Full Waveform Inversion (FWI) is a highly nonlinear and ill-posed problem that aims to recover subsurface velocity maps from surface-recorded seismic waveforms data. Existing data-driven FWI typically uses small models, as available datasets have limited volume, geological diversity, and spatial extent, leading to substantial concerns about overfitting. Although they perform well on synthetic datasets, current methods fail to generalize to more realistic geological structures. In this work, we show that a model trained entirely on simulated and relatively simple data can generalize remarkably well to challenging and unseen geological benchmarks. We provide a working recipe that tames a billion-parameter model for FWI through coordinated scaling across three axes: model capacity, data diversity, and training strategy. Our model achieves state-of-the-art performance on OpenFWI and significantly narrows the generalization gap in data-driven FWI. Across six challenging geophysical benchmarks, including Marmousi, 2D SEG/EAGE Salt and Overthrust, 2004 BP, Sigsbee, and SEAM Phase I, it infers complex structures absent from the training set and delivers significant performance improvements (SSIM from 0.5844 to 0.7669). Overall, our results demonstrate that with an appropriate scaling strategy, large models trained on simple synthetic data can achieve substantial generalization to more complex and realistic geological structures.

We introduce DLT-Corpus, the largest domain-specific text collection for Distributed Ledger Technology (DLT) research to date: 2.98 billion tokens from 22.12 million documents spanning scientific literature (37,440 publications), United States Patent and Trademark Office (USPTO) patents (49,023 filings), and social media (22 million posts). Existing Natural Language Processing (NLP) resources for DLT focus narrowly on cryptocurrency price prediction and smart contracts, leaving domain-specific language underexplored despite the sector's ~$3 trillion market capitalization and rapid technological evolution. We demonstrate DLT-Corpus' utility by analyzing patterns of technology emergence and market-innovation correlations. Findings reveal that technologies first appear in our scientific literature subset before reaching patents and social media, following traditional technology transfer patterns. While social media sentiment remains overwhelmingly bullish even during crypto winters, scientific and patent activity grows less tied to short-term sentiment, tracking overall market expansion in a virtuous cycle in which research precedes and enables economic growth that, in turn, funds further innovation. We release the DLT-Corpus and companion artifacts: LedgerBERT (+23% over BERT-base on DLT-specific Named Entity Recognition (NER) task), a sentiment analysis dataset of 23,301 crypto news headlines and descriptions, tools, and code.

Discrete diffusion language models (dLLMs) provide a fast and flexible alternative to autoregressive models (ARMs) via iterative denoising with parallel updates. However, their evaluation is challenging: existing metrics conflate denoiser approximation error with sampler-induced error from the sampling dynamics, a problem that does not arise for ARMs whose autoregressive sampling exactly reflects the learned probability model. We introduce a sampler-centric oracle framework that replaces learned denoisers with an exact Hidden Markov Model posterior derived from a ground-truth Markov chain, isolating sampler-induced error in a controlled setting. We show that few-step discrete diffusion samplers are not distributionally correct even under an oracle denoiser, with transition-level mismatch that vanishes only as the number of steps approaches the sequence length. Moreover, improvements in negative log-likelihood (NLL), generative perplexity (GenPPL), or MAUVE do not imply correct sampling. Code is available at https://luhantang.github.io/dllm_sampler

Despite high semantic alignment, modern text-to-image (T2I) generative models still struggle to synthesize diverse images from a given prompt. In this work, we enhance the T2I diversity through a geometric lens. Unlike most existing methods that rely primarily on entropy-based guidance to increase sample dissimilarity, we introduce Geometry-Aware Spherical Sampling (GASS) to enhance diversity by explicitly controlling both prompt-dependent and prompt-independent sources of variation. Specifically, we decompose the diversity measure in CLIP embeddings using two orthogonal directions: the text embedding, which captures semantic variation related to the prompt, and an identified orthogonal direction that captures prompt-independent variation (e.g., backgrounds). Based on this decomposition, GASS increases the geometric projection spread of generated image embeddings along both axes and guides the T2I sampling process via expanded predictions along the generation trajectory. Our experiments on different frozen T2I backbones (U-Net and DiT, diffusion and flow) and benchmarks demonstrate the effectiveness of disentangled diversity enhancement with minimal impact on image fidelity and semantic alignment.

