The paper introduces Prototype-Mediated Process Supervision (PMPS), a method that uses learnable reasoning prototypes to provide structural supervision for latent chain-of-thought embeddings, addressing representation collapse and uneven information distribution. PMPS aligns latent and explicit CoT embeddings in a shared prototype space via many-to-many soft assignment and employs a Progressive Sequential Alignment module to guide training from positional priors to adaptive matching. Experiments show PMPS reduces output token length to under 50% of explicit CoT on GSM8K-Aug and improves accuracy by 2.08% over SIM-CoT, even surpassing CoT-SFT on GPT-2 and achieving the highest accuracy among latent reasoning methods on larger models and harder tasks.
By Yiqi Li, Xu Chen, Chen Ju, Jiangchao Yao, Zhaoyang Li, Jinsong Lan, Xiaoyong Zhu, Bo Zheng, Yu Wang
The paper introduces TRACE, a digital‑advertising diagnostic environment that uses simulated interventions to generate verifiable rewards for training reasoning agents. By injecting controlled interventions into a simulator, the hidden cause of anomalies becomes an oracle label, enabling agents to learn to identify root causes and affected segments through noisy, confounded evidence. Experiments show that reinforcement learning with these synthesized rewards outperforms large prompted baselines, achieving higher accuracy while using fewer tool calls.
By Rui Sun, Zhan Shi, Bing He
The paper investigates how large language models (LLMs) share a common Fisher‑Rao geometry in their next‑token probability distributions, revealing that behaviour largely determines this geometry while activation geometry depends on coordinate choices. Across transformer, state‑space, and recurrent architectures, output geometries align more closely than activation geometries, and this shared structure facilitates semantic‑category transfer and improves agreement with human word choices as models scale and train. The study further demonstrates that geometry can guide minimum‑disturbance interventions, enabling reusable control that preserves behaviour better than Euclidean methods and enhances steering, editing, attribution, dictionary learning, and fine‑tuning.
By Dario Picozzi
Nuha‑Speech is a new initiative aimed at creating general‑purpose Arabic speech‑large language models (speech‑LLMs). It includes the construction of a large Arabic Speech Question‑Answering corpus with over 1.5 million samples for instruction tuning, supervised fine‑tuning of Qwen‑Omni model variants at various scales, and a systematic evaluation framework with diverse tasks and tailored metrics. The project seeks to establish foundational infrastructure for Arabic speech‑LLMs amid limited Arabic speech resources.
By Yingzhi Wang, Reem Alhazzani, Muhammad Alqurishi
The paper proposes a new technique that averages the logits of a frozen reference policy (such as one obtained via supervised fine‑tuning) and a trainable policy, integrating this approach into Group Relative Policy Optimization (GRPO). Unlike methods that use Kullback–Leibler regularization or a critic, the proposed method couples the policies solely through logit averaging, aiming to combine the reasoning strengths of the trainable policy with the formatting benefits of the reference policy. Experiments on MATH, cn‑k12, and MMLU demonstrate that this approach achieves higher or comparable accuracy to the standard KL‑regularized GRPO.
By Xingwei Gan, Ying Zhu
The paper investigates correctness‑gated multi‑teacher distillation, comparing a weighted arm to unfiltered distillation across eight experimental arms. While the weighted arm shows modest gains in accuracy (+0.1660) and macro‑F1 (+0.1323) and a reduction in unsafe action rate (−0.4979), it also exhibits lost label functionality, such as zero Refuted recall and over‑assignment of NotEnoughInfo. A subsequent grounding audit was inconclusive, failing to demonstrate a clear improvement in evidence grounding or overall system performance.
By Xiaofei Feng
The paper introduces SHAPER, a self‑evolving framework that enables embodied agents to adapt to new environments without retraining the underlying foundation model. SHAPER keeps model parameters frozen and instead evolves reusable skills and a context‑code harness through rollouts in the target environment, allowing the same model to act as both planner and optimizer. Experiments on VLABench and ESI‑Bench demonstrate that this skill‑and‑harness optimization outperforms pure execution, supervised fine‑tuning, and test‑time scaling baselines, showing that self‑evolving adaptation is viable when model training is costly or impractical.
