agent chunking. By exploring the accelerated decay of model fidelity with input length, we also explain why, for large inputs, a weaker model configured with chunk-based processing can surpass a more advanced model like GPT4o applied in a single shot. Overall, we present a principled understanding framework and our results highlight a direct pathway to handling long contexts in LLMs with carefully managed chunking and aggregator strategies.

cs.CL↗

Grounding Bodily Awareness in Visual Representations for Efficient Policy Learning

Learning effective visual representations for robotic manipulation remains a fundamental challenge due to the complex body dynamics involved in action execution. In this paper, we study how visual representations that carry body-relevant cues can enable efficient policy learning for downstream robotic manipulation tasks. We present $\textbf{I}$nter-token $\textbf{Con}$trast ($\textbf{ICon}$), a contrastive learning method applied to the token-level representations of Vision Transformers (ViTs). ICon enforces a separation in the feature space between agent-specific and environment-specific tokens, resulting in agent-centric visual representations that embed body-specific inductive biases. This framework can be seamlessly integrated into end-to-end policy learning by incorporating the contrastive loss as an auxiliary objective. Our experiments show that ICon not only improves policy performance across various manipulation tasks but also facilitates policy transfer across different robots. The project website: https://inter-token-contrast.github.io/icon/

cs.RO↗

FSP-Diff: Full-Spectrum Prior-Enhanced DualDomain Latent Diffusion for Ultra-Low-Dose Spectral CT Reconstruction

Spectral computed tomography (CT) with photon-counting detectors holds immense potential for material discrimination and tissue characterization. However, under ultra-low-dose conditions, the sharply degraded signal-to-noise ratio (SNR) in energy-specific projections poses a significant challenge, leading to severe artifacts and loss of structural details in reconstructed images. To address this, we propose FSP-Diff, a full-spectrum prior-enhanced dual-domain latent diffusion framework for ultra-low-dose spectral CT reconstruction. Our framework integrates three core strategies: 1) Complementary Feature Construction: We integrate direct image reconstructions with projection-domain denoised results. While the former preserves latent textural nuances amidst heavy noise, the latter provides a stable structural scaffold to balance detail fidelity and noise suppression. 2) Full-Spectrum Prior Integration: By fusing multi-energy projections into a high-SNR full-spectrum image, we establish a unified structural reference that guides the reconstruction across all energy bins. 3) Efficient Latent Diffusion Synthesis: To alleviate the high computational burden of high-dimensional spectral data, multi-path features are embedded into a compact latent space. This allows the diffusion process to facilitate interactive feature fusion in a lower-dimensional manifold, achieving accelerated reconstruction while maintaining fine-grained detail restoration. Extensive experiments on simulated and real-world datasets demonstrate that FSP-Diff significantly outperforms state-of-the-art methods in both image quality and computational efficiency, underscoring its potential for clinically viable ultra-low-dose spectral CT imaging.

cs.CV↗

Continuity-driven Synergistic Diffusion with Neural Priors for Ultra-Sparse-View CBCT Reconstruction

The clinical application of cone-beam computed tomography (CBCT) is constrained by the inherent trade-off between radiation exposure and image quality. Ultra-sparse angular sampling, employed to reduce dose, introduces severe undersampling artifacts and inter-slice inconsistencies, compromising diagnostic reliability. Existing reconstruction methods often struggle to balance angular continuity with spatial detail fidelity. To address these challenges, we propose a Continuity-driven Synergistic Diffusion with Neural priors (CSDN) for ultra-sparse-view CBCT reconstruction. Neural priors are introduced as a structural foundation to encode a continuous threedimensional attenuation representation, enabling the synthesis of physically consistent dense projections from ultra-sparse measurements. Building upon this neural-prior-based initialization, a synergistic diffusion strategy is developed, consisting of two collaborative refinement paths: a Sinogram Refinement Diffusion (Sino-RD) process that restores angular continuity and a Digital Radiography Refinement Diffusion (DR-RD) process that enforces inter-slice consistency from the projection image perspective. The outputs of the two diffusion paths are adaptively fused by the Dual-Projection Reconstruction Fusion (DPRF) module to achieve coherent volumetric reconstruction. Extensive experiments demonstrate that the proposed CSDN effectively suppresses artifacts and recovers fine textures under ultra-sparse-view conditions, outperforming existing state-of-the-art techniques.