Multi-turn jailbreak attacks have proven effective against text-only large language models (LLMs), where malicious content is gradually introduced to bypass safety alignment. However, effectively extending such attacks to large vision-language models (LVLMs) remains underexplored. In this paper, we find that naively incorporating visual inputs can make multi-turn jailbreaks easier to defend against; for example, overly malicious visual content will easily trigger the defense mechanism in safety-aligned LVLMs, resulting in more conservative responses. Based on this finding, we propose multi-turn adaptive prompting attack (MAPA) that 1) at each turn, alternates text-vision attack actions to elicit the most malicious response; and 2) across turns, adjusts the attack trajectory through iterative back-and-forth refinement to gradually amplify response maliciousness. This two-level design enables MAPA to consistently outperform state-of-the-art methods, improving attack success rates by 15-30% on recent benchmarks against LLaVA-v1.6-Mistral-7B, Qwen2.5-VL-7B-Instruct, Llama-3.2-Vision-11B-Instruct and GPT-4o-mini. Our code is available at: https://github.com/thomaschoi143/MAPA.

A line of recent training-free methods for mitigating hallucinations in large vision-language models (LVLMs) operates by amplifying attention to visual tokens during autoregressive generation within a single forward pass. We refer to this paradigm as visual attention amplification (VAA). In this paper, we identify a dual failure pattern in existing VAA methods caused by their use of a fixed amplification factor across generation steps: it can be too weak at some steps, leaving hallucinations unresolved, while too strong at others, introducing new hallucinations. Motivated by this finding, we propose Step-wise Adaptive Visual Attention Amplification (SAVAA), a new VAA framework that estimates hallucination risk for each generated token and uses the estimated risk to adaptively amplify visual attention at the next generation step. Specifically, we introduce Visual Grounding Entropy (VGE), a lightweight hallucination-risk estimator that augments predictive entropy with visual grounding, assigning higher risk to tokens that are uncertain, weakly grounded in the image, or both. Guided by VGE, SAVAA uses the estimated risk to calibrate the VAA factor for the next generation step, applying stronger amplification to higher-risk steps and weaker amplification to lower-risk steps. Across LLaVA-NeXT-7B, Qwen3-VL-8B, and InternVL3.5-8B, SAVAA significantly outperforms baseline methods on generative hallucination benchmarks such as CHAIR, SHR and AMBER. Code is available at: https://github.com/JiachengZ01/SAVVA.

Mechanically characterizing the human-machine interface is essential to understanding user behavior and optimizing wearable robot performance. This interface has been challenging to sensorize due to manufacturing complexity and non-linear sensor responses. Here, we measure human limb-device interaction via fluidic innervation, creating a 3D-printed silicone pad with embedded air channels to measure forces. As forces are applied to the pad, the air channels compress, resulting in a pressure change measurable by off-the-shelf pressure transducers. We demonstrate in benchtop testing that pad pressure is highly linearly related to applied force ($R^2 = 0.998$) and confirmed strong linear relationships to isometric knee torque in a clinical dynamometer with strategic pad placement. We built on these idealized settings to test pad performance in more unconstrained settings, including during cyclic dynamic and stepwise isometric bicep curls. Finally, we integrated the sensor into a lower-extremity robotic exoskeleton and recorded pad pressure during repeated squats with the device unpowered. Pad pressure tracked squat phase and overall task dynamics consistently. Collectively, our preliminary results suggest fluidic innervation is a readily customizable sensing modality with high signal-to-noise ratio and temporal resolution for capturing human-machine interaction. In the long-term, this modality may provide an alternative real-time sensing input to control / optimize wearable robotic systems and to capture user function during device use.

Reward-maximizing RL methods have shown to be capable of enhancing the reasoning performance of LLMs, but often lead to reduced generation diversity. Recent works address this issue by adopting GFlowNets, training LLMs to match a target distribution while jointly learning its partition function. In contrast to prior works that treat this partition function solely as a normalizer, we reinterpret it as a per-prompt expected-reward (i.e., online accuracy) signal, leveraging this unused information to improve sample efficiency. Specifically, we first establish a theoretical relationship between the partition function and per-prompt accuracy estimates. Building on this key insight, we propose Partition Function-Guided RL (PACED-RL), a post-training framework that leverages accuracy estimates to prioritize informative question prompts during training, and further improves sample efficiency through an accuracy estimate error-prioritized replay. Crucially, both components reuse information already produced during GFlowNet training, effectively amortizing the compute overhead into the existing optimization process. Extensive experiments across diverse benchmarks demonstrate strong performance improvements over GRPO and prior GFlowNet approaches, highlighting PACED-RL as a promising direction for a more sample efficient distribution-matching training for LLMs.

