U.S. immigration law spans thousands of pages of official policy, federal regulations, and procedural guidance that change frequently and carry high stakes for petitioners who lack legal representation. We describe the construction of ImmigrationQA, a source-grounded question-answering dataset of 17,058 pairs across 13 immigration subdomains, and the fine-tuning of a Llama 3.2 3B Instruct model on that dataset using parameter-efficient LoRA. The corpus was assembled from 11 primary and secondary sources -- including the USCIS Policy Manual, 8 CFR, BIA precedent decisions, and community Q&A -- yielding 10,056 validated canonical documents and 18,308 text chunks. Structured QA pairs were generated from these chunks using Claude Sonnet 4.6 via five mode-specific prompts, with 22 pairs rejected for insufficient source-span overlap. The fine-tuned model was evaluated against a held-out split of 993 pairs using LLM-as-judge scoring on a 101-example stratified sample. The fine-tuned model scored a mean of 1.08/3.0 (16.8% fully correct; 101-example stratified eval) versus the Llama 3 8B base model at 0.85/3.0 (4% fully correct), a relative improvement of 27% in mean score; a zero-shot Claude Sonnet baseline scored 1.52/3.0 (25% fully correct). The fine-tuned model shows concentrated improvement in procedural subdomains (travel documents, adjustment of status, nonimmigrant visas) while remaining weak on complex legal reasoning and time-sensitive statistics. The full pipeline ran for approximately $29 in cloud compute. All artifacts -- dataset, model, code, and prompt templates -- are publicly released. The system is not a substitute for legal counsel and does not reflect regulatory changes after the corpus crawl date.

Chain-of-thought (CoT) reasoning has significantly improved the reasoning ability of large vision-language models (LVLMs) by verbalizing intermediate reasoning steps in natural language. However, such discrete textual rationales are often insufficient for encoding continuous visual evidence. Recent work addresses this limitation by moving reasoning into continuous latent space. Despite promising progress, existing methods leave latent reasoning insufficiently connected to the compositional and relational structure of visual evidence. To address this gap, we introduce ReGuLaR, a relation grounded latent reasoning framework that explicitly grounds latent states in these critical yet overlooked visual evidence. ReGuLaR uses a training-time ReGFormer to focus latent reasoning on question-relevant objects and inter-object relations, while at inference time the model reasons and generates answers without invoking the ReGFormer. To support training ReGuLaR, we construct RGROUNDING-351K, a real-world vision-language dataset annotated with key object bounding boxes and inter-object relations. Extensive experiments across diverse benchmarks show that ReGuLaR consistently outperforms existing approaches and achieves state-of-the-art performance. We include our code in the submission and will release the code and training data publicly upon acceptance.

Effective prognostics and health management of modern engines relies on accurate turbine gas temperature predictions and robust uncertainty quantification to ensure reliability and safety. This paper investigates five major approaches for constructing prediction intervals -- namely the Delta method, Bayesian Monte Carlo Dropout, Bootstrap method, Lower-Upper Bound Estimation, and Mean-Variance Estimation -- as a means of capturing the uncertainty in neural network predictions of turbine gas temperature. Each approach is implemented within a unified experimental framework that employs cross-validation for hyperparameter selection, repeated train-test splits for performance robustness, and multiple metrics to evaluate both the accuracy and tightness of the intervals. In particular, Coverage Probability, Normalized Mean Prediction Interval Width, and the Coverage Width-based Criterion are measured to comprehensively assess each method's reliability and sharpness. Experiments conducted on a representative turbine gas temperature dataset reveal distinct trade-offs among the five methods in terms of interval coverage, width, and stability. These findings provide a practical guide for selecting and tuning prediction interval methods in engine health management and prognostics, ensuring both interpretability and precision in real-world applications.

