SPREG: Structured Plan Repair with Entropy-Guided Test-Time Intervention for Large Language Model Reasoning

Large Language Models (LLMs) are prone to logical hallucinations and stochastic drifts during long-chain reasoning. While Classifier-Free Guidance (CFG) can improve instruction adherence, standard static implementations often cause semantic dilution and linguistic degradation. We propose SPREG (Structured Plan-guided Real-time Entropy Gating), a lightweight inference-time framework for surgical error rectification. SPREG employs an adaptive dual-threshold mechanism to monitor real-time entropy, identifying sudden ``entropy spikes'' as reliable indicators of logical failure. Upon detection, it triggers a dynamic repair by replacing uninformative null-priors with reference distributions synthesized from historical high-confidence states. By modulating guidance intensity according to structured reasoning stages (e.g., Action, Observation), SPREG steers the model back to a stable manifold without compromising fluency. Our experiments demonstrate significant gains, notably a 20.0% absolute accuracy improvement on AIME25, while effectively suppressing uncontrolled entropy drift in complex tasks.

THEMIS: Towards Holistic Evaluation of MLLMs for Scientific Paper Fraud Forensics

We present THEMIS, a novel multi-task benchmark designed to comprehensively evaluate multimodal large language models (MLLMs) on visual fraud reasoning within real-world academic scenarios. Compared to existing benchmarks, THEMIS introduces three major advances. (1) Real-World Scenarios and Complexity: Our benchmark comprises over 4,000 questions spanning seven scenarios, derived from authentic retracted-paper cases and carefully curated multimodal synthetic data. With 60.47% complex-texture images, THEMIS bridges the critical gap between existing benchmarks and the complexity of real-world academic fraud. (2) Fraud-Type Diversity and Granularity: THEMIS systematically covers five challenging fraud types and introduces 16 fine-grained manipulation operations. On average, each sample undergoes multiple stacked manipulation operations, with the diversity and difficulty of these manipulations demanding a high level of visual fraud reasoning from the models. (3) Multi-Dimensional Capability Evaluation: We establish a mapping from fraud types to five core visual fraud reasoning capabilities, thereby enabling an evaluation that reveals the distinct strengths and specific weaknesses of different models across these core capabilities. Experiments on 16 leading MLLMs show that even the best-performing model, GPT-5, achieves an overall performance of only 56.15%, demonstrating that our benchmark presents a stringent test. We expect THEMIS to advance the development of MLLMs for complex, real-world fraud reasoning tasks.

Reflection in the Dark: Exposing and Escaping the Black Box in Reflective Prompt Optimization

Automatic prompt optimization (APO) has emerged as a powerful paradigm for improving LLM performance without manual prompt engineering. Reflective APO methods such as GEPA iteratively refine prompts by diagnosing failure cases, but the optimization process remains black-box and label-free, leading to uninterpretable trajectories and systematic failure. We identify and empirically demonstrate four limitations: on GSM8K with a defective seed, GEPA degrades accuracy from 23.81% to 13.50%. We propose VISTA, a multi-agent APO framework that decouples hypothesis generation from prompt rewriting, enabling semantically labeled hypotheses, parallel minibatch verification, and interpretable optimization trace. A two-layer explore-exploit mechanism combining random restart and epsilon-greedy sampling further escapes local optima. VISTA recovers accuracy to 87.57% on the same defective seed and consistently outperforms baselines across all conditions on GSM8K and AIME2025.

Compositional Oil Spill Detection Based on Object Detector and Adapted Segment Anything Model from SAR Images

Semantic segmentation-based methods have attracted extensive attention in oil spill detection from SAR images. However, the existing approaches require a large number of finely annotated segmentation samples in the training stage. To alleviate this issue, we propose a composite oil spill detection framework, SAM-OIL, comprising an object detector (e.g., YOLOv8), an adapted Segment Anything Model (SAM), and an Ordered Mask Fusion (OMF) module. SAM-OIL is the first application of the powerful SAM in oil spill detection. Specifically, the SAM-OIL strategy uses YOLOv8 to obtain the categories and bounding boxes of oil spill-related objects, then inputs bounding boxes into the adapted SAM to retrieve category-agnostic masks, and finally adopts the Ordered Mask Fusion (OMF) module to fuse the masks and categories. The adapted SAM, combining a frozen SAM with a learnable Adapter module, can enhance SAM's ability to segment ambiguous objects. The OMF module, a parameter-free method, can effectively resolve pixel category conflicts within SAM. Experimental results demonstrate that SAM-OIL surpasses existing semantic segmentation-based oil spill detection methods, achieving mIoU of 69.52%. The results also indicated that both OMF and Adapter modules can effectively improve the accuracy in SAM-OIL.