InnovatorBench: Evaluating Agents' Ability to Conduct Innovative LLM Research

AI agents could accelerate scientific discovery by automating hypothesis formation, experiment design, coding, execution, and analysis, yet existing benchmarks probe narrow skills in simplified settings. To address this gap, we introduce InnovatorBench, a benchmark-platform pair for realistic, end-to-end assessment of agents performing Large Language Model (LLM) research. It comprises 20 tasks spanning Data Construction, Filtering, Augmentation, Loss Design, Reward Design, and Scaffold Construction, which require runnable artifacts and assessment of correctness, performance, output quality, and uncertainty. To support agent operation, we develop ResearchGym, a research environment offering rich action spaces, distributed and long-horizon execution, asynchronous monitoring, and snapshot saving. We also implement a lightweight ReAct agent that couples explicit reasoning with executable planning using frontier models such as Claude-4, GPT-5, GLM-4.5, and Kimi-K2. Our experiments demonstrate that while frontier models show promise in code-driven research tasks, they struggle with fragile algorithm-related tasks and long-horizon decision making, such as impatience, poor resource management, and overreliance on template-based reasoning. Furthermore, agents require over 11 hours to achieve their best performance on InnovatorBench, underscoring the benchmark's difficulty and showing the potential of InnovatorBench to be the next generation of code-based research benchmark.

Interaction as Intelligence Part II: Asynchronous Human-Agent Rollout for Long-Horizon Task Training

Large Language Model (LLM) agents have recently shown strong potential in domains such as automated coding, deep research, and graphical user interface manipulation. However, training them to succeed on long-horizon, domain-specialized tasks remains challenging. Current methods primarily fall into two categories. The first relies on dense human annotations through behavior cloning, which is prohibitively expensive for long-horizon tasks that can take days or months. The second depends on outcome-driven sampling, which often collapses due to the rarity of valid positive trajectories on domain-specialized tasks. We introduce Apollo, a sampling framework that integrates asynchronous human guidance with action-level data filtering. Instead of requiring annotators to shadow every step, Apollo allows them to intervene only when the agent drifts from a promising trajectory, by providing prior knowledge, strategic advice, etc. This lightweight design makes it possible to sustain interactions for over 30 hours and produces valuable trajectories at a lower cost. Apollo then applies supervision control to filter out sub-optimal actions and prevent error propagation. Together, these components enable reliable and effective data collection in long-horizon environments. To demonstrate the effectiveness of Apollo, we evaluate it using InnovatorBench. Our experiments show that when applied to train the GLM-4.5 model on InnovatorBench, Apollo achieves more than a 50% improvement over the untrained baseline and a 28% improvement over a variant trained without human interaction. These results highlight the critical role of human-in-the-loop sampling and the robustness of Apollo's design in handling long-horizon, domain-specialized tasks.

GeoSDF: Plane Geometry Diagram Synthesis via Signed Distance Field

Plane Geometry Diagram Synthesis has been a crucial task in computer graphics, with applications ranging from educational tools to AI-driven mathematical reasoning. Traditionally, we rely on computer tools (e.g., Matplotlib and GeoGebra) to manually generate precise diagrams, but it usually requires huge, complicated calculations cost. Recently, researchers start to work on learning-based methods (e.g., Stable Diffusion and GPT4) to automatically generate diagrams, saving operational cost but usually suffering from limited realism and insufficient accuracy. In this paper, we propose a novel framework GeoSDF to automatically generate diagrams efficiently and accurately with Signed Distance Field (SDF). Specifically, we first represent geometric elements in the SDF, then construct a series of constraint functions to represent geometric relationships, next we optimize such constraint functions to get an optimized field of both elements and constraints, finally by rendering the optimized field, we can obtain the synthesized diagram. In our GeoSDF, we define a symbolic language to easily represent geometric elements and those constraints, and our synthesized geometry diagrams can be self-verified in the SDF, ensuring both mathematical accuracy and visual plausibility. In experiments, our GeoSDF synthesized both normal high-school level and IMO-level geometry diagrams. Through both qualitative and quantitative analysis, we can see that synthesized diagrams are realistic and accurate, and our synthesizing process is simple and efficient. Furthermore, we obtain a very high accuracy of solving geometry problems (over 95\% while the current SOTA accuracy is around 75%) by leveraging our self-verification property. All of these demonstrate the advantage of GeoSDF, paving the way for more sophisticated, accurate, and flexible generation of geometric diagrams for a wide array of applications.