We present Caspar, a library that makes the power of modern GPUs more accessible in robotics and provides a state-of-the-art nonlinear GPU solver that can be applied to a wide range of different optimization problems. Caspar bridges the gap between expressive symbolic programming in Python and high-performance GPU runtimes in C++ by automatically generating optimized CUDA kernels from symbolic expressions. Building on the SymForce library, users can easily define and combine symbolic expressions, including Lie group operations, to generate custom CUDA kernels. To use Caspar as a solver, users need only define the symbolic residual functions; Caspar then uses symbolic differentiation to generate the necessary GPU kernels and interfaces to perform nonlinear optimization. In this paper, we present the core components of Caspar and showcase its performance by performing bundle adjustment on the Bundle Adjustment in the Large (BAL) dataset. We benchmark Caspar against other state-of-the-art bundle adjusters and show that it is 5 to 20 times faster than the best alternative, requires less memory, and achieves similar accuracy. This illustrates the benefit of our symbolic GPU programming approach. Caspar is released as part of SymForce and is freely available at https://github.com/symforce-org/symforce

While platforms like Google Scholar and Semantic Scholar track citations for academic papers, no comparable infrastructure exists for monitoring dataset usage in research literature, leaving the landscape of data use largely opaque. Addressing this gap is critical for transparency, reproducibility, and monitoring of impact, yet progress is hindered by inconsistent citation practices, scarce labeled data, and ambiguous references to datasets in the wild. Traditional NLP approaches struggle with these challenges, motivating the shift toward more adaptive, semantically rich models. Building on prior work using LLMs for data mention detection and synthetic data for bootstrapping training, this paper presents an updated methodology for scalable dataset monitoring. We introduce a multitask GLiNER-based framework that jointly performs dataset mention extraction, relation identification, and usage-context classification. To address label scarcity, the pipeline leverages synthetic data generation to produce training examples and LLM-based revalidation to filter incorrect mentions and enforce labeling consistency, together improving reliability, coverage, and output consistency across the training pipeline. This work advances the development of open-source tools for monitoring data use in research literature, contributing to the broader goal of generalizable, unconstrained dataset citation tracking.

Speculative decoding has become a crucial component of large language model (LLM) inference, enabling faster generation by drafting multiple tokens and verifying them in parallel. However, small draft models tend to suffer from disproportionately poor multilingual capabilities. Thus, when generating text in a non-English language, speculative decoding is far less effective. We compare three strategies to improve speculative decoding efficiency for eleven languages: finetuning the draft model on task-specific data (translation); finetuning the draft model on unlabeled monolingual corpora; and training simple n-gram draft models on the same monolingual corpora. We evaluate efficiency on translation (from English into the target language) and the held-out task of story generation. We find that while task-specific distillation can significantly improve efficiency, distilled models generalize poorly to a new task. Meanwhile, n-gram draft models, despite lower acceptance rates, consistently provide large speed-ups due to much faster draft generation.

Exploration in reinforcement learning for autonomous driving is inherently unsafe: agents must experience novel behaviors to learn, yet exploration can lead to collisions or off-road driving. We propose an uncertainty-aware framework that leverages expert advice to guide exploration while avoiding long-term dependence. Advice is triggered when epistemic or aleatoric uncertainty exceeds adaptive thresholds derived from rolling buffers, ensuring advice evolves with the agent's confidence. A commitment-cooldown strategy with a stochastic early-stop heuristic regulates the duration and frequency of guidance, exposing the agent to coherent maneuvers without exhausting the advice budget. Expert and agent experiences are combined in a shared replay buffer within an off-policy implicit quantile network (IQN) backbone, enabling efficient reuse of expert trajectories. Experiments in CARLA show that our method outperforms the IQN baseline, improving success by 5-7% and reducing failures, demonstrating that risk-sensitive uncertainty coupled with regulated expert integration enables safer and more efficient exploration for sensor-based RL policy learning in unsignalized intersection navigation.

Prior KV cache compression schemes empirically demonstrate that the prompt cache is partially redundant during decoding, dropping or summarising entries with little accuracy loss. We ask when and what kind of redundancy: at which layers, after how many decoding steps, and in what form can the prompt span KV cache be replaced without breaking the task. A controlled splice intervention swept over layer cutoff and decoding steps shows this redundancy is about form (chat template scaffolding) rather than content. Replacing the upper layer prompt span KV cache with KV cache from a chat template scaffold whose user content is a neutral filler recovers near clean accuracy, while zeroing the same slots collapses accuracy. The dissociation replicates across the Qwen3, Gemma 3, and Llama 3 families on multiple datasets.

