The paper presents a synthetic pipeline that converts Japanese care handoffs into six-field structured notes, using 182 synthetic audio clips to adapt a 1.47 B‑parameter audio model. Parameter‑efficient adaptation via rank‑16 LoRA achieves 97.7 % of the full‑fine‑tuning performance while training only 12.4 M parameters (≈0.85 % of the backbone). Both full tuning and LoRA outperform the unadapted model on a synthetic test set, demonstrating that a compact model can learn a narrow audio‑to‑structure transformation from limited synthetic data.
By Sidi Chang, Peiying Zhu
The paper introduces LambdaDockScore, a protein‑protein docking scoring function that leverages the LambdaLoss loss from Learning‑to‑Rank to improve pose ranking. By fine‑tuning the energy prediction head of DFMDock on 2.9 million decoy poses from the DIPS dataset, LambdaDockScore outperforms the state‑of‑the‑art EuDockScore on CAPRI benchmark targets, achieving higher accuracy in top‑1 and top‑5 predictions. It also enhances ranking for antibody‑antigen and protein‑protein complexes with extreme binding interface sizes.
By Richard Zhu, Darren Xu, Lee-Shin Chu, Jeffrey J. Gray
The paper introduces CRAFT, a method for improving compositional generalization in vision‑language‑action (VLA) models. It addresses the issue where models rely on visual shortcuts during fine‑tuning, leading them to execute demonstrated skill combinations that match observations rather than the instructed ones. By training with counterfactual instruction–observation pairs and transferring supervision through reusable skill representations, CRAFT enhances success on unseen skill combinations while preserving performance on demonstrated ones across multiple VLA models and benchmarks.
By Taegeun Yang, Youngju Na, Yoonki Cho, Sung-Eui Yoon
The paper introduces an unsupervised algorithm that separates content and style in language model outputs, enabling explicit style control. Applied to over 100K traces from nine teacher models, the method uncovers six recurring, imbalanced styles and fine‑tunes student models to follow them. This style conditioning improves Pass@$k$ on six math reasoning benchmarks and shows that style influences problem‑solving success, offering a new avenue for controlling and enhancing model performance.
By Ioana Marinescu, Eric Karl Oermann, Kyunghyun Cho
The paper introduces a continual‑learning framework for single‑view 6‑DoF grasp synthesis with a parallel‑jaw gripper in cluttered scenes. Instead of fine‑tuning a large parametric model, the method updates grasp scores via memory in a learned embedding space and optionally incorporates user demonstrations to generate new candidate grasps. Experiments in simulation and real‑world trials (over 1500 grasps) show that the approach matches baseline performance before adaptation, improves online on unseen objects, and achieves over 90% success on challenging categories after just 50 online attempts.
By Giulio Schiavi, Andrei Cramariuc, Michael Pantic, Roland Siegwart
FedFit introduces a federated fine‑tuning framework for large language models that reduces communication overhead by using a disjoint shared vector‑bank parameterization to reconstruct adapter matrices from two compact global vector banks. It resolves the aggregation dilemma between Sum‑of‑Products and Product‑of‑Sums through an alternating optimization schedule that alternates between accurate single‑bank updates and joint updates corrected by a Residual Spectral Aggregation mechanism. The method also incorporates blockwise quantization with client‑side error feedback and provides theoretical convergence guarantees, achieving perplexity comparable to standard federated LoRA while delivering up to 100× higher compression ratios on Qwen2.5 models.
By Hang Zou, Chao Zhang, Yuzhi Yang, Yu Tian, Samson Lasaulce, M\'erouane Debbah
The paper introduces a joint simulation–experiment training framework that treats simulation and experimental data as separate objectives in a multi‑objective learning problem. Experiments on four fluid systems show that joint training outperforms both simulation‑only and experiment‑only baselines, as well as the conventional simulation‑to‑experiment fine‑tuning approach, by achieving a more balanced performance across domains and better retaining simulation‑specific information. The authors demonstrate that joint training preserves simulation‑only fields that are absent from experimental measurements, leading to improved overall predictive accuracy.
