The study evaluates how well three sentiment‑analysis tools (TextBlob, VADER, Twitter‑roBERTa‑base) and three large language models (Qwen3‑32B, GPT‑OSS‑120B, Llama‑4‑Maverick‑17B) agree with six human raters on 100 tweets. Agreement was measured with Cohen’s and Fleiss’ kappa, revealing only fair inter‑human agreement and higher concordance for binary sentiment labels than for three‑class labels. Twitter‑roBERTa‑base achieved the strongest alignment with humans, especially for negative versus non‑negative sentiment, while the LLMs showed substantial agreement among themselves and moderate to substantial alignment with humans, particularly for positive versus non‑positive classifications.
The findings emphasize that domain‑specific fine‑tuning and human‑centered evaluation are essential for reliable social media sentiment analysis.
By Himarsha R. Jayanetti, Sivakanesan Dhanushkanda, Shuai Hao, Michael L. Nelson, Michele C. Weigle
arXiv:2610.09665v1 Announce Type: new
Abstract: On-policy post-training can improve large language models by learning from their own trajectories, but requires costly rollout generation. We ask wheth...
By Tianle Wang, Jiayu Liu, Ruizhi Zhao, Ning Miao
PrismSSL is a Python library that consolidates leading self‑supervised learning methods for audio, vision, graphs, and cross‑modal tasks into a single, modular codebase. It allows users to install, configure, and run pretext training with minimal code, reproduce compact benchmarks, and extend the framework with new modalities or methods via clean trainer and dataset abstractions. The library, available on PyPI under an MIT license, integrates with HuggingFace Transformers and offers features such as distributed PyTorch training, Optuna hyperparameter search, LoRA fine‑tuning, animated embedding visualizations, Weights & Biases logging, and a Flask‑based graphical dashboard for easy pipeline configuration.
By Melika Shirian, Kianoosh Vadaei, Kian Majlessi, Audrina Ebrahimi, Peyman Adibi, Hossein Karshenas
The paper introduces RASA, a routing‑aware safety alignment framework for Mixture‑of‑Experts (MoE) language models. RASA repairs Safety‑Critical Experts by selectively fine‑tuning only those experts that are disproportionately activated during jailbreak attacks, while keeping routing fixed and enforcing routing consistency with safety‑aligned contexts. Experiments on two MoE architectures show that RASA achieves near‑perfect robustness against diverse jailbreak attacks, generalizes across attacks, reduces over‑refusal, and maintains performance on standard benchmarks such as MMLU, GSM8K, and TruthfulQA.
By Jiacheng Liang, Yuhui Wang, Tanqiu Jiang, Ting Wang
The paper introduces BehaviorTrace, an evaluation framework for assessing training‑data attribution in online reinforcement‑learning fine‑tuning of language models. Using a planted behavior on Qwen2.5‑1.5B, the authors show that many attribution signals are confounded by factors such as gradient magnitude and model fluency, and that only the gradient of trigger tokens consistently aligns with the true behavior across seeds. They provide a checklist for evaluating attribution methods and test existing estimators without proposing a new one.
By Amit Nautiyal
SoccerNet‑FoulRet is a new benchmark for retrieving semantically similar soccer foul videos, addressing inconsistencies in refereeing decisions by allowing comparison of a contentious foul with past precedents. The task requires retrieving relevant fouls regardless of camera angle, teams, or appearance, differing from prior video‑to‑video retrieval that focuses on visual similarity. The benchmark, built from the SoccerNet‑MVFoul dataset, evaluates zero‑shot video and vision‑language embeddings and a fine‑tuned baseline on 693 human‑verified queries, revealing that even the best zero‑shot model achieves less than 5% HitRate@10, highlighting the challenge of semantic foul retrieval.
By Jacobus Arthur, Ahmad Sait, Batool Hani, Merey Ramazanova, Jan Held, Marc Van Droogenbroeck, Bernard Ghanem, Anthony Cioppa, Silvio Giancola
Q-PACE is a dynamic precision allocation method for quantization-aware training that uses a second-order sensitivity model to predict loss increases from quantization noise. By periodically recalculating per-layer curvature coefficients and reassigning precision, it improves mixed-precision training for large language models up to 4 B parameters, achieving comparable loss with lower memory usage. The approach shows that quantization sensitivity is predictable by depth and layer type, enabling infrequent, inexpensive recalibration.