Sampling from high-dimensional, non-log-concave distributions with unnormalized densities remains a fundamental challenge in machine learning, particularly in black-box settings where gradient information is inaccessible or computationally prohibitive. While Langevin dynamics provides a principled framework for sampling when gradients are accessible, its extension to the black-box settings suffers from high variance and lacks non-asymptotic convergence guarantees for non-log-concave sampling. To address these limitations, we propose a variance-reduced zeroth-order Langevin sampling method. Our method employs a gradient estimator that substantially reduces the variance of the classical batched zeroth-order estimator and eliminates the unfavorable dimensional dependence of the batch size required for accurate estimation, enabling practical and stable sampling. We establish the first non-asymptotic convergence guarantees for zeroth-order non-log-concave sampling in terms of $\varepsilon$-relative Fisher information, and, under a Poincaré inequality assumption, squared total variation distance. We further propose ZO-APMC, a posterior sampling algorithm for black-box inverse problems with pre-trained score-based generative priors, establishing the first non-asymptotic convergence guarantees for such methods. We validate our theory through synthetic experiments and demonstrate strong empirical performance on practical linear and nonlinear inverse problems.

We investigate the application of MAP-Elites (a well-known quality diversity algorithm) to design levels for First-Person Shooter (FPS) games. We consider two well-known map representations (All-Black and Grid-Graph) and introduce two novel representations (Point-Line and Spatial-Layout) that improve the characterization of FPS maps. We define a series of metrics to describe maps' topological properties (which solely depend on maps' layout), and emergent properties (which must be evaluated through actual gameplay). We perform an in-depth analysis to identify the most suitable features to guide MAP-Elites illumination process. We apply MAP-Elites with Sliding Boundaries (MESB) to evolve populations of FPS maps. Our results show that the new representations can generate maps with higher diversity and quality than the representations previously used for evolving FPS maps.

Robotic manipulation dexterity is often pursued by building increasingly complex high-DoF multifingered hands. While many robotic hands are designed to replicate human morphology, the functional role of human hands suggests a different perspective: much of their complexity may exist to enable tool use and tool making. This observation motivates Any-ttach, a tool-centric manipulation framework that treats quick end-effector swapping as a mechanism for dexterity with simplicity. Any-ttach combines a low-cost automatic swapping mechanism for an open-close robot interface, a handheld device for collecting human demonstrations, and a task planning framework that composes learned, parameterized, and planned tool-use skills. The system supports diverse tools and end-effector modules, including daily tools, articulated tools such as scissors, Fin Ray fingers, and a low-cost anthropomorphic hand, through the same shared interface. Our experiments show that Any-ttach improves tool-swapping reliability, increases demonstration efficiency, reduces tool-pose variability, and supports diverse tool-use skills. In two long-horizon tasks, making a sandwich and preparing a cucumber, Any-ttach executes six tool-use subskills through end-effector switching and execution monitoring. These results suggest that robots can expand manipulation capability not only through more complex end-effectors, but also through rapidly exchangeable tools and end-effector modules. More details and videos are available at https://any-ttach.github.io/.

LLM-as-a-Judge is a scalable alternative to human evaluation, yet existing rubric-based methods rely on human-annotated data such as reference answers or expert-crafted rubrics. We propose to automatically generate fine-grained evaluation rubrics without any human annotation. Our training-free method generates rubrics at dataset-specific and instance-specific granularities, achieving performance competitive with existing methods across four benchmarks. We further present a method that iteratively fine-tunes a rubric generator model via meta-judge reward signals. The fine-tuned generator outperforms all existing baselines in both pairwise and pointwise evaluation. Notably, a fine-tuned 14B rubric generator outperforms a much larger proprietary model at rubric generation, showing the effectiveness of our fine-tuning strategy.

