The paper introduces a unified conditional-flow framework that integrates text-driven motion generation, semantic editing, and intra-structural retargeting into a single rectified-flow model. By treating editing as a change in semantic condition and retargeting as a change in skeletal condition, the approach eliminates fragmented pipelines and allows a single model to perform generation, zero‑shot editing, and zero‑shot retargeting on articulated 3D motion data. Experiments on SnapMoGen and a Mixamo subset demonstrate that the model can handle all three tasks without task‑specific fine‑tuning, preserving both motion semantics and skeletal structure.
By Junlin Li, Xinhao Song, Siqi Wang, Haibin Huang, Yili Zhao
The paper evaluates large language models (LLMs) for annotating German parliamentary debates on migration, achieving macro‑F1 scores comparable to human agreement, particularly with GPT‑5 and gpt‑oss‑120B. It combines soft‑label outputs with Design‑based Supervised Learning to mitigate systematic bias and applies the method to a 150‑plus‑year corpus, revealing high solidarity post‑war and a sharp rise in anti‑solidarity since 2015. The study demonstrates that LLMs can enable large‑scale social‑scientific analysis while highlighting the need for rigorous validation and bias correction.
By Aida Kostikova, Ole P\"utz, Steffen Eger, Olga Sabelfeld, Benjamin Paassen
The paper investigates whether the high costs of training chain-of-thought reasoning models can be reduced through algorithmic design. It introduces an autocurriculum approach that lets the model select which problems to focus on during training, showing that this method provably improves both supervised fine‑tuning and reinforcement learning. For supervised fine‑tuning, autocurriculum requires exponentially fewer reasoning demonstrations by targeting prompts where the model struggles, while for reinforcement learning it decouples computational cost from the quality of the reference model, making the burn‑in cost nearly independent of target accuracy.
By Nived Rajaraman, Audrey Huang, Miro Dudik, Robert Schapire, Dylan J. Foster, Akshay Krishnamurthy
The paper proposes a method called Modular Expert Merging for Biomedical Retrieval, which combines independently trained domain‑specialized experts instead of large mixed‑domain training. Experiments across four decoder‑only LLM families (0.6B‑7B) and twelve retrieval tasks from MTEB show that merging experts consistently outperforms mixed‑domain training. The authors also introduce a Synthesize‑Train‑Merge (STM) framework that generates hard negatives with a top‑tier LLM, fine‑tunes experts via LoRA, and merges them, achieving strong biomedical retrieval performance while retaining competitive general‑domain results.
By Sameh Khattab, Jean-Philippe Corbeil, Osman Alperen \c{C}inar-Kora\c{s}, Amin Dada, Julian Friedrich, Jiawei He, Douglas Teodoro, Jens Kleesiek
RATIO (Retrieval Across Typed Ideation Operations) is a large-scale benchmark designed to evaluate how well retrieval systems can support scientific inspiration. It defines relevance through three ideation moves—Address, Broaden, and Specify—each targeting different levels of abstraction in literature retrieval. The benchmark is built from millions of full-text CS papers using a novel discourse-marker distant supervision method, and includes extensive LLM and human vetting to ensure quality.
By Maayan Sharon, Tom Hope
KISS-GS is a modular compression pipeline for 3D Gaussian Splatting (3DGS) scenes that separates compression from training. It first compacts a vanilla 3DGS scene by 15.7× using state‑of‑the‑art pruning, then encodes the result into the SOG‑XT image‑based format, achieving an additional 6.6× reduction. Optional encoding‑aware fine‑tuning can further cut the size by 2.2×, yielding total reductions of 85× to 319× on standard benchmarks while enabling web‑native decoding.
By Wieland Morgenstern, Friedrich Elias Branschke, Florian Fleischmann, Adrian Szatmari, Paul Schlack, Florian Barthel, Peter Eisert, Anna Hilsmann
The paper investigates whether a language model’s own confidence can replace labeled data for teaching it to abstain from uncertain answers. By fine‑tuning models with LoRA to answer only when their frozen confidence is high and to say “I’m not sure” otherwise, the authors show that this label‑free approach matches label‑supervised abstention tuning on short‑form factual QA. The method works across six open‑weight models (1B‑8B) and is effective except for confidently wrong facts, which the confidence signal cannot flag.