By Mahindra Rautela, Alexander Scheinker, Ayan Biswas, Diane Oyen, Nathan DeBardeleben, Earl Lawrence
Varda‑single‑1.0 is a medium‑range, data‑driven weather prediction system designed for Switzerland’s Alpine region, delivering hourly deterministic forecasts at 1 km resolution and global forecasts at 31 km. It uses two independently trained stretched‑grid Graph Transformer models—an autoregressive forecaster and a temporal downscaler—trained through a curriculum that starts with ERA5 reanalysis, proceeds to a 20‑year regional reanalysis, and ends with fine‑tuning on operational analyses. Over a one‑year verification, Varda‑single matches or surpasses MeteoSwiss’s numerical weather prediction baselines for most key metrics, though it underestimates local wind peaks and produces smoother convective precipitation fields.
whyItMatters":"The system demonstrates that high‑resolution machine‑learning models can rival traditional numerical weather prediction in complex terrain, offering a complementary tool for operational forecasting and research."
By Alberto Pennino, Francesco Zanetta, Michele Cattaneo, Claire Merker, Radi Radev, Jonas Bhend, Louis Frey, Hugues de Laroussilhe, Oph\'elia Miralles, Carlos Osuna, Daniele Nerini, Andreas Pauling, Daniel Hupp, Ulrich Hamann, Mary McGlohon, Marti Bosch, Luca Lanzilao, Marco Arpagaus, Lukas Jansing, Daniel Leuenberger, Mark A. Liniger, Katrin Ehlert, Matthew Chantry, H\aa{}vard Homleid Haugen, Gert Mertes, Ana Prieto Nemesio, Mario Santa Cruz, Jasper Wijnands, Gabriel Moldovan, Harrison Cook, Oliver Fuhrer
Invent-A-Dataset is a prompt‑based system that generates realistic, large‑scale datasets from a description, targeting the zero‑data regime where no initial data exists. The authors benchmarked it against five leading model APIs across eight task types and up to 20,000 samples, finding that it outperforms competitors with 17% higher quality and 19% more diverse samples. The diversity advantage grows with dataset size, leading to better downstream training performance and consistently higher rankings for fine‑tuned models.
By Shivalika Singh, Andrija Djurisic, Gbemileke Onilude, Sudip Roy, Sara Hooker
The paper introduces Function-Structured Graph Reinforcement Learning (FSG‑RL), a framework that links subproblem graphs to Python code and uses multiple verifiers for feedback. It first trains a policy via supervised fine‑tuning to generate code from function graphs, then refines it with Group Relative Policy Optimization (GRPO) that employs answer‑gated rewards and span‑level credit assignment. On a benchmark combining GSM8K, MathQA, MATH, and Omni‑MATH, GRPO raises final‑answer accuracy from 43.25 % to 67.50 % and full‑solution success from 32.25 % to 52.25 %, with further improvements when teacher supervision is added.
By Zihan Liu, Xurong Xie
The paper introduces LoRA‑Norm, a post‑training normalization technique for Low‑Rank Adaptation (LoRA) that rebalances the gains of learned singular directions without altering the directions themselves. LoRA‑Norm uses spectral rebalancing and nuclear‑norm restoration to preserve total spectral mass, requiring no calibration data or extra training and adding no inference overhead. Experiments on two backbones and three adaptation tasks show that LoRA‑Norm improves both specialization and capability retention, outperforming other post‑hoc spectral pruning and gradient‑guided editing methods.
By Zailong Tian, Yanzhe Chen, Zhuoheng Han, Houfeng Wang, Lizi Liao
The paper investigates how fine‑tuning large language models with a small number of harmful examples can erode their refusal behavior, and explores whether localizing safety‑related behavior to specific layers or directions can provide robust defenses. Experiments across six checkpoints from four model families show that harmful and benign prompts remain linearly separable after attack, and that patching clean hidden states or freezing layers up to a transition depth can restore refusal. However, attackers can bypass these defenses by spreading updates or targeting singular directions, indicating that adaptive fine‑tuning can defeat localized repairs and highlighting the need for multiple defensive checks.
By Jungseob Lee, Dongyub Jude Lee, Sugyeong Eo, Seongtae Hong, Seungyoon Lee, Heuiseok Lim
GLoC-EHR is a multimodal language model that processes electronic health records by combining a fixed-size global memory of the entire patient trajectory with a local memory of selected events. It generates hospital-course summaries and masked concept descriptions, then is fine‑tuned to cite evidence before answering clinical questions, using group relative policy optimization to reward correct, evidence‑supported responses. On MIMIC‑IV outcome tasks, GLoC‑EHR achieves the highest macro AUROC among compared models when answering directly, and maintains strong performance with evidence‑cited reasoning while adding distinct supported findings from the local memory.