By Peidong Wang, Zhiming Ma, Ying Chang, Xufang Luo, Yiqun Zhang, Zihan Wang, Xiaocui Yang, Shi Feng, Yuqing Yang, Dongsheng Li
NCP-ArchPreview is a latent‑space language model that extends standard next‑token prediction (NTP) with a Next Concept Prediction (NCP) objective, allowing the model to predict discrete concepts spanning multiple tokens. The architecture builds a product‑quantized concept vocabulary from hidden states, uses a dedicated Concept Module to forecast future concepts, and feeds these predictions back to guide token‑level generation, all trained jointly end‑to‑end. Trained on 5.73 T tokens with 8.9 B parameters, it achieves the final pretraining loss of OLMo‑3‑7B using only 51.3 % of the tokens, outperforms OLMo‑3‑7B on downstream tasks (including a 5.99‑point GSM8K gain), and demonstrates that the learned latent space enables lightweight domain adaptation and improved drafting performance.
By NCP Team, Jiaqi Cao, Chiyu Chen, Shuang Cheng, Xu Cheng, Beiya Dai, Yufan Feng, Kewen Ge, Ruijun Ge, Jiayi Huang, Yang Jiao, Dahua Lin, Zhouhan Lin, Yifan Liu, Yuliang Liu, Biqing Qi, Mowen Ruan, Junzhe Shen, Yunchong Song, Hao Sun, Zhongbo Tian, Yixuan Wang, Rubin Wei, Jiaxin Xiong, Kangyu Yang, Qian Yao, Qi Zhang, Bowen Zhou
The study investigates how fine‑tuning large language models on synthetic stories can imprint human character traits onto AI assistants. Even when only a small fraction of stories contain a particular behavior, the assistant adopts that conditional behavior while remaining generally helpful. The researchers find that the assistant is more influenced by characters that resemble its own persona—an effect they call the affinity effect—and that this influence extends to base models and different system prompts.
By Jorio Cocola, Lev McKinney, Harry Mayne, Jan Betley, Owain Evans
The paper introduces a framework for out-of-distribution (OOD) detection that addresses the trade‑off between detection performance and classification accuracy caused by fine‑tuning with auxiliary outlier data. It optimizes three factors—model reminder, data sampling, and representation learning—by proposing Self‑Knowledge Distillation to preserve accuracy, Semi‑hard Outlier Sampling to enhance detection with minimal data, and Outlier‑aware Supervised Contrastive Learning to improve ID‑OOD separability. The combined approach yields cumulative gains, outperforming existing methods on diverse benchmarks, especially in long‑tailed scenarios, and offers a robust baseline for real‑world OOD detection.
By Hyunjun Choi, JaeHo Chung, Hawook Jeong
The Eloquence team presents three methods for the Interspeech 2026 MLC‑SLM Task 2, a multilingual MCQA challenge covering 21 languages. They fine‑tune Voxtral‑Mini‑3B with LoRA and data augmentation, achieving 0.72 macro‑accuracy; they use multimodal in‑context learning on Voxtral‑24B to correct label bias, reaching 0.81; and they deploy a training‑free retrieval system with a voice‑anchored memory, scoring 0.68. All approaches surpass the official baseline.
By Jordi Luque, Lorenzo Concina, Marco Matassoni, Alessio Brutti, Filippo Vella
The paper introduces Murmur2Vec, a lightweight, alignment‑free embedding that uses k‑mer counts hashed with MurmurHash to create a compact representation for biological sequences. It provides a full theoretical analysis, including bias/variance formulas, a Johnson–Lindenstrauss‑style concentration bound, and an excess‑risk bound that clarifies the trade‑off between hash‑table size and classifier performance. Empirically, Murmur2Vec matches or surpasses a fine‑tuned 650M‑parameter ESM‑2 protein language model across several classification tasks, including SARS‑CoV‑2 spike lineage and HIV‑1 Env subtype identification.
By Sarwan Ali, Taslim Murad, Imdadullah Khan, Safi Faizullah
The paper investigates UAV‑mounted reconfigurable intelligent surface (RIS) assisted device‑to‑device (D2D) communication with stochastic link activation. It models UAV motion, attitude, time‑varying Rician angles, and angle‑dependent RIS reflection, and formulates a joint optimization of UAV trajectory, attitude, and RIS phases to maximize average sum rate under mobility, energy, and hardware constraints. The authors employ deep reinforcement learning and a Decision Transformer trained on expert trajectories from multiple scenarios, showing that zero‑shot transfer outperforms direct DRL transfer and that online fine‑tuning achieves competitive performance with fewer interactions.