Factored tasks are a classical planning representation that extends SAS+ with limited forms of disjunctive preconditions, conditional effects, and angelic nondeterminism. This allows for a more compact representation of tasks than traditional formalisms such as STRIPS or SAS+, and supports a wide range of task transformations. However, existing planning approaches for factored tasks have been limited to heuristic search methods. In this work, we investigate how to encode factored tasks in SAT. We propose several ways to encode the tasks, focusing on different strategies for translating the factored transition relation into propositional logic. We also analyze how to exploit parallelism at various levels in this setting and study the impact of common task transformations on the performance of SAT-based planners.

Vision Language Models (VLMs) enable a unified model to solve various vision tasks through prompting. They have shown promising performance in semantic understanding. However, 3D understanding still largely relies on expert vision models with complex task-specific designs. The key argument this work wants to make is that VLMs are native 3D learners. Our in-depth large scale study shows that 1) focal length unification, 2) text-based pixel reference and 3) data mixture and scaling, are all you need for effective 3D learning. Model architecture changes, large models, heavy data augmentations, and complex losses including the regression formulation, many of which form the foundation of expert vision models, are actually not necessary conditions. As a result, we propose VLM3, a scalable method with the simplest design that enables standard VLMs to master diverse 3D tasks. VLM3 not only advances the VLM depth estimation accuracy by a large margin (0.84 -> 0.9), but also enables diverse 3D tasks such as pixel correspondence, camera pose estimation and object-level 3D understanding, matching expert vision model accuracy while maintaining standard architectures and text-based training. We believe VLM3 opens up a new paradigm for simple and scalable 3D learning.

Spatial reasoning is a fundamental capability for vision-language models (VLMs) deployed in real-world environments. However, visual observations are inherently limited representations of a 3D world: occlusion can render objects invisible, and perspective can make geometric properties misleading. Despite this, existing spatial reasoning benchmarks typically assume that observations are sufficient and reliable, focusing on whether models produce correct answers rather than whether they recognize when a question cannot be answered and what additional observations would be needed. In this work, we challenge this assumption by constructing a controlled evaluation framework, SpatialUncertain, and introducing two types of observation challenges: (1) occlusion, which hides target information, and (2) perspective ambiguity, which produces misleading visual cues. For each configuration, we design spatial questions that are answerable under clean observations but require abstention under the introduced challenges. We further evaluate whether models can identify which additional viewpoints would resolve perspective ambiguity. Our results across a diverse set of frontier open- and closed-source VLMs reveal two consistent failure modes. First, models are prone to overconfident answering, attempting to solve spatial reasoning tasks even when visual evidence is incomplete or misleading, with average accuracy around 30\% under occlusion and below 10\% under perspective ambiguity. Second, even when additional views are available, some models perform near random chance in identifying which would provide reliable evidence. Together, our findings call for moving beyond answer correctness toward evaluating whether models know when to abstain and how to seek reliable evidence.

Random, untrained neural networks consistently match or exceed trained networks in representational similarity to early visual cortex. This puzzling finding challenges the assumption that learning improves brain alignment. We investigate it by tracking representational similarity analysis (RSA) alignment to human fMRI data across training for four learning rules: backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP). Using 720 object images from the THINGS database and fMRI data from three subjects across six visual ROIs, we measure Spearman correlations between model and brain representational dissimilarity matrices at eight training checkpoints (epochs 0-40). We find that (1) a single epoch of training reduces V1 alignment by 25-90%, depending on the learning rule; (2) backpropagation reduces V1 alignment most severely (delta r = -0.080), while predictive coding and STDP preserve substantially more (delta r ~ -0.04); and (3) a weaker, opposite tendency appears in object-selective cortex (LOC), where BP shows the largest increase in alignment during training, although the absolute change is small. These results suggest that untrained architectures capture low-level visual statistics through inductive biases alone, and that global error signals (BP) reshape early representations more aggressively than local learning rules (PC, STDP), which better preserve brain-like structure.