In federated learning (FL), $K$ clients jointly train a model without sharing raw data. Because each participant invests data and compute, clients need mechanisms to later prove the provenance of a jointly trained model. Model watermarking embeds a hidden signal in the weights, but naive approaches either do not scale with many clients as per-client watermarks dilute as $K$ grows, or give any individual client the ability to verify and potentially remove the watermark. We introduce $(t,K)$-threshold watermarking: clients collaboratively embed a shared watermark during training, while only coalitions of at least $t$ clients can reconstruct the watermark key and verify a suspect model. We secret-share the watermark key $τ$ so that coalitions of fewer than $t$ clients cannot reconstruct it, and verification can be performed without revealing $τ$ in the clear. We instantiate our protocol in the white-box setting and evaluate it on image classification tasks on both IID and non-IID partitions, as well as language models fine-tuning setting. Our watermark remains detectable at scale ($K=128$) with minimal accuracy loss and stays above the detection threshold ($z\ge 4$) under attacks including adaptive fine-tuning using up to 20% of the training data. Code is available at https://github.com/tameemalaa/collaborative-threshold-watermark.

We study the computational relationship between replicability (Impagliazzo et al. [STOC `22], Ghazi et al. [NeurIPS `21]) and other stability notions. Specifically, we focus on replicable PAC learning and its connections to differential privacy (Dwork et al. [TCC 2006]) and to the statistical query (SQ) model (Kearns [JACM `98]). Statistically, it was known that differentially private learning and replicable learning are equivalent and strictly more powerful than SQ-learning. Yet, computationally, all previously known efficient (i.e., polynomial-time) replicable learning algorithms were confined to SQ-learnable tasks or restricted distributions, in contrast to differentially private learning. Our main contribution is the first computationally efficient replicable algorithm for realizable learning of parities over arbitrary distributions, a task that is known to be hard in the SQ-model, but possible under differential privacy. This result provides the first evidence that efficient replicable learning over general distributions strictly extends efficient SQ-learning, and is closer in power to efficient differentially private learning, despite computational separations between replicability and privacy. Additionally, we leverage our parity learner to prove that, assuming $RP \neq NP$, converting replicability to pure differential privacy requires a strict loss in sample complexity. Our main building block is a new, efficient, and replicable algorithm that, given a set of vectors, outputs a subspace of their linear span that covers most of them.

The rapid adoption of large language models (LLMs) in digital health has been driven by a "scaling-first" philosophy, i.e., the assumption that clinical intelligence increases with model size and data. However, real-world clinical needs include not only effectiveness, but also reliability and reasonable deployment cost. Since clinical decision-making is inherently collaborative, we challenge the monolithic scaling paradigm and ask whether a Small Agent Group (SAG) can support better clinical reasoning. SAG shifts from single-model intelligence to collective expertise by distributing reasoning, evidence-based analysis, and critical audit through a collaborative deliberation process. To assess the clinical utility of SAG, we conduct extensive evaluations using diverse clinical metrics spanning effectiveness, reliability, and deployment cost. Our results show that SAG achieves superior performance compared to a single giant model, both with and without additional optimization or retrieval-augmented generation. These findings suggest that the synergistic reasoning represented by SAG can substitute for model parameter growth in clinical settings. Overall, SAG offers a scalable solution to digital health that better balances effectiveness, reliability, and deployment efficiency.

