Papers

Dose accumulation is increasingly important in adaptive radiation therapy and reirradiation, but its clinical validity depends on the performance of deformable image registration (DIR). Reirradiation of brain metastases (BMs) with stereotactic radiosurgery (SRS) provides a controlled but clinically meaningful DIR test case: intra-subject brain deformation is usually limited after rigid alignment, yet recurrent lesions can undergo substantial local shape and volume changes that rigid registration cannot capture and can affect dose accumulation. We benchmarked a wide range of learning-based and optimization-based DIR methods on 87 manually screened longitudinal contrast-enhanced T1-weighted MRI lesion pairs from an institutional BM SRS retreatment cohort. Learning-based methods pretrained on healthy-brain MRI were evaluated zero-shot and after instance-specific optimization (ISO) or tumor-proximity target-specific optimization (TSO). Registration was assessed using lesion overlap (Dice), surface distance metrics (HD95 and sASD), target-volume recovery, and runtime and memory. Pretrained learning-based methods showed variable zero-shot performance, while ISO/TSO improved all tested learning-based families. However, optimization-based methods remained the best-performing approach while maintaining reasonable runtime. These findings suggest that even state-of-the-art DIR methods do not yet provide sufficiently accurate and consistent registration for unmonitored use in brain metastasis reirradiation dose accumulation. Because accurate registration is a prerequisite for deformable dose accumulation, clinical application will require case-level quality control and direct assessment of how registration uncertainty affects downstream dose metrics.

AI agents increasingly perform long-term reasoning, planning, tool use, memory integration, and autonomous decision making, yet erroneous intermediate states can propagate and cause inconsistent decisions and unreliable outputs. Existing reasoning approaches mainly rely on iterative planning, self-reflection, augmented memory, or verification, but rarely localize and selectively repair faulty reasoning. We present ORDDAR (Observation-Driven Reasoning for Distortion-Resilient Decision, Action, and Cognitive Recovery), a reasoning framework that models reasoning as cognitive state transitions, detects localized distortions, retrieves related reasoning from prior experiences, and repairs only the affected states. ORDDAR therefore performs recovery at the local reasoning-transition level rather than regenerating the complete trajectory. Experiments across mathematical, commonsense, multi-hop, and clinical reasoning benchmarks demonstrate improved reasoning quality, recovery ability, and interpretability over multiple evaluated reasoning baselines.

Deformable 3D Gaussian Splatting (D-3DGS) reconstructs dynamic scenes by deforming a canonical set of 3D Gaussians through a deformation field of frame time. Replacing its MLP with a stack of Closed-form Continuous-time (CfC) cells-a Liquid Neural Network that solves the Liquid Timeconstant ODE in closed form-gives the field continuous-time behaviour at feed-forward cost. That closed form, however, is only the deterministic limit of a noise-driven system, and drops the stochastic term usually credited for the robustness of liquid networks. We put it back: a small Gaussian perturbation is added to the time gate of every CfC cell, turning the deterministic field into a simple stochastic (SDE) one. The noise is used only during training, needs no solver, and reduces exactly to the CfC when switched off. On the synthetic D-NeRF scenes the stochastic field is on par with the deterministic CfC and beats the MLP baseline on most scenes; on the real-world NeRF-DS scenes the deterministic limit is already best and adding noise does not help. The study thus gives both a clean way to read the CfC field as an SDE and an honest account of when a plain noise term helps and when it does not.

Diagram-to-graph topology extraction aims to extract a graph of entities and their connections from a structural diagram. This task remains challenging for current vision-language models because it requires both fine-grained perceptual grounding and topology-aware reasoning with global consistency. We present TopoBench-180, a human-verified benchmark for diagram-to-graph topology extraction, and TopoAgent, a structure-aware perception-to-reasoning framework for reliable topology extraction using large vision-language models. TopoBench-180 contains 180 structural diagrams spanning Web-style and Network-style categories, paired with canonical graph annotations. TopoAgent progressively extracts the target graph by combining grounded perception, global structural priors, canonical node inventory construction, node-centric local-to-global relation reasoning, and topological consistency enforcement. Experiments on TopoBench-180 show that TopoAgent outperforms strong vision-language model baselines and recent visual reasoning frameworks, especially on edge extraction. More broadly, this work fills an important gap in multimodal structured understanding by establishing a benchmark and framework for diagram-to-graph topology extraction. The benchmark and associated resources will be publicly released at https://huggingface.co/datasets/WayneGuo0011/TopoBench-180.