By Alexandra Volkova, Matin Ansaripour, Erik Schultheis, Christoph H. Lampert, Dan Alistarh
FinVector-Market-4B adapts the Qwen/Qwen3.5-4B language model using rank‑16 LoRA on a 22,000‑example financial corpus. On a 600‑example benchmark with implicit and explicit JSON‑schema contracts, the adapted model improves JSON validity from 0% to 91.3% and boosts task‑specific metrics—FinQA exact match from 14.7% to 40.0%, calculator‑expression correctness from 48.0% to 82.7%, scenario branch‑label agreement from 20.1% to 89.5%, and implication‑direction agreement from 52.4% to 87.2%. A post‑hoc audit shows that macro‑F1 differences stem from changing label sets, and the benchmark’s generalization claims are limited by filing overlap and calculator‑target inconsistencies.
By Alina Khaybullina
MaRK (Markov‑adapted Recurrent Kernels) is a framework that conditions pre‑trained State Space Models (SSMs) by mapping context vectors into bounded modulations of the SSM’s recurrence, read‑in, read‑out, skip, and discretization parameters. By treating the SSM as a linear‑parameter‑varying system, MaRK generates a context‑indexed family of Markov parameter sequences, enabling each diffusion timestep to reshape the model’s input‑output memory kernel. Implemented on a frozen 111 M‑parameter Hydra SSM, MaRK uses low‑rank adapters (Hypernet, Chebyshev polynomial, or Discrete Cosine Transform) to achieve parameter‑efficient fine‑tuning with only 6.3–11 M trainable auxiliary parameters, while providing an analytic stability certificate and demonstrating stable, timestep‑conditioned memory profiles in experiments.
By Syed Ibrahim Omer, Ginny Y. Wong. Xiangyu Zhao
The paper introduces CHASE, a framework that leverages Geometric and Spectral Alignment (GSA) to inform practical model design. GSA characterizes trained networks via spectral concentration, channel alignment, support structure, and singular basis changes, and CHASE applies these insights to six applications including parameter‑efficient fine‑tuning, structured pruning, and model merging. New methods—CAGA, SAKV, and CAPS—use GSA to share attention heads, compress KV caches, and prune structured outputs, demonstrating improved performance over baselines.
By Wei Wang, Wei Jiang, Ziran Liu
AutoAdapt is a modular framework that automatically discovers latent domains and trains per-domain Low‑Rank Adaptation (LoRA) adapters independently in parallel, without modifying other adapters. It uses parameter‑free routing to integrate new domains, achieving performance comparable to a LoRA adapter trained on all domains while avoiding full‑model retraining. The approach prevents domain interference by construction, enabling taxonomy‑free domain specialization across 14 benchmarks and GPT‑4o evaluations.
By Josh McGiff, Salma Mekaoui, Robert Shanahan, Nikola S. Nikolov
ORDERS is a personalized federated learning configuration that combines a shared backbone, private residual adapters, and a classifier, using geometric weights based on descending update norms, feature alignment, and private‑parameter perturbations. In extensive experiments on CIFAR‑10 and Sent140, ORDERS achieves slightly higher mean client accuracy than baseline methods, with a 5.47% reduction in parameter payload. Ablation studies show limited, dataset‑dependent benefits from norm ranking and alignment, and no clear advantage from perturbations.
By Koffka Khan
The paper introduces Q‑Learning with Scalar Adjoint Matching (SQAM), a method that replaces expensive per‑step vector–Jacobian products in flow policy fine‑tuning with a closed‑form scalar adjoint that scales the value gradient by flow time. SQAM also incorporates a value penalty on policy‑generated actions, which is shown to be crucial for effective learning. Experiments demonstrate that SQAM outperforms strong baselines on the hardest OGBench domains and improves supervised fine‑tuning on a real bimanual robot with a vision‑language‑action policy.
By Yonghoon Dong, Minsung Yoon, Jaehyuk Kim, Jungwoo Park, Changyeon Kim, Jinwoo Shin
The paper introduces the Expected Decoding Rounds (EDR) objective, a theoretical framework that models speculative decoding as a Markov reward process and directly optimizes the number of decoding rounds. EDR weights local rejection costs by state occupancies, eliminating auxiliary hyperparameters and enabling unbiased stochastic optimization via temporal‑difference gradients. Applying EDR to fine‑tune state‑of‑the‑art draft models (DSpark and DFly) improves mean accepted length and outperforms prior training objectives across nine benchmarks in math reasoning, code generation, and chat.