Pipeline integrity is critical to industrial safety and environmental protection, with Magnetic Flux Leakage (MFL) detection being a primary non-destructive testing technology. Despite the promise of deep learning for automating MFL interpretation, progress toward reliable models has been constrained by the absence of a large-scale public dataset and benchmark, making fair comparison and reproducible evaluation difficult. We introduce \textbf{PipeMFL-240K}, a large-scale, meticulously annotated dataset and benchmark for complex object detection in pipeline MFL pseudo-color images. PipeMFL-240K reflects real-world inspection complexity and poses several unique challenges: (i) an extremely long-tailed distribution over \textbf{12} categories, (ii) a high prevalence of tiny objects that often comprise only a handful of pixels and (iii) substantial intra-class variability. The dataset contains \textbf{249,320} images and \textbf{200,020} high-quality bounding-box annotations, collected from 12 pipelines spanning approximately \textbf{1,530} km. Extensive experiments are conducted with state-of-the-art object detectors to establish baselines. Results show that modern detectors still struggle with the intrinsic properties of MFL data, highlighting considerable headroom for improvement, while PipeMFL-240K provides a reliable and challenging testbed to drive future research. As the first public dataset and the first benchmark of this scale and scope for pipeline MFL inspection, it provides a critical foundation for efficient pipeline diagnostics as well as maintenance planning and is expected to accelerate algorithmic innovation and reproducible research in MFL-based pipeline integrity assessment.

Kabuki syndrome (KS) and Wiedemann-Steiner syndrome (WSS) are rare but distinct developmental disorders that share overlapping clinical features, including neurodevelopmental delay, growth restriction, and persistent fetal fingertip pads. While genetic testing remains the diagnostic gold standard, many individuals with KS or WSS remain undiagnosed due to barriers in access to both genetic testing and expertise. Dermatoglyphic anomalies, despite being established hallmarks of several genetic syndromes, remain an underutilized diagnostic signal in the era of molecular testing. This study presents a vision transformer-based deep learning model that leverages fingerprint images to distinguish individuals with KS and WSS from unaffected controls and from one another. We evaluate model performance across three binary classification tasks. Across the three classification tasks, the model achieved AUC scores of 0.80 (control vs. KS), 0.73 (control vs. WSS), and 0.85 (KS vs. WSS), with corresponding F1 scores of 0.71, 0.72, and 0.83, respectively. Beyond classification, we apply attention-based visualizations to identify fingerprint regions most salient to model predictions, enhancing interpretability. Together, these findings suggest the presence of syndrome-specific fingerprint features, demonstrating the feasibility of a fingerprint-based artificial intelligence (AI) tool as a noninvasive, interpretable, and accessible future diagnostic aid for the early diagnosis of underdiagnosed genetic syndromes.

Tree-boosting is a widely used machine learning technique for tabular data. However, its out-of-sample accuracy is critically dependent on multiple hyperparameters. In this article, we empirically compare several popular methods for hyperparameter optimization for tree-boosting including random grid search, the tree-structured Parzen estimator (TPE), Gaussian-process-based Bayesian optimization (GP-BO), Hyperband, the sequential model-based algorithm configuration (SMAC) method, and deterministic full grid search using $59$ regression and classification data sets. We find that the SMAC method clearly outperforms all the other considered methods. We further observe that (i) a relatively large number of trials larger than $100$ is required for accurate tuning, (ii) using default values for hyperparameters yields very inaccurate models, (iii) all considered hyperparameters can have a material effect on the accuracy of tree-boosting, i.e., there is no small set of hyperparameters that is more important than others, and (iv) choosing the number of boosting iterations using early stopping yields more accurate results compared to including it in the search space for regression tasks.

Iterative Direct Preference Optimization (DPO) has emerged as a widely used paradigm for aligning Large Language Models on reasoning tasks. Existing approaches typically rely on Best-of-N sampling ($N\geq8$) to mine positive trajectories from the distribution tail. In this work, we show that in mathematical reasoning, increasing $N$ yields diminishing returns while increasing verifier-induced false-positive risk and the distribution shift required for policy updates. To address this, we introduce PACE (Proximal Alignment via Corrective Exploration), a generation-based corrective framework that replaces exhaustive mining with low-budget exploration ($2\leq N\leq3$). Rather than searching for increasingly rare positive samples, PACE synthesizes high-fidelity preference pairs from failed explorations through corrective hindsight refinement and verification-guided filtering. Empirically, PACE matches or exceeds the performance of DPO-R1 ($N=16$) while using about $1/5$ of the compute, and remains robust under 20\% label corruption, where high-$N$ baselines exhibit substantially higher noise exploitation.

