Purpose: Surgical operative reports support procedural documentation, surgical training, performance assessment, and workflow review. However, extracting meaningful information often requires extensive manual review and clinical expertise. Automated surgical scene descriptions could reduce this burden by leveraging recurring surgical activities and prior knowledge. Methods: We propose GRAC (Graph-Based Retrieval-Augmented Captioning), a framework that represents surgical scenes as graphs capturing relationships among instruments, actions, and anatomical structures. Given an input scene graph, GRAC learns to retrieve relevant graph-caption pairs from training data. It compares instrument-tissue interactions between retrieved examples and the input scene, using comparison features to condition retrieved-caption representations during generation. Results: On the EndoVis-18 benchmark, GRAC achieved a CIDEr-D of 5.88, compared with 4.23 for SGT and 3.36 for the strongest conventional Transformer baseline. Exact instrument-action-target triplet-set accuracy reached 52.6%, compared with 32.7% for SGT and 26.2% for the strongest conventional baseline. These results demonstrate improvements in language quality and factual agreement. On TORS, GRAC achieved a CIDEr-D of 5.06, compared with 3.80 for the strongest baseline. Conclusion: GRAC demonstrates that combining structured surgical scene representations with retrieved examples improves surgical scene captioning and produces more contextually meaningful descriptions.
The Global Ecosystem Dynamics Investigation (GEDI) mission measures three-dimensional forest structure with a LiDAR aboard the International Space Station. GEDI observations have geolocation errors of about 10 m, which cause misalignment with other remote sensing data and reduce the accuracy of downstream tasks. Matching GEDI ground elevations against high-resolution digital terrain models (DTMs) from airborne LiDAR is one way to correct these errors, but such DTMs exist in few regions. To widen this method's applicability, we super-resolved a more widely available low-resolution DTM and used it for the matching. Our question is whether a super-resolved DTM yields the same correction as a 1 m-resolution airborne LiDAR DTM and, if not, why. In steep forests of western Japan, we super-resolved an existing 10 m-resolution DTM to 1 m. Improving its accuracy required fusion with a global 30 m-resolution DTM from a different source. Using the resulting DTM, we corrected GEDI observation positions first by a single shift per ground track, then per observation. For most tracks, the shifts nearly matched those from the airborne LiDAR DTM, likely because DTM errors canceled out across many observations. Per-observation corrections, however, did not match. At a single observation, the error does not cancel out and is too large for the correction. This error likely originates mainly from the input DTM. Therefore, even without a 1 m-resolution DTM, a roughly 10 m-resolution DTM, fused with a global DTM and super-resolved to 1 m, should yield track-wise corrections nearly matching those from a 1 m-resolution DTM.
The extent to which demographic-associated information is detectable across organs and disease states in routine clinical histopathology remains unclear. Here, we analyzed 1.3 million hematoxylin and eosin whole-slide images from 257,322 patients across 25 organs, using pathology foundation-model representations and multiple-instance learning to systematically map information associated with sex, age group, and self-reported race. Across 69 organ-demographic combinations, demographic attributes were broadly recoverable, with pan-organ mean AUROCs of 0.828 for sex, 0.733 for age group, and 0.662 for self-reported race, and substantial variation across organs. Demographic-class balancing produced relatively small changes in performance, while evaluation within diagnosis-text clusters showed modest attenuation. Under training with diagnosis-based grouping, demographic-associated signals remained detectable across multiple organs, including sex in gastric tissue (AUROC 0.865), age group in breast (0.816), and self-reported race in skin (0.728). Literature-guided analysis of these settings generated candidate histomorphological and contextual explanations for further investigation. Together, these findings provide a systematic map of demographic-associated information in routine histopathology, informing investigations of its sources, assessments of potential shortcut learning, and the development of more interpretable computational pathology models.
Can a sender that only computes hashes, with a language model only at the receiver, transmit short human text more cheaply than a classical sender trained on the same domain? We answer with one pre-registered run on 1,000 English user utterances from Taskmaster-1. The primary endpoint is the cost per message to deliver 100% of the messages (hash bits, plus a 2-byte negative acknowledgement and the raw text when the first attempt fails), paired against an in-domain PPM order-1 sender. The hash-only sender delivered 857 of 1,000 messages on the first attempt and cost 192.8 bits per message, 18.4 bits more than PPM order 1 (95% bootstrap interval 10.9 to 26.3); by the pre-registered rule it spends more and is not equivalent within one byte. As secondary results we map measured sender memory against bits. The hash-only sender (8,190 B on a host build, keyed BLAKE2b tag included, as in every method) was the cheapest point in bits from 8,190 B up to, but excluding, PPM order 1 at 24,844 B, 45 to 46 bits below smaz2 and Unishox2; from PPM order 1 upwards, in-domain senders spend fewer bits, except order 2 pruned to 16 KiB. In airtime it loses to PPM order 1 on all five links priced, to Unishox2 on four, and to raw text with a tag on two, because each failure costs an acknowledgement frame and a fallback frame. A wider receiver search recovers half of the failures and makes the cost 7.7 bits lower than PPM order 1, at about three minutes of receiver time per failure, and still loses in airtime on four of five links. We report the measurement as a delimitation, with every deviation of the run.