Understanding long-form video remains a fundamental challenge for multimodal large language models (MLLMs). Sparse frame sampling fails to capture fine-grained visual details, while dense sampling quickly exceeds context length limits. Retrieval-augmented generation (RAG) offers a promising middle ground by selectively retrieving relevant video segments for grounded generation, yet its effectiveness critically depends on the quality of the video segments used as retrieval units. In this paper, we investigate RAG for movie understanding, which demands story-level reasoning over characters, events, and narrative arcs spanning hours of content. Scene segmentation, a long-studied problem that partitions movies into semantically coherent units, is a natural candidate for defining such retrieval units. We reexamine whether existing methods actually serve this role through comprehensive evaluation on downstream movie understanding tasks, and find that they do not provide a consistent or dominant advantage over naive uniform temporal chunking. Our audit of the most standard scene segmentation benchmarks reveals why: current annotations prioritize visually salient transitions over narrative event structure. Motivated by this mismatch, we introduce NarraScene, a narrative-centric scene segmentation dataset annotated with a three-level cognitive taxonomy spanning physical, character, and narrative change, where every valid boundary requires a narrative-level shift. When used as retrieval units, these narrative-grounded segments outperform uniform chunking on downstream movie understanding tasks, suggesting that the central challenge for scene segmentation in movie RAG is not detecting boundaries, but identifying the narrative event units that matter for movie understanding. The dataset is available at https://huggingface.co/datasets/queez0405/NarraScene.

Compared with typical vision-language tasks, document parsing places stronger demands on fine-grained visual perception. Existing vision-language model (VLM)-based parsing approaches rely on globally compressed visual tokens, where fine-grained details are entangled within a single representation and repeatedly accessed during decoding. However, we observe that the visual evidence for each prediction is typically localized and conditioned on the current decoding state, whereas such representations must be accessed in full at every decoding step, resulting in inefficient computation. To address this mismatch, we formulate perception as state-conditioned visual evidence retrieval (SCVER) during autoregressive decoding. The model operates on a compact global representation for coarse structure and retrieves a small set of relevant high-resolution regions conditioned on the current token state. This coarse-to-fine design enables on-demand access to fine-grained visual cues, relieving globally shared representations from encoding all fine-grained details. We further find that learning such state-conditioned retrieval in VLMs is challenging and unstable. To stabilize this process, we introduce a Spatially-Guided Learning Objective (SGLO) to guide the retrieval process. Experiments on document parsing benchmarks show that SCVER improves robustness under reduced input resolution and achieves a better accuracy-efficiency trade-off, demonstrating the effectiveness of on-demand visual evidence retrieval for fine-grained perception.

Additive manufacturing (AM) enables compact hydraulic components whose internal fluid channels can follow free-form 3D paths rather than conventionally drilled holes. A representative case is multi-group channel layout in rotary direct-drive servo valves: once a channel is placed it permanently occupies volume, so later channels must clear earlier geometry–unlike classical multi-agent pathfinding (MAPF), where agents free space after moving. We study this setting as multi-group pipe routing under permanent geometric occupancy. Our contributions are a problem formalization with geometric dual-witness conflicts, a constructive 3D benchmark (two corridor generators x two obstacle painters, controlled difficulty, feasibility witnesses), and baseline results for CBS, PBS, and priority planning adapted to this coupling. We evaluate by success rate under a fixed time budget. PlaneSlice hard instances clearly rank CBS above PBS and PP, while easier cells validate the pipeline. The goal is a reproducible problem definition, suite, and classical baselines–not a new optimal MAPF algorithm.