cs.CV↗

DSGym: A Holistic Framework for Evaluating and Training Data Science Agents

Data science agents promise to accelerate discovery and insight-generation by turning data into executable analyses and findings. Yet existing data science benchmarks fall short due to fragmented evaluation interfaces that make cross-benchmark comparison difficult, narrow task coverage and a lack of rigorous data grounding. In particular, we show that a substantial portion of tasks in current benchmarks can be solved without using the actual data. To address these limitations, we introduce DSGym, a standardized framework for evaluating and training data science agents in self-contained execution environments. Unlike static benchmarks, DSGym provides a modular architecture that makes it easy to add tasks, agent scaffolds, and tools, positioning it as a live, extensible testbed. We curate DSGym-Tasks, a holistic task suite that standardizes and refines existing benchmarks via quality and shortcut solvability filtering. We further expand coverage with (1) DSBio: expert-derived bioinformatics tasks grounded in literature and (2) DSPredict: challenging prediction tasks spanning domains such as computer vision, molecular prediction, and single-cell perturbation. Beyond evaluation, DSGym enables agent training via execution-verified data synthesis pipeline. As a case study, we build a 2,000-example training set and trained a 4B model in DSGym that outperforms GPT-4o on standardized analysis benchmarks. Overall, DSGym enables rigorous end-to-end measurement of whether agents can plan, implement, and validate data analyses in realistic scientific context.

cs.AI↗

Observation of Topological Hall Effect in Synthetic Antiferromagnetic Skyrmion System

Synthetic antiferromagnetic (SAF) skyrmions have emerged as promising candidates for next-generation high-speed and highly integrated spintronic devices, owing to their exceptional properties such as high driving velocity, nanoscale dimensions, and the absence of the skyrmion Hall effect. In this work, we report the observation of the topological Hall effect in both compensated and non-compensated synthetic antiferromagnetic skyrmion systems based on [Pt/Co/Ru]2 bilayers. The antiferromagnetic skyrmions are demonstrated to be robust in these synthetic antiferromagnets under zero-field. Our first principal calculations and micromagnetic simulations demonstrate that the formation of the antiferromagnetic skyrmions are due to nonuniformity of RKKY coupling associated with the proximity effect induced magnetic moments in the Pt and Ru layers. The skyrmions in the Pt and Ru layers adjacent to the Co layers lead to the observed topological Hall effect. This work not only provides insight into the effect of the magnetic proximity effect and RKKY coupling to the SAF skyrmions, but also an effective detection method for the SAF skyrmion systems, thereby laying a foundation for the practical application of antiferromagnetic skyrmions in spintronic devices.

cond-mat.mes-hall↗

Towards Competent AI for Fundamental Analysis in Finance: A Benchmark Dataset and Evaluation

Generative AI, particularly large language models (LLMs), is beginning to transform the financial industry by automating tasks and helping to make sense of complex financial information. One especially promising use case is the automatic creation of fundamental analysis reports, which are essential for making informed investment decisions, evaluating credit risks, guiding corporate mergers, etc. While LLMs attempt to generate these reports from a single prompt, the risks of inaccuracy are significant. Poor analysis can lead to misguided investments, regulatory issues, and loss of trust. Existing financial benchmarks mainly evaluate how well LLMs answer financial questions but do not reflect performance in real-world tasks like generating financial analysis reports. In this paper, we propose FinAR-Bench, a solid benchmark dataset focusing on financial statement analysis, a core competence of fundamental analysis. To make the evaluation more precise and reliable, we break this task into three measurable steps: extracting key information, calculating financial indicators, and applying logical reasoning. This structured approach allows us to objectively assess how well LLMs perform each step of the process. Our findings offer a clear understanding of LLMs current strengths and limitations in fundamental analysis and provide a more practical way to benchmark their performance in real-world financial settings.

q-fin.ST↗