The paper introduces LAST, a Looped Audio Spectrogram Transformer that processes all tokens once and then reuses the same blocks to refine only the class token over fixed audio features, making subsequent passes inexpensive. On AudioSet, a ten‑pass LAST outperforms a twelve‑layer sequential transformer by 2.1% relative mean average precision while using 49.4% fewer parameters, 42% fewer MACs, and achieving 9.8% higher throughput. Increasing the pass count from two to ten improves accuracy with only a 1.2% increase in computation, and the model shows enhanced robustness to temporal masking and other auditory augmentations across music, environmental, and event sound classification tasks.
By Haider Al-Tahan, Sean O'Brien, Anastasia Razdaibiedina, N. Apurva Ratan Murty
ConQuR introduces a lightweight post‑training rotation calibration for large language model activation quantization. By learning orthogonal rotations that align normalized activations with the corners of an inscribed hypercube, the method distributes activation energy evenly and can be updated online without storing activations. Experiments on Llama‑2 and Llama‑3 models (3B–70B) show competitive or improved perplexity and reasoning performance while avoiding costly training or large offline storage.
By Chayne Thrash, Ali Abbasi, Soheil Kolouri
The paper proposes constraining the query and key projection matrices in Transformer attention to the Stiefel manifold and optimizing them with a Riemannian Adam optimizer. It demonstrates that this geometric constraint yields significant performance gains on a CIFAR‑10 patch benchmark, with the constrained model outperforming standard AdamW by up to +6.79 percentage points. The authors also show that weight decay has no effect on the constrained frames and that the improvement is driven by a scale‑free step size rather than the manifold projection or equivariance properties.
By Rub\'en Dar\'io Guerrero
The paper proposes a sensitivity‑aware residual‑stream pruning method for large language models that goes beyond minimizing activation reconstruction error. By using a second‑order approximation of output KL divergence, the authors derive a spectral upper bound that selects pruning subspaces based on both activation covariance and output sensitivity, enabling efficient eigendecomposition. Experiments on instruction‑tuned language models show that this approach consistently reduces calibration KL divergence, improves perplexity, and enhances downstream task performance across various compression levels.
By Chayne Thrash, Kevin Chen, Soheil Kolouri
The paper introduces MERITED, a framework that combines Instance-Based Learning Theory (IBLT) with Energy Based Models (EBMs) to enable metacognitive reasoning about computational effort. It allows an EBM to dynamically allocate resources based on uncertainty estimates, addressing limitations of large language models that cannot predict uncertainty before responding. The authors present a 191‑million‑parameter reasoning EBM and demonstrate how MERITED uses an IBL model for efficient, uncertainty‑driven compute allocation.
By Tailia Malloy, Prateek Kumar Rajput, Serge Lionel Nikiema, Cleotilde Gonzalez, Tegawend\'e F. Bissyand\'e
The paper introduces the Denoising Workload Surface (DWS), a two‑dimensional probability surface that captures the block‑autoregressive generation structure of diffusion large language models (dLLMs). By preserving both output block and within‑block denoising step information, DWS enables a lightweight, prompt‑only predictor to estimate per‑request inference cost accurately, even on a single CPU core. In real‑world serving experiments, DWS reduces cost‑prediction error by up to 2.5× and improves end‑to‑end latency for online chatbots by up to 1.92×.
By Haoyu Zheng, Fangcheng Fu, Binhang Yuan, Yongqiang Zhang, Liang Deng, Hao Wang, Yuanyuan Zhu, Xiao Yan, Jiawei Jiang
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
CrossGMN introduces a graph metanetwork that processes a trained source network and an initialized target network simultaneously, enabling equivariant cross‑architecture weight‑space transformations. By preserving symmetry through cross‑network message passing, CrossGMN can refine target network initializations while remaining invariant to source permutations and equivariant to target permutations. Experiments demonstrate that CrossGMN accelerates knowledge distillation, transfers across datasets without retraining, and unifies compression from diverse source architectures into a common target architecture.
By Adir Dayan, Yam Eitan, Haggai Maron
arXiv:2610. 02191v1 Announce Type: new Abstract: While Large Language Models (LLMs) have demonstrated striking capabilities on frontier mathematical problems, it remains unclear whether they possess the structural mathematical understanding underlying their solutions.
By Shuo Xing, Zilin Dai, Chengyuan Qian, Fangzhou Lin, Wenjing Chen, Ping He, Pan Lu, Alvaro Velasquez, Mohit Bansal, Zhengzhong Tu
AURAL is a speech language model that performs adaptive latent reasoning by modeling multiple plausible reasoning continuations in latent space and jointly predicting chunks of future states, thereby reducing sequential forward passes and latency. The authors introduce a large bilingual dataset, AuralReason-683K, containing concise chain‑of‑thought annotations for emotion recognition, empathetic dialogue, and general reasoning, and use reinforcement learning (AURAL‑RL) to reward concise, high‑quality reasoning that adapts to problem difficulty. Experiments on two backbones show that AURAL‑RL matches or exceeds chain‑of‑thought reinforcement learning while achieving significant latency reductions, such as an 11.8× speed‑up on Qwen2.5‑Omni.
"whyItMatters":"The work demonstrates that latent reasoning can match the performance of explicit chain‑of‑thought methods while dramatically cutting response time, addressing the trade‑off between intelligence and speed in speech language models."