Visual in-context learning (ICL) with multimodal large language models (MLLMs) is effective for fine-grained visual classification, but each retrieved image example consumes several hundred context tokens, making large-$K$ settings prohibitively expensive at inference scale. We propose Instruction Distillation: an offline procedure in which the MLLM itself generates, for each individual training image, a structured identification instruction encoding general appearance cues, features that differentiate the class from visually similar ones, and a common confusion point. Unlike prior work that produces a single description per class, our instructions are generated per training image, preserving the intra-class visual diversity that per-class descriptions collapse. At inference time, we study five configurations sharing a single CLIP retrieval index: zero-shot, image ICL, instruction-only ICL, and two hybrid variants in which retrieved neighbors are split between images and instructions. Across seven fine-grained benchmarks and two MLLM backbones, instruction based pipelines match, or exceeds image ICL at $K{=}1$ and reduces per-query tokens by $2.9\times$ and inference latency by $3.3\times$ at $K{=}5$. Hybrid configurations further show that visual and textual ICL signals are complementary, images give visual patterns to learn and see, while instructions give explicit rules and logic. When both of these are provided, the quality of context improves, which is noticeable in the performance.

Image retrieval has traditionally been formulated as a point-wise matching problem, where each candidate image is scored in isolation. However, this atomic paradigm fails to capture the complexity of human search intent within personal photo collections, where users often seek compact visual stories bound by structural relations rather than isolated snapshots. To address this limitation, we introduce **Image Bundle Composition (IBC)**, a novel paradigm that shifts the objective from ranking individual images to dynamically composing cohesive image bundles from a massive, unstructured photo pool. Since target bundles are not predefined, IBC presents a severe combinatorial explosion challenge and demands modeling non-decomposable joint relevance. To establish this paradigm, we construct **IBCBench**, the first IBC benchmark dataset containing 109,467 images and 667 verified queries, built via a semi-automated verification pipeline. Furthermore, we propose **BundleWeaver**, an agentic framework that reformulates IBC as query-conditioned incremental hyperedge discovery. By employing a Large Language Model to adaptively search for missing relational roles and utilizing a Vision-Language Model for whole-bundle verification, BundleWeaver effectively navigates the combinatorial space. Extensive experiments demonstrate that while state-of-the-art embedding models and static decompose-and-rerank paradigms suffer from relational blindness, BundleWeaver achieves substantial performance gains, highlighting the necessity of shifting from atomic scoring to dynamic relational composition. Our dataset and code are available.

Long-horizon video generation is evaluated with whole-frame metrics that reward motion and temporal consistency. For fixed-camera nature scenes this creates an ambiguity: motion of water, fire, smoke, or rain is desirable, whereas motion of the background is an error. A system can therefore score well on motion while its scene drifts, or on consistency while its flow stagnates. We introduce SNF-Bench, an evaluation framework for long-horizon fixed-camera generation that partitions each scene into static support and dynamic flow and reports static fidelity, flow persistence with absolute magnitude, and drift leakage separately, never as one score. Drift leakage is interpretive context rather than a headline measurement. Each factor is validated mechanistically rather than by correlation with preference: we inject global translation, rotation, and scale drift and progressive late freezing at known severity into real generations, and require each factor to respond in its stated direction and to remain selective against corruptions it does not target. Auditing publicly released long-horizon text-conditioned checkpoints under one recorded common inference configuration, plus an image-conditioned track with released-pipeline references and a deployment-sensitivity panel, we find that whole-frame motion and static-region drift induce near-opposite orderings of the same outputs. At maximum controlled translation, fBD and NBF rise to $1.86\times$ and $1.32\times$ baseline, but whole-frame Dynamic Degree reaches only $1.07\times$—rewarding the corruption. SNF-Bench measures where motion occurs and whether it persists; it does not measure physical realism. Project page: https://minar09.github.io/snfbench/.

Enabling humanoid robots to respond to human speech with synchronized and semantically meaningful gestures is fundamental to natural human-robot interaction. However, this task faces three critical barriers: the scarcity of semantically rich datasets, the "modality eclipse" where models ignore audio cues in favor of kinematic inertia, and the sim-to-real gap regarding physical safety. We propose RoboGesture, a robot-centric framework that co-designs data, modeling, and control to power a complete interactive human-humanoid system in which the robot listens, responds, and gestures in real time. We first establish the RoboGesture dataset featuring over 300 gesture categories and develop an automated pipeline to synthesize large-scale collision-free, robot-specific audio-motion pairs. Our architecture features a Hierarchical Semantic-Acoustic Aligner that extracts multi-granular prosodic and semantic cues directly from raw audio tokens. These cues drive a Streaming Conditional Motion Generator based on a diffusion transformer with conditional flow matching. To ensure high responsiveness, we introduce Anti-Inertia CFG Masking, which prevents the model from collapsing into repetitive historical patterns by compelling it to proactively mine control signals from the audio modality. Finally, an MPC-based safety filter ensures real-time, collision-free execution on physical hardware. Experiments on a Unitree G1 humanoid demonstrate that RoboGesture generates safer, more rhythmic, and more semantically appropriate responses compared to state-of-the-art baselines.