WGSR-Bench: Wargame-based Game-theoretic Strategic Reasoning Benchmark for Large Language Models

Recent breakthroughs in Large Language Models (LLMs) have led to a qualitative leap in artificial intelligence' s performance on reasoning tasks, particularly demonstrating remarkable capabilities in mathematical, symbolic, and commonsense reasoning. However, as a critical component of advanced human cognition, strategic reasoning, i.e., the ability to assess multi-agent behaviors in dynamic environments, formulate action plans, and adapt strategies, has yet to be systematically evaluated or modeled. To address this gap, this paper introduces WGSR-Bench, the first strategy reasoning benchmark for LLMs using wargame as its evaluation environment. Wargame, a quintessential high-complexity strategic scenario, integrates environmental uncertainty, adversarial dynamics, and non-unique strategic choices, making it an effective testbed for assessing LLMs' capabilities in multi-agent decision-making, intent inference, and counterfactual reasoning. WGSR-Bench designs test samples around three core tasks, i.e., Environmental situation awareness, Opponent risk modeling and Policy generation, which serve as the core S-POE architecture, to systematically assess main abilities of strategic reasoning. Finally, an LLM-based wargame agent is designed to integrate these parts for a comprehensive strategy reasoning assessment. With WGSR-Bench, we hope to assess the strengths and limitations of state-of-the-art LLMs in game-theoretic strategic reasoning and to advance research in large model-driven strategic intelligence.

WDMamba: When Wavelet Degradation Prior Meets Vision Mamba for Image Dehazing

In this paper, we reveal a novel haze-specific wavelet degradation prior observed through wavelet transform analysis, which shows that haze-related information predominantly resides in low-frequency components. Exploiting this insight, we propose a novel dehazing framework, WDMamba, which decomposes the image dehazing task into two sequential stages: low-frequency restoration followed by detail enhancement. This coarse-to-fine strategy enables WDMamba to effectively capture features specific to each stage of the dehazing process, resulting in high-quality restored images. Specifically, in the low-frequency restoration stage, we integrate Mamba blocks to reconstruct global structures with linear complexity, efficiently removing overall haze and producing a coarse restored image. Thereafter, the detail enhancement stage reinstates fine-grained information that may have been overlooked during the previous phase, culminating in the final dehazed output. Furthermore, to enhance detail retention and achieve more natural dehazing, we introduce a self-guided contrastive regularization during network training. By utilizing the coarse restored output as a hard negative example, our model learns more discriminative representations, substantially boosting the overall dehazing performance. Extensive evaluations on public dehazing benchmarks demonstrate that our method surpasses state-of-the-art approaches both qualitatively and quantitatively. Code is available at https://github.com/SunJ000/WDMamba.

Unlock Pose Diversity: Accurate and Efficient Implicit Keypoint-based Spatiotemporal Diffusion for Audio-driven Talking Portrait

Audio-driven single-image talking portrait generation plays a crucial role in virtual reality, digital human creation, and filmmaking. Existing approaches are generally categorized into keypoint-based and image-based methods. Keypoint-based methods effectively preserve character identity but struggle to capture fine facial details due to the fixed points limitation of the 3D Morphable Model. Moreover, traditional generative networks face challenges in establishing causality between audio and keypoints on limited datasets, resulting in low pose diversity. In contrast, image-based approaches produce high-quality portraits with diverse details using the diffusion network but incur identity distortion and expensive computational costs. In this work, we propose KDTalker, the first framework to combine unsupervised implicit 3D keypoint with a spatiotemporal diffusion model. Leveraging unsupervised implicit 3D keypoints, KDTalker adapts facial information densities, allowing the diffusion process to model diverse head poses and capture fine facial details flexibly. The custom-designed spatiotemporal attention mechanism ensures accurate lip synchronization, producing temporally consistent, high-quality animations while enhancing computational efficiency. Experimental results demonstrate that KDTalker achieves state-of-the-art performance regarding lip synchronization accuracy, head pose diversity, and execution efficiency.Our codes are available at https://github.com/chaolongy/KDTalker.