By Ali Asaria, Tony Salomone, Deep Gandhi
The paper introduces a unified large language model workflow for modeling and inverse-design of metasurfaces across multiple families. By converting geometries, design parameters, and optical responses into a shared instruction‑following text format, the authors fine‑tune Gemma‑2‑9B on eight distinct metasurface families. Compared to single‑family models, the joint model predicts all families’ optical responses simultaneously and reduces mean‑squared error by an average of 56.5%.
"whyItMatters":"The approach demonstrates that a shared sequence‑based LLM interface can streamline cross‑family metasurface design, eliminating the need for separate surrogate architectures for each geometry class."
By Huanshu Zhang, Lei Kang, Yuyan Chen, Luxiang Wang, Zhaolong Cao, Douglas H. Werner
The paper investigates how differential privacy (DP) affects the ability to merge independently fine‑tuned task models into a single multi‑task model. It identifies two geometric challenges—local sharpness and reference drift—that hinder mergeability of private models. To address these, the authors propose DP‑Merging, a framework that uses a sharpness‑aware objective and a reference‑based alignment regularizer to improve mergeability while maintaining DP guarantees.
By Jin Liu, Junkang Liu, Ning Xi, Yinbin Miao, Dawei Wei, Ke Cheng, Jianfeng Ma
The paper introduces a production‑grade large language model (LLM) pricing system for the tourism industry that separates structured extraction and policy selection from deterministic numeric pricing. Policies are compiled into interpretable condition trees, allowing new clauses and evolving rules to be added without code changes while maintaining auditability. Deployed across 12 business categories and 1,500 operators, the system handled 3,960 orders in six months, cutting the order‑management team from 15‑20 to 3 and reducing per‑order handling time from 10 minutes to under 2 minutes.
By Ziqiang Zhang, Jing Ma, Zilong Wang, Jiayuan Chen, Yi Qiao, Yu He, Wei Zhang, Dai Cheng, Xiaoyu Shen
SIMPLER is a pre‑fine‑tuning method that reduces inference and deployment costs for Earth Observation foundation models by pruning redundant layers. It uses layer‑wise representation similarity on unlabeled task data to identify and remove up to 79% of parameters without requiring gradients, magnitude heuristics, or hyperparameter tuning. Experiments on Prithvi‑EO‑2, TerraMind, and ImageNet‑pretrained ViT‑MAE show that SIMPLER retains 94% of baseline performance while achieving 2.1× faster training and 2.6× faster inference.
By V\'ictor Barreiro, Johannes Jakubik, Francisco Arg\"uello, Dora B. Heras
CRAM (Centroid‑Routing and Adaptive MoE) is a method for Multimodal Continual Instruction Tuning that isolates task‑specific patterns into independent modules to reduce catastrophic forgetting. It uses adaptive‑rank instantiation to allocate only the necessary parameters for new tasks, and centroid‑guided routing with an orthogonality penalty to reuse existing experts while preventing interference. Experiments on diverse benchmarks show CRAM outperforms existing approaches.
By Jun-Tao Tang, Zhen-Hao Xie, Yu-Cheng Shi, Da-Wei Zhou
The paper investigates how the order of candidate documents in a prompt affects the decisions made by large‑language‑model (LLM) scorers, even when their ranking quality is similar. It shows that five scorers with only a 0.010 nDCG@10 difference can produce retained‑set overlaps as low as 0.66–0.84, and that existing rerankers still exhibit significant order dependence. The authors propose Order‑Consistency SFT (OC‑SFT), a training method that reduces this dependence, maintaining ranking quality while improving decision stability across multiple tasks.