I present diffusion models as part of a family of machine learning techniques that withhold information from a model's input and train it to guess the withheld information. I argue that diffusion's destroying approach to withholding is more flexible than typical hand-crafted information withholding techniques, providing a rich training playground that could be advantageous in some settings, notably data-scarce ones. I then address subtle issues that may arise when porting reinforcement learning techniques to the diffusion context, and wonder how such exploration problems could be addressed in more diffusion-native ways. I do not have definitive answers, but I do point my fingers in directions I deem interesting. A tutorial follows this thesis, expanding on the destroy-then-generate perspective. A novel kind of probabilistic graphical models is introduced to facilitate the tutorial's exposition.

Adaptive systems often need to make task-specific decisions about people from limited evidence: a tutor may need to anticipate how a learner will approach a new problem, a game may need to adapt when a player enters a new level, and a human-AI system may need to infer whether a partner will persist with a plan or switch goals. These decisions depend on person-level tendencies that shape how people solve related tasks, but such tendencies are difficult to infer from standard behavioral evidence. One approach is to use aggregate outcome summaries, such as scores, completion rates, or productivity; these summaries are compact and available across tasks, but can collapse distinct behavioral processes into similar outcomes. Another approach is to use process-level traces, which record how behavior unfolds; however, process modeling within one task can entangle stable person-level tendencies with task-specific layout and affordances. In this work, we study early cross-task behavioral inference: whether partial source-task process traces can reveal transferable person-level structure that predicts strategy in a held-out target task. We introduce a Process-Level Latent Variable Model (PLVM), which encodes task-specific traces and fuses them into a shared person-level latent representation for cross-task prediction. In PowerWash Simulator, a naturalistic telemetry dataset of human gameplay, PLVM uses partial traces from two cleaning tasks to predict locally persistent Zone Planner behavior versus frequent Zone Hopper behavior in the held-out Fire Station level. Controlled simulations with known latent types show that cross-task fusion helps when source tasks reveal complementary dimensions of a shared latent process. These results suggest that process-level cross-task modeling can support early prediction of target-task strategy when observing sufficient target-task behavior is impractical.

Korean grammatical error correction (K-GEC) presents a structural mismatch between word-based evaluation and the morpheme-level locus of many learner errors. Postpositions and verbal endings are bound to lexical hosts, but they encode grammatical relations that must be represented in correction and evaluation. This paper refines word-based grammatical error annotation for L2 Korean by addressing three connected problems in existing resources: surface target realization, Korean-specific edit annotation, and single-reference evaluation. We reconstruct target sentences from the National Institute of Korean Language (NIKL) L2 corpus under morphologically constrained realization rules and convert its morpheme-level annotations into word-level \texttt{m2} edits. We then define a Korean ERRANT-style annotation scheme that preserves the MRU core while distinguishing functional morpheme errors, spelling errors, word boundary errors, and word order errors. We also augment the KoLLA corpus with an additional reference correction, yielding a multi-reference evaluation setting for Korean GEC. Empirical validation shows that the refined NIKL targets yield lower perplexity, the converted \texttt{m2} files achieve higher agreement with source-target edit representations, and the refined resources improve KoBART-based correction under the same model setting. Multi-reference KoLLA evaluation further reduces the penalty imposed on valid corrections that diverge from a single reference, especially for neural and prompted GEC systems. These results show that Korean GEC evaluation depends not only on correction models, but also on reference data and edit annotations that reflect Korean morphology, spacing, and correction variability.

Visual monitoring systems that rely on cloud-based AI inference expose raw image data to external services, creating fundamental tensions with the data-minimisation principle of the General Data Protection Regulation (GDPR). This paper presents a proof-of-concept privacy-by-design pipeline that resolves this tension by confining all inference entirely to the edge device. A YOLOv5n-seg model compiled for a Hailo-8L AI accelerator delivers real-time object detection on a Raspberry Pi 5, from which raw pixel buffers are immediately discarded after inference. A stateful trigger engine forwards minimal JSON event payloads to a locally hosted instance of Phi-3 Mini (3.8B parameters, Q4_0 quantisation), which synthesises one-to-two sentence natural-language alerts for a human operator. No image data crosses the network boundary at any point; only the generated text alert is transmitted. We describe the full system architecture and implementation, report measured inference latency and resource utilisation on the target hardware, and present representative generated alerts. The results demonstrate that combining a dedicated neural-network accelerator with an on-device large language model on a single-board computer is not only feasible but produces practically deployable, human-readable monitoring output while aligning with GDPR Art. 5(1)(c) by design.