Consistency Diffusion Models for Single-Image 3D Reconstruction with Priors

This paper delves into the study of 3D point cloud reconstruction from a single image. Our objective is to develop the Consistency Diffusion Model, exploring synergistic 2D and 3D priors in the Bayesian framework to ensure superior consistency in the reconstruction process, a challenging yet critical requirement in this field. Specifically, we introduce a pioneering training framework under diffusion models that brings two key innovations. First, we convert 3D structural priors derived from the initial 3D point cloud as a bound term to increase evidence in the variational Bayesian framework, leveraging these robust intrinsic priors to tightly govern the diffusion training process and bolster consistency in reconstruction. Second, we extract and incorporate 2D priors from the single input image, projecting them onto the 3D point cloud to enrich the guidance for diffusion training. Our framework not only sidesteps potential model learning shifts that may arise from directly imposing additional constraints during training but also precisely transposes the 2D priors into the 3D domain. Extensive experimental evaluations reveal that our approach sets new benchmarks in both synthetic and real-world datasets. The code is included with the submission.

XRAG: eXamining the Core -- Benchmarking Foundational Components in Advanced Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) synergizes the retrieval of pertinent data with the generative capabilities of Large Language Models (LLMs), ensuring that the generated output is not only contextually relevant but also accurate and current. We introduce XRAG, an open-source, modular codebase that facilitates exhaustive evaluation of the performance of foundational components of advanced RAG modules. These components are systematically categorized into four core phases: pre-retrieval, retrieval, post-retrieval, and generation. We systematically analyse them across reconfigured datasets, providing a comprehensive benchmark for their effectiveness. As the complexity of RAG systems continues to escalate, we underscore the critical need to identify potential failure points in RAG systems. We formulate a suite of experimental methodologies and diagnostic testing protocols to dissect the failure points inherent in RAG engineering. Subsequently, we proffer bespoke solutions aimed at bolstering the overall performance of these modules. Our work thoroughly evaluates the performance of advanced core components in RAG systems, providing insights into optimizations for prevalent failure points.

Towards Cross-device and Training-free Robotic Grasping in 3D Open World

Robotic grasping in the open world is a critical component of manufacturing and automation processes. While numerous existing approaches depend on 2D segmentation output to facilitate the grasping procedure, accurately determining depth from 2D imagery remains a challenge, often leading to limited performance in complex stacking scenarios. In contrast, techniques utilizing 3D point cloud data inherently capture depth information, thus enabling adeptly navigating and manipulating a diverse range of complex stacking scenes. However, such efforts are considerably hindered by the variance in data capture devices and the unstructured nature of the data, which limits their generalizability. Consequently, much research is narrowly concentrated on managing designated objects within specific settings, which confines their real-world applicability. This paper presents a novel pipeline capable of executing object grasping tasks in open-world scenarios even on previously unseen objects without the necessity for training. Additionally, our pipeline supports the flexible use of different 3D point cloud segmentation models across a variety of scenes. Leveraging the segmentation results, we propose to engage a training-free binary clustering algorithm that not only improves segmentation precision but also possesses the capability to cluster and localize unseen objects for executing grasping operations. In our experiments, we investigate a range of open-world scenarios, and the outcomes underscore the remarkable robustness and generalizability of our pipeline, consistent across various environments, robots, cameras, and objects. The code will be made available upon acceptance of the paper.

Dynamic PET Image Prediction Using a Network Combining Reversible and Irreversible Modules

Dynamic positron emission tomography (PET) images can reveal the distribution of tracers in the organism and the dynamic processes involved in biochemical reactions, and it is widely used in clinical practice. Despite the high effectiveness of dynamic PET imaging in studying the kinetics and metabolic processes of radiotracers. Pro-longed scan times can cause discomfort for both patients and medical personnel. This study proposes a dynamic frame prediction method for dynamic PET imaging, reduc-ing dynamic PET scanning time by applying a multi-module deep learning framework composed of reversible and irreversible modules. The network can predict kinetic parameter images based on the early frames of dynamic PET images, and then generate complete dynamic PET images. In validation experiments with simulated data, our network demonstrated good predictive performance for kinetic parameters and was able to reconstruct high-quality dynamic PET images. Additionally, in clinical data experiments, the network exhibited good generalization performance and attached that the proposed method has promising clinical application prospects.