In closed forests, neighboring tree crowns often stop short of touching, leaving a network of narrow gaps known as crown shyness. The pattern belongs to the stand rather than to any single tree, which makes it a natural probe of a question in generative modeling: can a learned model produce a set of objects whose defining structure exists only between them? We formulate stand-level canopy generation as set generation with a flow-matching model, in which attention between trees is the only channel through which coupling can arise. Trained on stands grown by a resource-competition simulation that is provably not reducible to per-tree geometry, the joint model halves the clearance distribution error of an identical-capacity model that generates each tree alone, and the advantage persists at stem densities outside the training range. Against field measurements of a tropical oak forest, a single calibrated scalar yields held-out agreement in gap magnitude and crown asymmetry. The directional statistics of the gaps are controlled by stem placement rather than by the growth rule, and match the field once stem jitter is calibrated. Crown shyness, in both the simulation and the learned model, is a property of the stand and not of the tree.

We study audio-visual conflict as a compositional generalization test for AV-LLMs: the model must combine synchronized but semantically incompatible audio and video evidence and decide whether the pair matches. On VideoLLaMA 2-7B-AV, three alignment configurations remain nearchance on the scored exact-string Yes/No subset of AVHBench, even though their output priors shift substantially. Similarly, off-the-shelf InternVideo2 experienced a 32.3% accuracy decrease specifically under cross-modal conflict, accompanied by a 17.3% instruction-following failure. We call this failure mode prior dominance: late-layer commitment to an internally preferred answer pattern that is weakly grounded in the conflicting inputs. To explain this behavior, we conduct a mechanistic interpretability analysis and find that commitment remains concentrated at 25.5 $\pm$ 1 layers. We show that stronger temporal alignment changes answer bias, but do not improve compositional conflict resolution. Code and data to reproduce our mechanistic audit and behavioral evaluations are available at https://github.com/AdarshSudheer09/AVHBench-dmai.

Speech language models (speech LLMs) can generate plausible outputs from audio that contains no usable speech evidence. We study this failure as a pre-generation support-estimation problem and present SURE-Voice, a training-free front end that decides whether an audio prompt contains intelligible speech evidence before calling a speech LLM. We build SURE-Challenge with a 640-example SURE-Core split and a 1,920-example SURE-Extended split derived from 120 LibriSpeech source utterances. Using one fixed operating point, an energy screen plus Whisper token confidence raises unsupported accuracy on the held-out Extended test from 0.000–0.133 to 0.919 for six non-degenerate speech LLM backbones, while supported accuracy remains 0.919–0.970 and downstream calls fall from 480 to 287. A 500-clip ESC-50 sanity set shows the same pattern on real environmental audio, with vocal non-speech as a residual failure mode. An overlap diagnostic shows that source attribution remains separate from speech-evidence filtering. The evidence supports a controlled benchmark baseline and a deployment-oriented analysis; it does not establish universal robustness to semantic answerability, gain variation or natural conversations.

