Parameter-efficient fine-tuning (PEFT) enables low-cost adaptation of large language models but often suffers from limited representational flexibility. To address this, we incorporate a Mixture-of-Experts (MoE) design and propose Efficient and Expressive split-path experts that enhance specialization while maintaining low resource overhead. Split-Path Adaptive Representation Mixture-of-Experts (SparMoE) replaces discrete hard routing with a soft routing and fully-activated mixture, enabling stable optimization. Each expert is parameterized as a split-path modulation module, consisting of a scaling path that promotes expert specialization and a bias path that preserves expert-specific signals. This design significantly enhances expressive capacity while maintaining strict parameter efficiency and architectural compatibility with PEFT. Extensive evaluations on GLUE, GSM8K, MBPP, and a text rewriting task from SmolTalk show that our approach consistently outperforms or matches state-of-the-art PEFT methods under comparable parameter budgets, achieving a favorable trade-off between adaptability and efficiency.

Large language models (LLMs) are increasingly deployed as web-browsing agents, yet current benchmarks mostly evaluate explicit queries over clean factual evidence. We introduce Melonbench, a 665-item benchmark for coded entity resolution over noisy open-web user-generated content, instantiated with mid-to-long-tail public figures in Chinese entertainment blind-item discourse. Solving Melonbench requires agents to decode obfuscated clues, retrieve fragmented evidence, and identify the target entity. The benchmark is constructed through a five-stage semi-automatic pipeline with atomic event stratification, adversarial faithfulness auditing, and expert verification. Across 30 systems, the best non-search model reaches 25.1% accuracy and the best search-enabled system reaches 49.9%. Controlled comparisons of the same models with and without web search confirm consistent gains from web search, but also substantial remaining headroom. Diagnostic experiments identify semantic de-obfuscation and open-ended entity recall as key bottlenecks, and show that reasoning-centric search can outperform report-oriented agents. An 80-item finance pilot provides preliminary evidence of transfer beyond entertainment within a Chinese-language setting. Melonbench offers a challenging initial testbed for search agents under intentional obfuscation and noisy evidence. The dataset is publicly available at https://huggingface.co/datasets/janehaha/MelonBench.

Recent reinforcement learning (RL) approaches have advanced radiology report generation (RRG), yet two core limitations persist: (1) report-level rewards offer limited evidence-grounded guidance for clinical faithfulness; and (2) current methods lack an explicit self-improving mechanism to align with clinical preference. We introduce clinically aligned Evidence-aware Self-Correcting Reinforcement Learning (ESC-RL), comprising two key components. First, a Group-wise Evidence-aware Alignment Reward (GEAR) delivers group-wise, evidence-aware feedback. GEAR reinforces consistent grounding for true positives, recovers missed findings for false negatives, and suppresses unsupported content for false positives. Second, a Self-correcting Preference Learning (SPL) strategy automatically constructs a reliable, disease-aware preference dataset from multiple noisy observations and leverages an LLM to synthesize refined reports without human supervision. ESC-RL promotes clinically faithful, disease-aligned reward and supports continual self-improvement during training. Extensive experiments on two public chest X-ray datasets demonstrate consistent gains and state-of-the-art performance.

Radiology Report Generation (RRG) is an important research topic for relieving radiologists’ heavy workload. Existing RRG models mainly rely on supervised fine-tuning (SFT) based on different model architectures using data pairs of radiological images and corresponding radiologist-annotated reports. Recent research has shifted focus to post-training improvements, aligning RRG model outputs with human preferences using reinforcement learning (RL). However, the limited data coverage of high-quality annotated data poses risks of overfitting and generalization. This paper proposes a novel Online Iterative Self-Alignment (OISA) method for RRG that consists of four stages: self-generation of diverse data, self-evaluation for multi-objective preference data, self-alignment for multi-objective optimization and self-iteration for further improvement. Our approach allows for generating varied reports tailored to specific clinical objectives, enhancing the overall performance of the RRG model iteratively. Unlike existing methods, our framework significantly increases data quality and optimizes performance through iterative multi-objective optimization. Experimental results demonstrate that our method surpasses previous approaches, achieving state-of-the-art performance across multiple evaluation metrics.