By Yuxiang Wang, Kunyu Feng, Yuancheng Wang, Zihang Liu, Shengbo Cai, Qinke Ni, Wan Lin, Tao Feng, Yingda shen, Ming-Hao Hsu, Zhixian Zhao, Liqiang Zhang, Teddy Sun, Steve Yves, Zhizheng Wu
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
The paper examines how large language models (LLMs) express uncertainty compared to humans, noting that humans use verbal markers like "possible" or "likely" to convey metacognitive awareness. By curating a corpus of human uncertainty markers and benchmarking LLMs against it, the authors find that LLMs encode these markers with numerical levels that differ substantially from human usage. They introduce METHODNAME, an optimization-based algorithm that learns an optimal uncertainty profile over verbal markers directly from LLM outputs, enabling a direct comparison of confidence semantics and revealing systematic disparities in verbal expressions.
By Jinhao Duan, Zicheng Liu, Zijie Liu, Kaidi Xu, Tianlong Chen
The paper investigates using short polynomial approximations to accelerate special‑function operations in large language models on NVIDIA Blackwell GPUs. By replacing native sigmoid, tanh, and SiLU with degree‑3 or degree‑4 bfloat16 programs, the authors achieve up to 2.19× speed‑ups in isolated FP16 benchmarks and modest training‑step throughput gains (2.7–8.0%) across four integration tasks. The study also evaluates model behavior, finding negligible training‑loss differences within 100 billion tokens.
By Robert Hu
EchoPress is a training‑free method for pruning key‑value caches in large language models. It approximates the reconstruction attention used by KVzip by leveraging queries and keys from standard prefill, reconstructing only the first chunk to calibrate importance scores for the rest of the context. Experiments on LongBench and RULER with Qwen3‑8B and Llama‑3.1‑8B‑Instruct show that EchoPress matches KVzip’s task accuracy across eviction ratios from 50% to 90%, while reducing compression overhead by 1.7–19.6× and total prefill time by up to 2.9×.
By Jiawei Lin, Saibo Geng, Thomas Bourgeat
The paper introduces attention kernels that replace the exponential function in transformer softmax to better handle operator learning on probability measures with heavy-tailed (polynomial) distributions. Two new benchmarks with closed‑form targets are constructed to evaluate how different kernel growth rates and data preprocessing affect performance. The study finds that slower‑growing kernels prevent ensemble collapse on heavy‑tailed tasks, while softmax with symlog preprocessing only succeeds on a subset of problems, and that all kernels perform similarly on Gaussian data.
By Kailen Hargenrader, Edoardo Calvello, Bohan Chen
The paper shows that the geometric patterns seen in language model embeddings—such as circles for months and saddle-shaped manifolds—arise from group-invariant statistics in word co‑occurrence data. By extending previous work on translation symmetry to arbitrary finite, compact, and homogeneous groups, the authors prove that embeddings correspond to matrix elements of the irreducible representations of the symmetry group. They validate this theory experimentally with the cyclic group <Z_{12}> for months, a dihedral group for musical chords, and a spherical‑harmonic embedding for celestial objects.
By Liam Storan, Andreas Tolias, Nina Miolane
The report introduces AnyJev, a method that extracts typed decisions from a pretrained instruction‑tuned language model by restricting the next‑token distribution to a set of option tokens. It corrects two defects—label bias and position bias—without any gradient updates by dividing out a label prior and averaging log‑probabilities over cyclic rotations of the option list. Experiments on two 20‑option tasks show that rotations reduce order‑flip rates and improve accuracy across all tested models, while a stopping rule reduces the number of required rotations and increases decision throughput.
By Jiamu Zhang, Tianze Yang, Yucheng Shi, Evan Chen, Zixiang Nie, Kelly Wan, Liangjie Hong, Ninghao Liu, Liang Wu
AF‑Muon is an AdamW‑free extension of the Muon optimizer that retains Muon’s matrix update for hidden weights while applying a support‑aware finite‑cap linear minimization oracle to tied vocabulary tables and an RMS‑normalized update for one‑dimensional auxiliary parameters. This design eliminates second‑moment state, reducing optimizer‑state memory by about 20% compared to Hybrid Muon. Across nine tied‑token settings—including decoder‑only language models, T5‑style encoder‑decoders, and ImageGPT‑style variants—AF‑Muon consistently improves mean validation loss and perplexity over both Hybrid Muon and a SCION‑style Sign endpoint, with robust gains confirmed by long‑horizon runs and hyperparameter studies.
By Arash Lagzian, Paniz Halvachi, Junming Zhang, Zhouhan Lin, Dianbo Liu
The paper presents a method for distilling tabular foundation models (TFMs) into lightweight, dataset‑specific students. By using the full labeled training set as teacher context and training students on both observed and synthetic queries, the authors achieve significant performance gains over traditional supervised models on TabArena and TALENT benchmarks. The distilled students also provide substantial inference speedups, reducing the cost of repeated inference.
By Minho Jeong, Dooho Lee, Jinmo Lee, Jaemin Yoo
The paper investigates how treating the optimization objective as a design variable can improve Remaining Useful Life (RUL) prediction in predictive maintenance. Five model architectures are compared under single‑objective and multi‑objective hyperparameter optimization, with the latter using NSGA‑II and Entropy‑CRITIC weighting to balance accuracy and prediction timeliness. Results on NASA C‑MAPSS and BackBlaze datasets show that multi‑objective optimization reduces directional imbalance in predictions and can alter model rankings, highlighting the importance of objective choice in RUL modeling.
By Tugrul Cabir Hakyemez, Ener Uras Gokhan