Memorization in large language models is measured through a zoo of definitions whose formal relations are unknown, and differential privacy (DP) is treated as a proxy against all of them at once. We pin down the exact DP constant for the two that carry the practical weight, counterfactual memorization and adaptive extraction, and show that they do not control each other. Under $f$-DP, every adaptive extraction protocol with list budget $m$ succeeds with probability at most $1-f(\kappa)$ for the oblivious baseline $\kappa$, and the bound is tight on a dense set of baselines: DP uniformly controls extraction exactly up to a threshold in how well the secret can be guessed a priori. Min-entropy certifies that baseline distribution-free, since $H_\infty\ge\epsilon\log_2 e+\log_2(m/\tau)$ holds extraction below a risk level $\tau\le1/2$ under pure $\epsilon$-DP for every prior, and is exact on uniform priors. On the memorization side, $f$-DP caps the counterfactual memorization of any bounded score at an advantage functional $\eta(f)$, equal to $\tanh(\epsilon/2)$ under pure DP; for $k\ge2$ duplicated copies the naive $\epsilon\mapsto k\epsilon$ bound $\tanh(k\epsilon/2)$ is unattainable, the exact constant being a closed-form staircase attained by geometric noisy counting. That cap is attained inside the local score class used in practice, and it is there that the two measures separate: one mechanism is memorized yet unextractable, another fully extractable yet exactly invisible to every loss-based score. The two-sided blind spot this opens for loss-based auditing and unlearning verification survives on billion-parameter models: a reserved-trigger release is recovered verbatim from one prompt while the audits practitioners deploy certify it clean.

Recent severe storms on the U.S. Gulf coast have demonstrated the challenges presented by compound flooding, such as the interactions between rainfall runoff and storm surge. Historically, many studies have neglected these nonlinear interactions, but we propose to use a discontinuous Galerkin shallow water equation solver, which allows for incorporation of rainfall inputs directly onto the finite element mesh. In this work, we analyze the use of parametric rainfall for forecasting scenarios, using Hurricane Beryl (2024) as a case study. Beryl led to extensive flooding due to rainfall and storm surge along the Gulf. We use a collection of the National Oceanic and Atmospheric Administration's short-term advisories along with the best track data to demonstrate the efficacy of the parametric rainfall model for forecasting. Results show that the parametric rainfall input allowed for much more accurate inundation. Areas with heavy rainfall and low surge were affected the most, with many areas peaking over 50 cm above the baseline surge model. Almost none of the available high water marks from Beryl were captured by the standard models, but the compound flooding models capture many of them, the majority of which show relative errors under 10 percent. Results from the advisory forecast simulations were shown to be much closer to the best track hindcast simulation when rainfall forcing was used, even while early forecasts predicted the storm's trajectory much less accurately. The advisory simulations improved even further as Beryl neared the Texas coast, with sampled peak elevations most closely approximating the best track at Advisory 38, a few hours before landfall.

We study the edge-power cost of the labelled tree generated by the $\gamma$-strategy, a constant-gain rule for sequential Euclidean connections. Starting with $x_0=p_0$, each input point $p_i$ is attached to $x_{i-1}$, and the state is updated by $x_i=\gamma x_{i-1}+(1-\gamma)p_i$. Retaining $x_i$ subdivides the insertion segment into a spine edge and a leaf edge. The memory parameter $\gamma$ controls how long earlier points influence subsequent attachment points. We minimize the sum of the $\alpha$-powers of the edge lengths under independent uniform input and arbitrary input sequences. For uniform points in the unit ball, the stationary problem has a transition at $\alpha=1$. Its continuous extension is minimized at the boundary for $0<\alpha\leq1$, while every global minimizer is interior for $\alpha>1$. We determine the finite optimizer in the joint window $\alpha_N=1+\varepsilon_N$, $\varepsilon_N\log N\to\lambda$. Below an explicit threshold it lies on the $N^{-1/2}$ scale, at the threshold its scale is $\sqrt{\log N/(N\log\log N)}$, and above the threshold it approaches an explicit stationary root with two computable corrections. A second threshold identifies the governing correction, and differentiated estimates prove eventual uniqueness. At $\alpha=3d+8$, the linear coefficient at the stationary endpoint changes sign and a branch of strict local maxima enters the parameter interval. For arbitrary input sequences, the optimal parameter and asymptotic worst-case edge-power cost per point are explicit for $0<\alpha\leq3$. At high powers, periodic antipodal block inputs give explicit lower bounds which, with a separation argument, show that the optimized cost is asymptotic to $2\log2/\log\alpha$. Exact results for powers two and four, a rational recursion for every even power, and a high-dimensional expansion complete the analysis.