World models for embodied AI must be physically viable: constructed to answer intervention queries by representing the physical structure governing action outcomes, rather than merely predicting future observations. Existing observation-predictive world models can produce visually plausible but physically wrong rollouts. This failure is structural; distinct physical systems can look identical yet diverge under intervention. We expose this problem with controlled benchmarks that fix the visible scene while varying latent physics. We show that such models may recommend infeasible actions, mispredict interaction outcomes, or certify unsafe behavior. We argue that embodied AI requires world models that identify the simplest physical abstraction sufficient to answer an intervention query. Such a model comprises modular components, including environment representation, latent state and parameter estimation, action specification, interventional dynamics, and query-level response. An autonomous orchestrator should identify the relevant abstraction and compose compatible learned and structured components per query. When closed-form physics is unavailable, uncertain, or costly, the transition model may be analytic, simulated, learned, or hybrid, but it must preserve the structure that determines interventional outcomes. This decomposition makes the model interpretable, its components verifiable, and its outputs auditable against the query. It also provides a design principle for new world models and a feasibility test for existing ones: the right abstraction is not the most detailed model of the world, but the simplest model that preserves the distinctions relevant to the query. We demonstrate this approach on queries that existing systems fail to answer correctly, and outline how an orchestrator can dynamically assemble and adapt physically viable models for planning, control, and verification.

Full waveform inversion (FWI) is the gold standard for subsurface imaging, with applications from carbon sequestration to energy and mineral exploration to earthquake hazard assessment. Machine learning approaches to FWI need field-scale, geologically diverse, and physically realistic training data, but existing resources such as Marmousi, SEAM, and OpenFWI fall short on spatial extent, temporal extent, geological diversity, and physical realism. We address these limitations with SubsurfaceGen, a GPU-accelerated generator for 3D velocity models and seismic data. Along with SubsurfaceGen, we release a paired dataset of 4,276 2D velocity slices, 5 s wavefields, and 8 s shot gathers drawn from 42 realistic, field-scale 3D velocity models, each spanning 10 km x 10 km laterally and 6.19 km deep at 10 m resolution. The dataset spans six geological settings -- four built with SubsurfaceGen and two drawn from prior sources -- relevant for carbon sequestration and hydrocarbon exploration. We use this dataset to evaluate neural operators on wavefield prediction and encoder-decoders on end-to-end velocity inversion, holding out one geological setting for out-of-distribution testing. These experiments surface failure modes at field-scale and demonstrate how SubsurfaceGen and the associated dataset can impact ML-based FWI.

Disasters are inevitable and increasingly costly, and effective response depends on querying structured tabular data: precise, information-dense records of hazard, exposure, vulnerability, and lifeline infrastructure that underpin disaster management. Current text-to-SQL methods enable natural-language access to such tables but transfer poorly to the disaster domain, where queries span heterogeneous geospatial schemas and require reasoning over causal relations. We introduce DisasterLex, a knowledge-graph-mediated framework that inserts an Expert Knowledge Graph (EKG) of curated concepts and typed causal edges between the user query and the database, bridged to schema by concept-to-table links. The orchestration runs four stages (identifying query entities, routing to the operational domain, planning over causal edges, and grounding the SQL), restricting the schema passed to the model at each step. We instantiate it on a disaster-analytics database (36 geospatial tables, 150 columns) with an EKG of 107 concepts, 117 causal edges, and 52 concept-to-schema links, evaluated on a 75-query test set. On all seven base models spanning proprietary and open-weight families, DisasterLex beats four state-of-the-art baselines (LightRAG, HippoRAG 2, ReFoRCE, CHESS) by 1.4x to 2.75x, with absolute scores of 1.65 to 3.56 (of 5.0). Error analysis shows baseline failures cluster in routing and multi-table SQL composition, the operations our orchestration explicitly addresses. Code, data, and the EKG artifact are available at https://github.com/YimingXiao98/DisasterLex and on Zenodo at https://doi.org/10.5281/zenodo.20388029.