In many areas of AI, general-purpose models have increasingly replaced handcrafted methods and specialized architectures. We argue that embodied AI is poised for a similar shift. Our key claim is that embodied tasks can be formulated as interaction with digital software. Specifically, we build software that renders an agent's view of a physical environment on a screen and provides controls for acting in it. A computer-use agent (CUA) can then perform embodied tasks through the same screenshots and keyboard and mouse actions that it uses to operate ordinary software. Using a single CUA design and changing only the software, our entries win first-, second-, third-, and sixth-place prizes across four ECCV workshop competitions, without any competition-specific agent tuning. These results suggest that the skills CUAs learn on software, such as visual grounding, precise clicking, and exploration, already carry over to embodied tasks, giving CUAs a far broader scope in embodied AI than previously realized. We therefore advocate that computer-use agents can be embodied generalists. Further, we advocate that, by following the training strategy that made computer-use agents general, embodied AI can converge on a single general model as well, ushering in a new era of general embodied AI. Website: https://pranjal2041.github.io/cua-embodied
Plug-and-play (PnP) methods reconstruct images by alternating a data-fitting step with an image denoiser. Their performance depends on two parameters: the step size $\gamma$ and the measurement noise level $\sigma$. In practice, $\gamma$ is often tuned by grid search over validation images and must be retuned whenever the problem setting changes, and $\sigma$ is assumed known or estimated separately. We propose AdaptivePnP, a PnP framework that jointly estimates $\sigma$ and $\gamma$ at every iteration from the data-fitting residual. To do so, we apply the PnP virtual model to the data-fitting proximal map and show that the residual is zero-mean Gaussian with covariance parameterized by $\gamma$ and $\sigma$. However, when one parameter is much larger than the other, the residual carries little information about the smaller one and using the maximum likelihood (ML) estimator can degrade the PnP iterations. We therefore introduce the identifiable maximum likelihood (IML) estimator, which restricts the ML estimator to an identifiable window in which the normalized Cram\'er--Rao bound of every parameter is bounded by a user-specified tolerance. For the virtual model, this window is an interval of ratios $\rho = \gamma/\sigma$, so computing the IML estimate reduces to a one-dimensional search over $\rho$. We apply AdaptivePnP to two standard PnP methods, yielding algorithms that require neither a tuned $\gamma$ nor a known $\sigma$. On image deblurring and super-resolution, AdaptivePnP matches or outperforms PnP baselines tuned with the true $\sigma$. Code is available at https://github.com/hdubasullivan/AdaptivePnP.
The pigs were bleeding. That was the point of the experiment: could airway pressure reveal hemorrhage before it became critical? In the UCR Time Series Classification Archive, the de facto benchmark for Time Series Classification (TSC), the same recordings ask a different question: "which of 52 pigs produced this trace?". The dataset PigAirwayPressure is not an outlier. Although the archive is the standard benchmark, whether its tasks resemble the problems TSC is meant to solve has rarely been examined. We find that several datasets are repurposed from their original task, built from toy setups, or have flawed test splits. We further question how representative the archive's ordered non-temporal sequences (e.g. spectra, shape outlines) are of real TSC, and the potential downsides from the decomposition of multivariate problems. Our findings call for more careful interpretation of archive-wide scores and greater attention to deployment-relevant evaluation. The audit is at https://github.com/automl/ucr-provenance-audit.
Reinforcement learning (RL) rewards are notoriously difficult to design and control, often leading to the model learning unintended behaviors such as reward hacking. One potential solution is to monitor for reward hacking and penalize it when detected; however, training against a monitor could lead to evasive behavior, and our general understanding of how to apply monitors effectively during training is limited. To study how best to use monitors to mitigate reward hacking, we introduce and open source three realistic environments where Qwen3-4B reward hacks: a coding environment hackable via test overwriting, a medical chat environment hackable via sycophancy, and a biography generation environment hackable via hallucination. We first focus on the coding environment, where we find that: (1) models can learn to evade highly accurate monitors by exploiting systemic flaws in probes and LLM judges; (2) monitors that leak more learning signal during RL suppress reward hacking but are more often evaded; and (3) including easier problems in training can decrease reward hacking. We apply our findings to build better reward hacking monitors for the medical chat and biography generation environments that improve upon naive baselines to reduce reward hacking rates across seeds to before-training levels in some interventions. Our results demonstrate that our takeaways translate to new settings and that better monitor intervention designs are possible.