We present a large-scale human evaluation benchmark for assessing cultural localisation in machine translation produced by state-of-the-art multilingual large language models (LLMs). Existing MT benchmarks emphasise token-level and grammatical accuracy, but often overlook the pragmatic and culturally grounded competencies required for real-world localisation. Building on a pilot study of 87 translations across 20 languages, we evaluate 7 multilingual LLMs across 15 target languages with 5 native-speaker raters per language. Each rater scored both full-text translations and segment-level instances of culturally nuanced language (idioms, puns, holidays, and culturally embedded concepts) on an ordinal 0-3 quality scale; segment ratings additionally included an NA option for untranslated segments. Across full-text evaluations, mean overall quality is modest (1.68/3): GPT-5 (2.10/3), Claude Sonnet 4 (1.97/3), and Mistral Medium 3.1 (1.84/3) form the strongest tier with fewer catastrophic failures. Segment-level results show sharp category effects: holidays (2.20/3) and cultural concepts (2.19/3) translate notably better than idioms (1.65/3) and puns (1.45/3), and idioms are most likely to be left untranslated. Inter-rater reliability was assessed using Krippendorff's α and Gwet's AC2, indicating moderate agreement overall (Krippendorff's α = 0.45) with the lowest agreement for puns. These findings demonstrate a persistent gap between grammatical adequacy and cultural resonance. To our knowledge, this is the first multilingual, human-annotated benchmark focused explicitly on cultural nuance in translation and localisation. The results highlight the need for culturally informed training data, improved cross-lingual pragmatics, and evaluation frameworks that support systematic benchmarking of culturally grounded translation.

Neural implicit mapping has emerged as a powerful paradigm for robotic navigation and scene understanding. However, real-world robotic deployment requires continual adaptation to changing environments under strict memory and computation constraints, which existing mapping systems fail to support. Most prior methods rely on replaying historical observations to preserve consistency and assume static scenes. As a result, they cannot adapt to continual learning in dynamic robotic settings. To address these challenges, we propose TACO (TemporAl Consensus Optimization), a replay-free framework for continual neural mapping. We reformulate mapping as a temporal consensus optimization problem, where we treat past model snapshots as temporal neighbors. Intuitively, our approach resembles a model consulting its own past knowledge. We update the current map by enforcing weighted consensus with historical representations. Our method allows reliable past geometry to constrain optimization while permitting unreliable or outdated regions to be revised in response to new observations. TACO achieves a balance between memory efficiency and adaptability without storing or replaying previous data. Through extensive simulated and real-world experiments, we show that TACO robustly adapts to scene changes, and consistently outperforms other continual learning baselines. Code is available at https://iconlab.negarmehr.com/TACO

Inference-time scaling via chain-of-thought (CoT) reasoning is a major driver of state-of-the-art LLM performance, but it comes with substantial latency and compute costs. We address a fundamental theoretical question: how many reasoning tokens are required to solve a problem as input size grows? By extending the bounded attention prefix oracle (BAPO) model--an abstraction of LLMs that quantifies the information flow required to solve a task--we prove lower bounds on the CoT tokens required for three canonical BAPO-hard tasks: binary majority, triplet matching, and graph reachability. We show that each requires $Ω(n)$ reasoning tokens when the input size is $n$. We complement these results with matching or near-matching upper bounds via explicit constructions. Finally, our experiments with frontier reasoning models show approximately linear reasoning token scaling on these tasks and failures when constrained to smaller reasoning budgets, consistent with our theoretical lower bounds. Together, our results identify fundamental bottlenecks in inference-time compute through CoT and offer a principled tool for analyzing optimal reasoning length.

Post-training of reasoning LLMs is a holistic process that typically consists of an offline SFT stage followed by an online reinforcement learning (RL) stage. However, SFT is often optimized in isolation to maximize SFT performance alone. We show that, after identical RL training, models initialized from stronger SFT checkpoints can significantly underperform those initialized from weaker ones. We attribute this to a mismatch typical in current SFT-RL pipelines: the distribution that generates the offline SFT data can differ substantially from the policy optimized during online RL, which learns from its own rollouts. We propose PEAR (Policy Evaluation-inspired Algorithm for Offline Learning Loss Re-weighting), an SFT-stage method that corrects this mismatch and better prepares the model for RL. PEAR uses importance sampling to reweight the SFT loss, with three variants operating at the token, block, and sequence levels. It can be used to augment standard SFT objectives and incurs little additional training overhead once probabilities for the offline data are collected. We conduct controlled experiments on verifiable reasoning games and mathematical reasoning tasks on Qwen 2.5 and 3 and DeepSeek-distilled models. PEAR consistently improves post-RL performance over canonical SFT, with pass at 8 gains up to a 14.6 percent on AIME2025. Our results suggest that PEAR is an effective step toward more holistic LLM post-training by designing and evaluating SFT with downstream RL in mind rather than in isolation.