We study a PDE-constrained optimization problem for an elliptic equation with the spectral fractional Laplacian and a linear combination of Dirac measures as the forcing term; the controls are the amplitudes of these singular sources. We prove existence and uniqueness of an optimal solution and derive first-order optimality conditions. We then propose a discretization based on finite elements. Since the set of admissible controls is finite dimensional, the control variable itself does not require discretization. We conclude by deriving a priori error bounds

Efficiently and robustly analyzing shape data is critical across many scientific disciplines. While chirality is a fundamental property in numerous applications - most notably in molecular science - existing shape analysis metrics fail to distinguish between a shape and its mirror image. To address this gap, we introduce a multilinear generalization of the Gromov-Wasserstein objective. Under mild assumptions, this objective yields a distance between shapes, represented as probability distributions quotiented by a symmetry group $G$. In particular, for $G = SO(d)$, we introduce the Chiral Gromov-Wasserstein ($\mathrm{CGW}$) distance, sensitive to chirality. We establish robustness properties for the multilinear Gromov-Wasserstein distances and develop efficient algorithms to compute them, reformulating the underlying optimization problem by projecting couplings onto a low-dimensional space. We derive algorithms for both local and approximate global solutions, yielding a fully polynomial-time approximation scheme for these problems. We validate the framework through numerical experiments that demonstrate the effectiveness of $\mathrm{CGW}$ as a shape metric for chiral objects.

We describe the design of a custom crowbar circuit that quickly removes power when a programmable current threshold is exceeded. This protects devices from permanent damage during single event latchup/burnout tests.

A language model with access to tools can commit to a final claim unsupported by the evidence it has seen, even when a single available tool call would resolve the uncertainty and its instructions explicitly forbid assumptions and guesses. We separate this failure into two precisely defined quantities: occurrence, how often the model makes an unsupported claim on its own, measured from the visible evidence and final claim without using the hidden correct answer; and conditional repair, how often those same naturally occurring unsupported claims are repaired when the missing evidence is supplied. On one fixed Qwen3-32B setup, 33 of 512 first responses to 256 new prompt templates ended with an unsupported established claim. We replayed each case from an exact copy of the state in which the claim occurred; within each matched replay, the alternative tool responses had the same structure and length and differed only in a one-character response code. Resolving evidence repaired 33 of 33 claims; a matched response carrying no useful information repaired 0 of 33. When the evidence supported the original answer, the model preserved 33 of 33, with no observed harm. In a separate experiment, on 64 cases where evidence was needed, an automatic checking rule added 21 evidence calls, corrected all 10 wrong unsupported claims, preserved the 11 that were correct by accident, and never changed a correct answer into a wrong one. On a fixed Gemma 4 setup using the same sampling settings, the model called the tool in all 512 first responses and never made an unsupported final claim, so conditional repair could not be measured for that setup. These results describe two local fixed model setups on two synthetic task families. They do not show how common this failure is in real-world deployments, nor that it reflects a general mechanism shared across models.

Combining secure multi-party computation (MPC) with differential privacy (DP) enables multiple parties to release aggregate statistics without a trusted curator, and the core primitive is the protocol to sample noise from a continuous distribution under finite-precision arithmetic. In this paper, we revisit the continuous noise sampling protocols and present several improvements in both security and efficiency. We start by identifying a vulnerability in widely used sample-and-scale constructions. We demonstrate that the scaling operation in arithmetic circuits confines the noise to a sparse, publicly known set of values, so that an adversary can observe the released noisy queries and decide which dataset produced them. As concrete demonstrations, we instantiate attacks on two systems employing such “flawed” sampling protocols: Orchard (OSDI'20) for DP secure aggregation and DP-BREM$^+$ (USENIX Sec'25) for DP federated learning. We report a near-$100\%$ attack success rate on both systems, under any noise scaler $s\geq 2$ used in practice. The leakage we reveal is intrinsic to the scaling operation, and direct repairs either substantially sacrifice utility or add significant precision bits to make the sampling more expensive. To address the security and efficiency issues together, we turn to discrete sampling at the granularity of individual biased bits. We make several optimizations to the sampler and prove its security. Our implementation achieves $4\times \sim 612\times$ speedup over existing secure discrete samplers and orders-of-magnitude speedup over the insecure sample-and-scale paradigm, with negligible utility loss compared to the ideal continuous mechanism.