By Markus Frohmann, Mahdiyar Alavi, Elizabeth Lingg, Navid Rekabsaz
The paper introduces sLoTh, a parameter‑efficient continual learning framework for sparse event‑based vision transformers. sLoTh freezes the backbone and limits plasticity to low‑rank attention updates (seLoRA) and shared neuronal threshold modulation, updating less than 1% of parameters without replay buffers. Experiments on CIFAR‑100, Tiny‑ImageNet, ImageNet‑100, and ImageNet‑R show competitive rehearsal‑free performance across up to 100 tasks while achieving roughly 6.5× lower energy consumption than dense vision transformers.
By Vaishnavi Nagabhushana, Kartikay Agrawal, Ayon Borthakur
PercepCap is a video captioning framework that explicitly models spatio‑temporal perception before generating captions. It follows a perceive‑describe chain, first producing a perception trace of object trajectories and temporal events, then generating the final caption conditioned on that trace. The method uses a two‑stage training strategy—supervised fine‑tuning followed by perception‑grounded reinforcement learning—and builds caption‑aligned perception data to ensure the perception trace and caption refer to the same objects and events.
By Yifan Xu, Zihao Wang, Zhixiao Wang, Jiaming Zhang, Yichun Yang, Desen Meng, Yuanxing Zhang, Pengfei Wan, Limin Wang
TelecomGPT‑R1‑9B is an open‑source large language model designed specifically for telecom reasoning tasks. It was trained on a 67,427‑example supervised fine‑tuning corpus that covers protocol, knowledge, modeling, and fault reasoning, and further refined with a two‑stage post‑training process involving low‑rank adaptation and policy optimization. The model tops the GSMA open telco leaderboard and matches state‑of‑the‑art closed‑source reasoners across seven public telecom benchmarks.
By Bohao Wang, Chenwei Wu, Haoyu Li, Hang Zou, Yu Tian, Lina Bariah, Li Wei, Chongwen Huang, Yongliang Shen, Zhaoyang Zhang, Merouane Debbah
The paper introduces Echo-GRPO, a method that rewrites privileged reasoning traces into a model’s own idiolect to align off‑policy supervision with the student policy’s vocabulary. By preserving semantics through Dual‑Reference Decoding, Echo‑GRPO mitigates gradient clipping on critical reasoning tokens and improves reasoning distillation. The approach is instantiated as VideoEcho‑R1 for video reasoning, yielding consistent gains across multiple multimodal LLM backbones and benchmarks, and it can be applied as a plug‑in to both RL and supervised fine‑tuning frameworks.
By Ji Soo Lee, Jinyoung Park, Seohyun Lee, Jongha Kim, Joonmyung Choi, Jinsung Yoon, Hyunwoo J. Kim
The paper introduces Visual Retrieval Heads (VRHs), a small fraction of attention heads in vision‑language models that are causally responsible for grounding text descriptions to image regions. By recasting head‑scoring methods and evaluating across eleven VLMs and five benchmarks, the authors show that masking the top 20 VRHs can drop grounding accuracy by up to 80 percentage points, while random masking has little effect. VRHs generalize across various visual reference tasks, preserve output format while corrupting localization, and transfer causally across models sharing an LLM backbone.
By Chanho Park, Daehyeon Choi, Jihyun Lee, Minhyuk Sung
The study evaluates literature reviews produced by large language models (LLMs) using short and long context windows, assessing their quality across 15 dimensions. Results show that while larger context windows allow LLMs to incorporate more information and maintain coherence, they also increase repetition, omission of key works, and a tendency toward descriptive rather than synthetic content. Human oversight remains essential for meeting academic publishing standards, and the authors suggest future work should blend human expertise with AI to mitigate these limitations.
By Muhammad Ali Chaudhry, Xinyuan Hao, Haifa Alwahaby
The paper introduces FrameFT, a parameter-efficient fine-tuning method for transformer models that represents weight updates using sparse coefficients in a Fusion Frame basis. This approach reduces memory usage by storing only the sparse coefficients, leading to significant compute advantages and formal convergence guarantees. Experiments on language and vision tasks show that FrameFT matches or surpasses state‑of‑the‑art PEFT techniques while requiring far fewer trainable parameters.
By Harshavardhan Adepu, Li Zhang, Sanjiv Kumar, Vikas Singh