Data selection is increasingly used to reduce the cost of large language model (LLM) fine-tuning, with recent methods prioritizing samples by current utility, diversity, quality, or influence. This paper studies a different question: when fine-tuning occurs over multiple stages, can selection strategies that look optimal now make the model less adaptable later? We introduce a long-horizon view of LLM data selection in which a selector is evaluated not only by immediate task performance, but also by future adaptation speed, forgetting, capability imbalance, and out-of-distribution robustness. We compare representative random, loss-based, gradient-based, diversity-based, quality-based, and utility-diversity selection families under a unified multi-stage protocol. Through controlled experiments designed to instantiate this protocol, we show how short-term selectors can exhibit rank reversal: they improve the current stage while slowing subsequent learning and increasing forgetting. We formalize this behavior as \emph{myopic selection}, provide a simple local analysis of why it can occur, and propose a diagnostic Long-Horizon Aware Selection (LHAS) objective that augments immediate utility with coverage, future-proxy transfer, and anti-concentration terms. The study argues that data selection should be evaluated as a training intervention that shapes the model's learning trajectory, rather than only as a local data-efficiency mechanism.

Sentence-embedding models for semantic search are overwhelmingly developed and evaluated on English corpora. When applied to clinical retrieval in other languages -- particularly retrieval of ICD-10-CM / CIE-10 codes -- recall degrades in ways often masked by aggregate benchmarks. We study whether large generative language models can serve as data factories to close this gap. We build a two-stage retriever (bi-encoder followed by cross-encoder reranker), fine-tuned from a Spanish biomedical encoder (PlanTL-GOB-ES/bsc-bio-ehr-es) on Gemini-generated synthetic data covering English, Spanish, Catalan, Italian, Portuguese and French, and evaluate against BioBERT-ST and the un-tuned Spanish encoder. The bi-encoder alone matches BioBERT-ST on MRR (0.876 vs. 0.866) and overtakes it on R@3 (0.650 vs. 0.626) and R@5 (0.804 vs. 0.790) without English biomedical pretraining. Adding a cross-encoder reranker lifts aggregate R@5 to 0.822 and dominates on four of five languages (+0.017 Spanish, +0.033 Catalan, +0.018 French, +0.037 Portuguese) at the cost of a small English regression. The trade-off is clinically acceptable: Portuguese reaches R@5 = 0.829 vs. BioBERT-ST's 0.714. Contributions: an open recipe for building domain-specific medical retrievers from LLM-generated data; quantification of the learning gain (MRR 0.755 to 0.876, +15.9% with ~19,500 synthetic pairs); and a characterisation of where gains concentrate by language and rank.

Aligned language models often exhibit a recognizable AI-like style, yet its connection to post-training and internal representations remains poorly understood. In this work, we study whether post-training introduces or amplifies AI-like stylistic regularities and whether these regularities have a localized internal signature. To this end, we compare human text, base-model generations, and aligned-model generations under matched human-source prefixes. Aligned generations show lower human-corpus affinity and higher AI-detection rates than base generations, suggesting that post-training shifts generated text away from human-corpus style and toward detector-visible AI-like text. We then introduce PASTA (Post-training Alignment Signature Targeted Ablation), a training-free method that estimates a post-training alignment signature from aligned-base residual contrasts and ablates the corresponding direction during decoding. Across 11 aligned models and 6 AI detectors, PASTA lowers the detection rate for most aligned models; this effect transfers well across detectors and is not reproduced by random directions. Qualitative analysis suggests that PASTA generations remain relevant and coherent while exhibiting greater stylistic variation. Together, these results show that AI-like stylistic effects of post-training can be measured, localized, and causally tested through activation ablation.