Recurrent fast-weight memories and selective state-space models compress an expanding context into a fixed-size recurrent state, making the state transition an online learning rule. We study this rule under read-after-write autoregressive semantics. For the prefix-prediction objective considered here, the local fast-memory example revealed at step $t$ is the prefix-aligned pair $(\mathbf{x}_t,\mathbf{y}_t)=(\phi(\mathbf{k}_{t-1}),\mathbf{v}_t)$. The common same-step association $(\phi(\mathbf{k}_t),\mathbf{v}_t)$ remains causal, but optimizes a different internal objective. We derive normalized first-order updates for squared-error regression and negative inner-product objectives. The regression family comprises Falcon-1 (a scalar NLMS update), Falcon-2 (its per-column extension), and Falcon-3 (a sliding-window mini-batch update); Falcon-1A/Falcon-2A/Falcon-3A are the corresponding inner-product variants. We provide recurrent, masked-parallel, and chunk-parallel forms, together with numerically stable positive-decay renormalization. Representative variants remain competitive in language modeling and improve length extrapolation on variable-digit addition. This framework separates temporal alignment, plasticity, forgetting, and bounded rehearsal in recurrent sequence models.

We consider the ability of transformer-based language models (LLMs) to learn what we call k-antilocal languages, i.e., languages that have no mutual information across any span of $k$ contiguous symbols. We construct such languages with increasing $k$, finding that LLMs trained on them achieve comparable cross-entropy loss regardless of antilocality, but converge more slowly on more antilocal languages. Our findings support the idea that non-local dependencies are more difficult to learn, but the evidence for this bias comes from learning speed rather than learning success.

Consistent query performance is essential for mission critical database applications, yet SQL execution plans can change due to factors such as database upgrades, DML changes, new indexes, etc. While plan stability mechanisms such as stored outlines prevent regressions by freezing execution plans, they also inhibit performance improvements by disallowing plan evolution. We introduced SQL Plan Management (SPM) in Oracle 11g to address this trade-off by maintaining a set of accepted execution plans and allowing plan evolution only when new plans demonstrably outperform existing baselines. However, prior implementations of SPM primarily rely on background performance verification processes, delaying regression detection and recovery. This issue is amplified in autonomous cloud database systems, where several automatic actions that could cause plan change driven regressions are performed with limited customer control. Timely detection and remediation is paramount, but the constrained background resources on cloud may not keep pace. To overcome these limitations, we introduce Real-Time SPM in Oracle 26ai, a novel extension of SPM that performs foreground verification of new execution plans during user query execution. Real-Time SPM leverages runtime session context to immediately validate plan changes, enabling rapid adoption of superior plans while promptly detecting and preventing regressions. This paper presents the architecture and design of Real-Time SPM - including technical challenges like reliably comparing performance of previous plans - and contrasts it with traditional background plan evolution.

Searchless chess networks reach human master strength from a single forward pass by imitating a stronger teacher: the strongest, Leela Chess Zero's (Lc0) released Chessformer, distills the visit counts of an AlphaZero-style Monte Carlo Tree Search (MCTS). Imitating a search is a poor proxy for playing without one, so we fine-tune for single-pass strength with self-play reinforcement learning (RL). Its exploration is usually supplied by an entropy bonus, the reverse Kullback-Leibler (KL) divergence to uniform. We replace it with a forward, mass-covering KL toward the network's own MCTS prior (prior-directed exploration), so exploration covers the moves the prior judges promising, and pair it with an entropy-adaptive sampling temperature, set by the value head's outcome uncertainty, that sharpens once a position is decided. In about two thousand steps it raises puzzle accuracy from 93.9% to 94.9% on a 100,000-puzzle suite and mate-in-four accuracy from 77% to 81% while holding searchless strength at or slightly above the base. Measuring tactical accuracy and playing strength together across a matched-compute sweep, we find the two dissociate: accuracy gains fall in a one-point band while ratings straddle the base, and a control fine-tuned on puzzles alone posts the study's largest tactical gains while shedding roughly 260 Elo; a better puzzle-solver is not thereby a stronger player. Distribution-level measurements show what anchoring buys: without a regularizer self-play collapses onto a single line of play, and the puzzles newly solved are the near misses whose winning move the prior kept alive. The forward-KL prior tops the rating ladder, statistically tied with a reverse-KL anchor that concentrates twice as hard and drops the hardest solutions the mass-covering prior keeps in support.