By Chaiho Shin, Kwangsoo Kim
The paper introduces TACO, a new optimizer for fine‑tuning large language models that drastically reduces optimizer state memory while preserving first‑order gradients. TACO selects the sign of the largest magnitude entry in each column of weight matrices, achieving a 174× reduction in persistent optimizer memory compared to AdamW8bit and a 2.9× decrease in peak training memory on OPT‑13B. This allows full‑parameter fine‑tuning of 30–32B‑parameter models on a single 80 GB GPU across multiple model families and tasks, with comparable accuracy and runtime to existing methods.
By Jichao Jiang (University of Central Florida), Cristian McGee (University of Central Florida), El Houcine Bergou (Mohammed VI Polytechnic University), Hanqin Cai (University of Central Florida), Aritra Dutta (University of Central Florida)
The paper reports a reproducible study of evasion attacks on image and text classifiers. A compact convolutional network on MNIST achieved 98.63% clean accuracy but dropped to 60.20% under FGSM with ε=0.15 and 1.72% with ε=0.30, while PGD reduced accuracy to 32.47% and 0.41%; a bit‑depth‑reduction defense only partially restored performance. In contrast, a DistilBERT model fine‑tuned on the SMS Spam Collection reached 98.75% accuracy and 94.96% F1‑score, yet a sequence of predefined perturbations produced only modest probability shifts and did not flip spam to ham predictions.
By Parker Hummel (Minot State University), Ryne Skabo (Minot State University), Muhammad Abusaqer (Minot State University)
The study evaluates 4‑bit quantization and low‑rank adapter fine‑tuning (QLoRA) on several large protein language models, finding that many model‑task pairs retain over 90% of full fine‑tuning performance while achieving up to 90% GPU memory savings. QLoRA preserves early‑layer representations and induces task‑specific changes in later layers, closely resembling full fine‑tuning with smaller representational shifts. For generative models, 4‑bit quantization largely maintains structural and sequence‑level properties, though token‑level analysis reveals model‑dependent changes in autoregressive output distributions.
By Ilan Yaniv Zeisler, Sebastian Clancy, Pouriya Bayat, Saaim Raad, Ivan Kraskov, Matthew Xie, Vivian White, Spencer Perkins, Serena Singh, Sepehr Bayat, Keith Pardee
ScaffoldM3C is a lightweight, multimodal, auto‑regressive framework that generates stable 3D block constructions by treating the task as a probabilistic next‑block generation problem. It incorporates text, image, and sketch conditioning, introduces a scaffold block token to aid intermediate stability, and uses Sequential Monte Carlo to explore multiple assembly sequences simultaneously. The model is four times smaller than existing baselines, achieving 5‑ to 20‑fold inference speedups while matching or surpassing state‑of‑the‑art construction quality and stability in both simulations and real‑world robot demonstrations.
By Gadiel Sznaier Camps, Chengyang He, Guillaume Sartoretti, Eduardo Montijano, Mac Schwager
The paper extends mechanistic interpretability of large language models by modeling concepts as low‑dimensional non‑linear manifolds rather than linear subspaces. It introduces a concept‑based alignment (CBA) score to compare these manifolds across layers and models, revealing block structures in intermediate layers, a shift from syntax‑dominated to mixed syntactic‑semantic concepts, and training‑dependent multilingual sharing. The study also shows that alignment patterns differ across model families and training stages, with adjacent stages aligning more closely than distant ones.
By Tido Specht, Elias Benedict Krey, Nils Neukirch, Nils Strodthoff
arXiv:2610. 00436v1 Announce Type: new Abstract: Online batch selection fine-tunes a language model on the most useful part of each candidate batch.
By Hongyu Chen, Xinyi Luo, Ming Zhao, Lin Tang, Zihan Xu, Jing Li, Yuxuan Wang, Haoran Deng, Wei Zhang
The paper investigates how pre‑training on random input‑output mappings (memorization tasks) can transfer to downstream tasks. It discovers two unexpected patterns: equivalent transfer, where each pre‑training epoch saves roughly one fine‑tuning epoch, and non‑equivalent transfer, where pre‑training on a mismatched task can be more efficient than training directly on the downstream task. Ablation studies reveal that transfer consists of a trivial magnitude‑driven effect in the last layer and a non‑trivial structure‑driven effect linked to covariance in other layers.
By Yimiao Yu, Florentin Guth