Agentic Reinforcement Learning (ARL) trains large language models to interleave reasoning with external tool execution to solve complex tasks. Most existing ARL methods train a single set of parameters to support both reasoning and tool-use behaviors, implicitly assuming that joint training leads to improved overall agent performance. Despite its widespread adoption, this assumption has rarely been examined empirically. In this paper, we systematically examine this assumption by introducing Capability Effect Attribution (CEA), which provides quantitative evidence of interference between reasoning and tool-use behaviors. Through an in-depth analysis, we show that these two capabilities often induce misaligned gradient directions, leading to training interference that undermines the effectiveness of joint optimization and challenges the prevailing ARL paradigm. To address this issue, we propose Disentangled Action--Reasoning Tuning (DART), a simple and efficient framework that explicitly decouples parameter updates for reasoning and tool use via separate low-rank adaptation modules. With this simple change alone, DART outperforms all joint-optimization baselines and approaches the 2-Agent upper bound across thirteen benchmarks on retrieval-augmented QA and NL2SQL, further supporting our finding of capability interference under shared optimization.

Recent advances in Large Audio Language Models (LALMs) have extended Text-to-Speech (TTS) to interactive role-play scenarios, which demand high expressiveness and strict adherence to role-play instructions. However, existing models struggle to maintain stylistic consistency with character profiles and scene descriptions across multi-turn dialogues. A critical bottleneck is the lack of objective metrics for quantifying speaking style. To bridge this gap, we propose Mean Continuation Log-Probability (MCLP) as both an evaluation metric and a reward signal, validated on LALM-based Role-Play TTS (RP-TTS) tasks. MCLP leverages the in-context learning capability of pretrained LALMs to measure the likelihood of ground-truth speech tokens conditioned on a contextual history consisting of the transcript, generated speech, and repeated transcript, serving as a proxy for stylistic continuity. Furthermore, we employ MCLP as a reinforcement learning reward to enhance the style alignment between generated speech and role-play instructions. To support this task, we construct a large-scale RP-TTS dataset with rich scene and character annotations. Experiments demonstrate that MCLP is well aligned with human judgments of stylistic consistency and serves as an effective reward for improving RP-TTS, leading to consistent gains in both objective metrics and subjective evaluations. Our code is publicly available at https://github.com/y-ren16/MCLP.

Because of the pervasive use of deep neural networks (DNNs), especially in high-stakes domains, the interpretability of DNNs has received increased attention. The general idea of rationale extraction (RE) is to provide an interpretable-by-design framework for DNNs via a select-predict architecture where two neural networks learn jointly to perform feature selection and prediction, respectively. Given only the remote supervision from the final task prediction, the process of learning to select subsets of features (or rationales) requires searching in the space of all possible feature combinations, which is computationally challenging and even harder when the base neural networks are not sufficiently capable. To improve the predictive performance of RE models that are based on less capable or smaller neural networks (i.e., the students), we propose REKD (Rationale Extraction with Knowledge Distillation) where a student RE model learns from the rationales and predictions of a teacher (i.e., a rationalist) in addition to the student's own RE optimization. This structural adjustment to RE aligns well with how humans could learn effectively from interpretable and verifiable knowledge. Because of the neural-model agnostic nature of the method, any black-box neural network could be integrated as a backbone model. To demonstrate the viability of REKD, we conduct experiments with multiple variants of BERT and vision transformer (ViT) models. Our experiments across language and vision classification datasets (i.e., IMDB movie reviews, CIFAR 10 and CIFAR 100) show that REKD significantly improves the predictive performance of the student RE models.