By Yunxiao Zhao, Changxiao Cai
The paper presents a plug‑and‑play framework that uses concept-based interpretation to debug medical imaging models. By aligning a single‑modality encoder with BioMedCLIP, the authors build a Concept Bottleneck Model that allows concept‑level interventions to distinguish causal from spurious features. Experiments on Mayo Clinic ultrasound and CheXpert chest X‑ray data show that these interventions can reliably diagnose model issues and even improve performance through guided fine‑tuning.
By Samrajya Thapa, Daniel J. Quest, Timothy L. Kline, Carrie L. Langstraat, Emanuel C. Trabuco, Wei Le
The paper introduces Physics‑Aligned Electronic Ground‑State Models (GSMs) that sit between machine‑learned interatomic potentials and Kohn‑Sham density functional theory in terms of computational cost and accuracy. By aligning GSM learning objectives with KS‑DFT equations and enforcing physical constraints, the authors develop new training techniques—OrthoNormal‑Loss (ON‑Loss), Grassmann Restricted Occupied‑Orbital Training (GROOT), and Residual Optimal‑gauge Conditioning‑aware KS‑Eq. Training (ROCKET)—which dramatically reduce energy and force errors in size‑extrapolation tests from QM9 to QM40. The models also achieve near‑chemical‑accuracy predictions on reactive chemistry benchmarks and demonstrate efficient label‑free self‑consistency fine‑tuning.
whyItMatters":"The work advances the cost‑accuracy frontier for electronic structure predictions, enabling more reliable out‑of‑distribution generalization in drug and material discovery."
By Eike S. Eberhard, Xaver Kainz, Viktor Kotsev, Abdulrahman Aldossary, Stephan G\"unnemann
The paper presents cross‑domain pretraining for neural surrogates in steady‑state computational fluid dynamics (CFD). By training models across varied geometries, boundary conditions, and fidelities, the authors show that such pretraining improves zero‑ and few‑shot performance compared to training from scratch or transferring from domain‑specific experts. Finetuning a pretrained cross‑domain model can reduce errors by 2–3× at the same sample size and require eight times fewer samples to reach the same error, with benefits that grow with model size and dataset diversity.
By Anthony Zhou, Amir Barati Farimani, Shirley Ho, Rudy Morel
HE‑OFT is a cryptographically secure one‑shot federated fine‑tuning protocol that protects privacy by exchanging only one encrypted contribution and never revealing the trained model. Each client fine‑tunes a low‑rank adapter and a classifier head on a frozen public backbone, keeps the adapter, and uploads an encrypted head displacement that the server combines using multiparty CKKS without decryption. On four text classification tasks and one vision task, HE‑OFT achieves 61–79 % accuracy, retaining 85–96 % of the accuracy of a disclosed model, while a test‑time query takes 443.1–1713.1 s on one core or 56.1–255.1 s on a GPU, with level restoration reducing traffic from 1.6 GiB to 13.5 MiB.
By Halil \.Ibrahim Kanpak, Sinem Sav, Alptekin K\"up\c{c}\"u
The paper investigates whether different parameter‑efficient fine‑tuning (PEFT) methods truly differ in their approach and impact. By comparing six methods on language and diffusion models, the authors find that many LoRA‑family techniques approximately preserve pretrained geometry, and that restoring slightly drifted singular‑value spectra largely maintains task performance, challenging the need for explicit geometric preservation. They also observe distinct adaptation–retention trade‑offs: LoRA limits forgetting best, DoRA scores higher on average, and PiSSA incurs more retention costs, with performance gains linked to modifications in specific spectral components.
By Yikuan Li, Pinyan Lu, Fanghui Liu
arXiv:2610.09985v1 Announce Type: cross
Abstract: We study a sequential pricing problem in which a seller jointly posts a price and an advertisement generated by a large language model (LLM). The sel...
By Alessandro Barro, Francesco Bacchiocchi, Francesco Emanuele Stradi, Alberto Marchesi