Large language models are now adapted through chains of post-training stages rather than through a single instruction-tuning pass. This paper studies whether such sequential post-training gradually compresses internal representations into low-rank, anisotropic, and homogeneous feature spaces. We define a measurement suite for hidden states, logits, token trajectories, and LoRA updates, and we use it to analyze supervised fine-tuning, preference optimization, safety/refusal tuning, math and code specialization, and long chain-of-thought tuning under controlled stage orderings. The central hypothesis is that excessive representation concentration is not merely a geometric curiosity: it predicts reduced plasticity during later adaptation, weaker out-of-domain generalization, and poorer calibration. We further evaluate lightweight interventions, including mixed-domain replay, feature refresh, representation diversity regularization, and LoRA update decorrelation, as ways to preserve future learnability without giving up the behavioral gains of post-training.

Recent work describes what transformers can and cannot compute through connections to boolean circuits, but existing results lack exact characterizations and are sensitive to modeling choices. Padded transformers -- to whose input filler symbols such as ``...'' are appended -- emerge as a useful gadget for establishing equivalences to circuit classes by providing polynomial space for adaptive parallel computation. However, only a limited set of padded transformer idealizations has been studied, leaving open how robustly these equivalences hold under changes to attention type, model width, and uniformity. We find that, under practical assumptions, padded transformers are surprisingly robust to all of these, and identify numeric precision and model depth as the main factors affecting expressivity. Concretely, we prove that polynomially padded $\text{L-uniform}$ constant-precision transformers are equivalent to $\text{L-uniform AC}^0$, while growing-precision ones achieve $\text{L-uniform TC}^0$ regardless of width. Furthermore, looping enables sequential processing analogous to circuits: $\log^d N$-looped constant-precision transformers reach $\text{FO-uniform AC}^d$, and growing-precision ones reach $\text{FO-uniform TC}^d$. Interestingly, growing width or precision beyond logarithmic does not increase expressivity, and all our results hold for both softmax and average hard attention transformers.

Autoregressive (AR) video generation extends videos by producing latent chunks sequentially, but scaling to long videos requires repeated access to a growing historical KV cache. Existing methods reduce this cost by truncating the KV cache or compressing it into implicit memory, but both lose explicit access to query-relevant historical details. We propose OmniMem, an explicit full-range memory retrieval framework that performs sparse KV retrieval over the historical cache. To make this practical for chunk-based AR video generation, OmniMem addresses two issues: (i) local bias in sparse KV selection and (ii) Union Explosion in memory access. Adaptive Window Exclusion removes local-window blocks from the selection candidates when sufficient long-range history is available, preserving the sparse budget for informative long-range retrieval. Query-Shared KV Selection reduces cross-query diversity, while Per-Head Scattered KV Access avoids expanding head-specific selections into a large selected KV buffer. This allows each attention head to retrieve non-contiguous KV blocks according to its own selection pattern. Experiments on long-video generation show that OmniMem improves Dynamic Degree by 52.3% and preserves strong consistency over strong baselines, while maintaining comparable memory usage.

Machine unlearning evaluation is structurally skewed: Why-type questions, which probe causal and relational knowledge, comprise less than 0.06% of CounterFact, 0.6% of ZSRE, and less than 1.3% of TOFU, MUSE, and WMDP-Cyber. This near-zero representation means that methods that fail on causal knowledge can score highly in aggregate, and this failure is undetectable without balanced evaluation. We present 5WBENCH, a balanced 5,000-sample benchmark with 1,000 examples per 5W category (Who, What, When, Where, Why), making causal unlearning failures quantifiable for the first time. Using 5WBENCH, we show that no existing baseline simultaneously achieves high forgetting and high retention on Why-type questions: aggressive forgetting degrades retained knowledge, while conservative methods fail to forget causal facts. Why-type difficulty stems from multi-hop reasoning chains (44% of Why entries vs. less than or equal to 2% for others) and gradient dilution over 40.1-token answer spans. We present MAAT (Multi-phase Adapter-Aware Targeted Unlearning), a three-phase framework operating on LoRA adapter weights, combining gradient-projected ascent, SVD rank-dimension pruning, task vector negation, and hybrid KL-hidden-state retain repair. MAAT is the first method to simultaneously achieve high forgetting and high retention on Why-type causal knowledge, reaching a new operating point on the forget-retain Pareto frontier. We make our code publicly available.