Recent post-training quantization (PTQ) methods have adopted block rotations to diffuse outliers prior to rounding. While this reduces the overhead of online full-vector rotations, the effect of block structure on outlier suppression remains poorly understood. To fill this gap, we present the first systematic, non-asymptotic analysis of outlier suppression for block Hadamard rotations. Our analysis reveals that outlier suppression is fundamentally limited by the geometry of the input vector. In particular, in the deterministic worst case, post-rotation outliers are minimized when the pre-rotation $\ell_1$ norm mass is evenly distributed across blocks. Guided by these insights, we introduce PeRQ (Permute, Rotate, then Quantize), a PTQ framework that redistributes activation mass via permutations prior to rotation. We propose a greedy mass diffusion algorithm to calibrate permutations by equalizing the expected blockwise $\ell_1$ norms. To avoid adding inference overhead, we identify permutation-equivariant regions in transformer architectures to merge these permutations into model weights before deployment. Experiments show that PeRQ consistently improves accuracy across all block sizes, recovering up to 90% of the full-vector rotation perplexity when quantizing Llama3 1B to INT4 with block size 16, compared to 46% without permutations.

Pretraining language models directly on web-scale corpora is the de facto paradigm. We study an alternative where the model is initially exposed to abstract structured data to ease the subsequent acquisition of rich semantic knowledge, much like humans learning simple logic and mathematics before higher reasoning. We focus on procedural data, generated by formal languages and other simple algorithms, as such abstract data. We first diagnose the algorithmic skills that different forms of procedural data can improve, often significantly. For example, the accuracy of context recall (Needle-in-a-haystack) jumps from 10 to 98% when a model is pretrained on Dyck sequences (balanced brackets). Second, we study how these gains are reflected in pretraining larger models (up to 1.3B). We find that front-loading as little as 0.1 to 0.3% procedural data significantly outperforms standard pretraining on natural language, code, and informal mathematics (C4, CodeParrot, and DeepMind-Math datasets). Notably, this also enables the models to reach the same loss value with only 55/67/86% of the original data and thus a comparable reduction in FLOPs. Third, we explore the mechanisms behind the benefits and find that procedural pretraining instills non-trivial structure in both attention and MLP layers. The former is particularly important for structured domains (e.g. code), and the latter for language. Finally, we lay a path for combining multiple forms of procedural data. Our results show that procedural pretraining is a simple, lightweight means of improving performance and accelerating language model pretraining, ultimately suggesting the promise of disentangling knowledge acquisition from reasoning in LLMs.

SWE-bench has emerged as the premier benchmark for evaluating Large Language Models on complex software engineering tasks. While these capabilities are fundamentally acquired during the mid-training phase and subsequently elicited during Supervised Fine-Tuning (SFT), there remains a critical deficit in metrics capable of guiding mid-training effectively. Standard metrics such as Perplexity (PPL) are compromised by the "Long-Context Tax" and exhibit weak correlation with downstream SWE performance. In this paper, we bridge this gap by first introducing a rigorous data filtering strategy. Crucially, we propose the Entropy Compression Hypothesis, redefining intelligence not by scalar Top-1 compression, but by the capacity to structure uncertainty into Entropy-Compressed States of low orders ("reasonable hesitation"). Grounded in this fine-grained entropy analysis, we formulate a novel metric, HE-SNR (High-Entropy Signal-to-Noise Ratio). We validate our approach on models with up to 560B parameters across different context windows (32K/128K). This work provides both the theoretical foundation and practical tools for optimizing the latent potential of LLMs in complex engineering domains.

Modern generative modeling methods have demonstrated strong performance in learning complex data distributions from clean samples. In many scientific and imaging applications, however, clean samples are unavailable, and only noisy or linearly corrupted measurements can be observed. Moreover, latent structures, such as manifold geometries, present in the data are important to extract for further downstream scientific analysis. In this work, we introduce Riemannian AmbientFlow, a framework for simultaneously learning a probabilistic generative model and the underlying, nonlinear data manifold directly from corrupted observations. Building on the variational inference framework of AmbientFlow, our approach incorporates data-driven Riemannian geometry induced by normalizing flows, enabling the extraction of manifold structure through pullback metrics and Riemannian Autoencoders. We establish theoretical guarantees showing that, under appropriate geometric regularization and measurement conditions, the learned model recovers the underlying data distribution up to a controllable error and yields a smooth, bi-Lipschitz manifold parametrization. We further show that the resulting smooth decoder can serve as a principled generative prior for inverse problems with recovery guarantees. We empirically validate our approach on low-dimensional synthetic manifolds and on MNIST.