Determining whether large language models derive answers from context or prior knowledge remains a fundamental challenge. Self-contained linguistic olympiad puzzles provide a controlled setting where all answers derive solely from expert-designed context examples without external knowledge. Removing individual context examples can eliminate information needed for specific questions while leaving the rest of the puzzle unchanged. We leverage this to introduce a diagnostic framework for analyzing individual context examples. Using 53 UK Linguistics Olympiad puzzles, we generate two modified variants by deleting a single context example: (1) uniform random deletion, and (2) targeted deletion (inspired by error-correcting codes) to remove a structurally load-bearing example uniquely carrying necessary information. We formalize this impact using a Question Damage Score to classify puzzles as fragile or robust. Evaluating three frontier LLMs under instructions to abstain when information is insufficient, we find they rarely abstain, often continuing to produce correct answers after load-bearing context is removed. These findings motivate further investigation into context-based reasoning, prior knowledge, memorization, and linguistic inference. Beyond abstention, the framework enables fine-grained analyses of context reliance, including causal interventions, stopping-set analysis, targeted contamination studies, and mechanistic interpretability.

We study a family of language operations based on insertion, shuffle, and crossover and investigate the undecidability of adjacent equality together with finite convergence and associated spectrum questions. Insertion and shuffle operations on formal languages arise in formal language theory, models of concurrency, and biologically inspired computation. This paper studies a different question from the usual closure problem, specifically whether an increasing sequence of languages generated by repeated insertion, or by increasing the permitted degree of bounded shuffle, reaches an instance of adjacent equality after finitely many stages. We show that several such adjacent equality questions are undecidable. In particular, reaching such an adjacent equality event is undecidable for each of the following: iterated insertion of a regular language into a context-free language; bounded shuffle of a regular language with a context-free language as the bound increases; and the corresponding self-insertion and self-bounded-shuffle hierarchies for context-free languages. The new reductions proceed directly from the undecidability of context-free-language universality, using separator-delimited block constructions and, for self-operations, an absorbing regular language of guard violations. Earlier trace-based proofs relied on mortality and uniform halting. More generally, we investigate finite-stage equality and stabilization (persistent equality) in hierarchies generated by insertion and bounded shuffle. In addition to giving substantially simpler proofs of earlier undecidability results, we obtain general criteria for one-step equality, develop new reductions for self-insertion, and identify several open problems, including structural questions concerning insertion depth and degree whose resolution determines whether adjacent equality necessarily implies permanent stabilization.

Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report the practical innovations each stage required. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latents activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations. The results also suggest a possible tension: the cross- seed stability might be selecting for some types of features, like structure-related ones, much more than others, such as function-related features.

Despite the rapid advancement of Vision-Language Models (VLMs), their linguistic reach remains largely confined to high-resource languages, leaving the majority of the world's 7,000+ living languages on the wrong side of a growing digital divide. This disparity is especially pronounced in Optical Character Recognition (OCR), where low-resource scripts lack the massive datasets required for traditional scaling laws. We investigate OCR adaptation in extreme data-scarce regimes (<10K real and <250K synthetic images), demonstrating that conventional fine-tuning strategies often reach a performance ceiling. Our key finding reveals a structural inefficiency in language-specific adaptation: while higher layers of specialized models diverge to capture unique script nuances, the lower layers learn redundant, highly similar features. Motivated by this observation, we propose PSMC (Pre-train, Specialize, Merge, and Co-train), a data-efficient framework that capitalizes on a cross-script "transfer effect". Our approach first derives language-specific experts from a high-resource base model, then employs task arithmetic to fuse these experts into a unified, high-performance multilingual back- bone. Extensive evaluation across 10 Indian scripts (supporting 20+ languages) shows that PSMC achieves a ~2% average improvement in Word Recognition Rate (WRR) over individual specialist models without increasing parameter count. Our results indicate that joint training in the merged latent space facilitates a constructive knowledge transfer that benefits all constituent scripts, providing a scalable pathway for inclusive VLM development. Source code and datasets will be released post publication.