By Yaxuan Liu
BruNet is a new cross‑domain transfer framework for automatic bruise segmentation that combines a ViT‑based visual encoder (either self‑supervised DINOv3 or pretrained LingBot‑Vision) with a SAM‑based mask decoder. The model is trained on the HAM10000 skin lesion dataset and evaluated on a separate bruise dataset without any fine‑tuning, achieving superior performance over CNN‑based models, state‑of‑the‑art segmentation models, ChatGPT‑4o/5‑assisted SAM2 zero‑shot baselines, and the medical‑oriented MedSAM. This work represents the first study to address pixel‑level localisation of bruises, demonstrating strong cross‑domain generalisation.
By Qiming Wang, Richard J. Motley, Ebube E. Obi, Xianfang Sun, Paul L. Rosin
The paper introduces Perturbation, a method that treats representations in language models as learning conduits rather than activation patterns. By fine‑tuning a model on a single adversarial example and observing how this perturbation spreads to other inputs, the approach avoids geometric assumptions and does not identify representations in untrained models. In trained models, Perturbation uncovers structured transfer across multiple linguistic scales, indicating that language models generalize along representational lines and acquire linguistic abstractions through experience.
By Joshua Rozner, Cory Shain
The study evaluates time‑series foundation models for continuous glucose monitoring (CGM) forecasting across eight public datasets covering Type 1, Type 2, and non‑diabetes populations. Zero‑shot foundation models did not consistently beat strong task‑specific baselines, but lightweight fine‑tuning of models like Chronos‑Bolt improved root‑mean‑square error by up to 18% in both in‑distribution and out‑of‑distribution settings. Incorporating multimodal dietary context via CGMacros and a residual‑based fusion framework further reduced overall RMSE by ~3% and postprandial RMSE by ~15%, indicating that dietary signals add clinically meaningful value beyond CGM alone.
By Bowen Zhang, Hsiu-Wen Cheng, Hongyu Yang, Evie L. Shen, Joleen Vansomphone, Yuna Li, Kerry Zhou, Zitian Qu, Suning Zhao, Xiangning Deng, Hua Zhou, Jin J. Zhou
The study investigates how the rank of Low‑Rank Adaptation (LoRA) affects diffusion model fine‑tuning on CIFAR‑10 using a DDPM U‑Net. Experiments with ranks 2, 4, 8, 16, and 32 show that moderate ranks—particularly rank 4—yield the best FID scores while keeping trainable parameters, runtime, and GPU memory low. Higher ranks offer only marginal improvements at a higher computational cost, suggesting that small‑to‑moderate ranks are efficient defaults for fixed training budgets.
By Iman Khazrak, Narges Nejad, Mostafa M. Rezaee, Robert C. Green II
The paper introduces a divergence-based similarity function (DSF) for multi-view contrastive learning, representing each set of augmented views as a distribution and measuring similarity via distribution divergence. DSF captures joint structure across all views, outperforming prior pairwise methods on tasks such as kNN classification, linear evaluation, transfer learning, and distribution shift. It also offers greater efficiency and eliminates the need for a temperature hyperparameter, unlike cosine similarity.
By Jaehyoung Jeon, Cheolsu Lim, Myungjoo Kang
The paper "Beyond Training: A Feasibility Taxonomy for Inference-Time AI Governance" presents a taxonomy of twenty inference‑time mechanisms for monitoring, verification, and enforcement, each evaluated on a four‑point readiness scale using evidence from four vendors. It applies this taxonomy to a two‑dimensional adversary model and maps the mechanisms to four governance scenarios, finding that most mechanisms are commercially available but only adequate against cooperative or low‑to‑medium‑capability users, not high‑capability state‑level deployers. The study also links inference‑stage controls to hardware‑stage mechanisms through a substitution principle and reports a second‑rater reliability of 0.74.
whyItMatters":"The work identifies the current gaps and readiness of inference‑time governance tools, highlighting that existing mechanisms are insufficient against powerful adversaries and thus informing future regulatory and technical development."
By Samar Ansari
The paper presents a pipeline that leverages large language models to extract grammatical rules, example sentences, and lexicons from descriptive grammar books, producing synthetic parallel corpora for fine‑tuning machine translation models. Evaluated on three low‑resource languages—Kalamang, Tuatschin, and Mandan—the synthetic data improves translation quality over seed‑data baselines in 75% of configurations for Kalamang and 59% for Tuatschin, achieving up to +8.8 ChrF++ gains. A factorial study across 96 configurations identifies which combinations of target part‑of‑speech, retrieval granularity, and sample volume drive performance gains and where they fail, demonstrating that static linguistic documentation can be repurposed for practical translation tools for severely under‑resourced languages.
By Varun Ghat Ravikumar, Sina Ahmadi, Lena J\"ager, Rico Sennrich