Deep Brain Stimulation (DBS) is an effective treatment for Parkinson's disease, but conventional fixed-parameter stimulation can reduce battery life and cause side effects while failing to adapt to changing neural dynamics. Recent reinforcement learning approaches improve adaptability, yet most rely on deep neural networks that require offline training and are computationally too expensive for implantable hardware. This paper presents a resource-conscious adaptive DBS framework based on a Time- and Threshold-Triggered Pruned Multi-Armed Bandit (T3P MAB) algorithm. The proposed method jointly tunes stimulation frequency and amplitude, avoids prior training, and remains transparent enough to support clinician-guided adjustment. Using a computational basal ganglia-thalamic model, we show that T3P converges faster than competing MAB methods and outperforms deep-RL baselines in suppressing pathological beta-band activity while reducing stimulation power. We implemented it on different microcontrollers and report detailed energy measurements, showing convergence in under two minutes and suitability for resource-constrained implantable systems. These results support lightweight bandit-based control as a practical path toward personalized, energy-efficient DBS.

Controllable generation requires language models to realize output characteristics such as reading level, politeness, and toxicity. Existing steering methods are often indirect, require access to internal activations, or depend on auxiliary trained models. We propose SWAI, a training-free inference-time method that addresses these limitations by steering directly in logit space using corpus-derived token statistics. SWAI computes z-normalized one-vs-rest log-odds scores from labeled corpora and biases high-scoring tokens only within the model's top-K candidate set, allowing control to favor target-characteristic tokens while preserving contextually plausible choices. Across readability, politeness, and toxicity control, SWAI consistently improves over prompt-based and prior logit-level baselines without modifying model parameters, accessing internal layers, or training an auxiliary model. Selectivity and lookup-table ablations show that the gains come from target-specific statistical scores rather than generic logit perturbation. These results indicate that effective steering does not require learned controllers when the logit intervention is guided by target-specific statistics under high-probability candidates.

Rubric-based text evaluation increasingly uses large language models (LLMs) as scalable judges, but aligning frozen black-box models with human scoring standards remains challenging. We formulate this challenge as a criteria-transfer problem: the goal is not merely to prompt an LLM to assign a score, but to transfer human rubric intent into a stable, auditable, and human-aligned scoring protocol. We identify three recurring failure modes in LLM-based rubric scoring: rubric execution drift, unverifiable score attribution, and human-scale misalignment. To address these failure modes, we introduce Rulers, a three-stage inference-time framework for reliable, evidence-grounded rubric-based text evaluation. Rulers first converts a human rubric into a locked task-level specification, then executes the specification with structured checklist decisions, typed evidence grounding, and extractive quote verification when applicable, and finally applies post-hoc calibration to align model-derived signals with human score boundaries. Across four rubric-governed benchmarks covering essay scoring, summarization assessment, EFL writing evaluation, and structured-input text generation, Rulers achieves stronger human-score agreement in most evaluated settings across multiple frozen backbone models. Further analyses show that Rulers better matches empirical human score distributions, improves stability under semantically equivalent rubric perturbations, and benefits from each of its three components. These results suggest that reliable LLM judging requires fixed criteria, traceable evidence, and calibrated score interpretation rather than prompt phrasing alone. Our code is available at https://anonymous.4open.science/r/Rulers_0525-3328.

Confidence estimation (CE) indicates how reliable the answers of large language models are and impacts user trust and decision-making. Existing evaluations mainly concern the alignment between confidence and correctness, but ignore the variability of language: confidence estimates should remain consistent under semantically equivalent prompts or answer variations, while changing when answer meaning differs, as this may indicate a change in correctness. Therefore, we introduce a novel evaluation framework based on three complementary properties: \textbf{robustness} to prompt perturbations, \textbf{stability} across semantically equivalent answers, and \textbf{sensitivity} to semantically different answers. We show that these metrics are largely independent from existing CE metrics, and that common CE methods often fail on them: while most methods achieve high robustness and stability, they struggle to distinguish semantically different answers, potentially because they do not effectively leverage generation-side information. Overall, our framework exposes overlooked limitations of current CE evaluations and provides guidance for selecting confidence estimators for real-world applications.