Machine learning (ML) methods have been explored as a means of optimizing lattice field theory calculations, with increasing interest in recent years. This chapter gives an introduction and summary to the use of ML methods to optimize lattice field theory without introducing any uncontrolled approximations. We focus on those directions which are clearly established as of late 2025, marking the 50th anniversary of the beginning of the study of lattice gauge theory. This chapter is a contribution to the online book ``Lattice QCD at 50 years,'' (LQCD@50), edited by Tanmoy Bhattacharya, Maarten Golterman, Rajan Gupta, Laurent Lellouch, and Steve Sharpe.
3D spatial reasoning from video remains challenging for VLMs because videos provide sequential, view-local observations rather than persistent metric scene records. Many questions must be answered in a query-specified perspective rather than the observed camera viewpoint, and prompting with a fixed set of RGB views may miss sparse or occluded visual cues. To address these gaps, we introduce BEV3SR, a training-free and backbone-agnostic framework for RGB-D videos with camera parameters. BEV3SR turns video observations into queryable metric bird's-eye-view (BEV) scene memory and formalizes each query into an evidence-aware spatial task. The Spatial Task Formalizer specifies the query's reference frame, spatial constraints, objective, and evidence requirement. The evidence requirement determines the appropriate reasoning pathway. A query treated as requiring geometric evidence is handled by the Geometric Reasoner, which deterministically computes the answer using the metric BEV records in the query-specified reference frame. Otherwise, a query treated as requiring visual evidence is passed to the Visual Evidence Aligner, which uses BEV-indexed object records and their associated supporting views to retrieve object-consistent and scene-context RGB frames for VLM grounding. Within a standard 16-frame budget, BEV3SR consistently improves different VLM backbones across multiple 3D scene QA benchmarks, delivering large gains on spatially intensive questions and significant improvements on open-ended scene QA. These results demonstrate that BEV3SR offers an effective, efficient, and interpretable framework for explicit 3D spatial reasoning from video.
Uncertainty bands on wind forecasts help inform reserve sizing and commitments, but often fail to deliver their stated coverage. We apply conformal prediction to native bands to see what it adds and what it costs, comparing six forecasters, from ARIMA to a tabular foundation model, at two wind farms in Great Britain (GB) under operational conditions. We show that calibration repairs empirical marginal coverage with little or no Winkler Score penalty, but this apparent success conceals under-coverage in high winds. Calibrating by forecast wind regime turns marginal coverage into regime-level coverage, cutting day-ahead out-of-band volume in high winds by over 40% relative to pooled calibration while narrowing bands overall.
A conversational agent that remembers has to decide what to keep. Retrieval-augmented generation keeps the transcript and searches it; the alternative is to compact each turn into atomic facts and search those. We compare the two on LoCoMo, with both arms answering all 1,985 questions in one window, against the same stores, with the same generator and judge, and both predictions on one row, which licenses McNemar's exact test on the 512 questions where they disagree. On the four categories comparable work reports, compaction scores 71.4 per cent against 64.6, a lead of 6.8 points (p = 5.5e-08) at 34 per cent fewer prompt tokens. On all five, including the adversarial set comparable work drops, the lead falls to 2.0 and stops being significant, because the baseline leads adversarial by 14.6. That loss has a structure. Our lead falls as the reference answer demands more source wording, 9.6 points ahead at one or two reference words and 3.1 behind past eight, with token overlap moving the same way, so the judge is not the cause. Compaction keeps what a fact asserts and discards how it was said, so it wins where the answer is a value and loses where it is a phrasing. We also report four negative results with their power calculations.
Graph similarity computation is an important task in graph mining with many real-world applications, such as chemical compound analysis, program dependence comparison, and social network analysis. The problem is computationally challenging: Graph Edit Distance (GED) and Maximum Common Subgraph (MCS), which quantify the distance between graphs and identify their largest shared subgraph, respectively, are NP-hard to compute exactly. To address these computational challenges, Graph Neural Networks (GNNs) have been applied to this problem. However, existing GNN-based approaches have primarily focused on small-scale graph datasets, while scalability to larger scales remains underexplored. In this work, we propose FiLSim, a scalable GNN-based framework, and investigate whether a simplified GNN pipeline can maintain competitive performance on standard benchmarks while scaling to large graphs. We additionally propose a sequential graph-editing strategy to construct large-scale graph datasets with reliable GED labels to evaluate scalability. The proposed framework is evaluated on four benchmark datasets and four newly created large-scale datasets, demonstrating its effectiveness and scalability. FiLSim remains accurate on graphs with more than 150{,}000 nodes, where the baseline methods run out of memory or fail to learn, and scores all test pairs within seconds on a single GPU.
Editing a video stream in real time, without access to future frames, is essential for interactive applications. Existing training-free video editing methods, however, assume full offline access to the entire sequence, while streaming approaches either rely on additional training or are limited to specific editing techniques. We show that this limitation can be removed by analyzing the self-attention dynamics of DMD-distilled autoregressive (AR) video diffusion models. We uncover a key asymmetry: queries and keys remain stable and aligned with their source counterparts across denoising timesteps, while values diverge and encode the evolving content. Inherited from the non-distilled bidirectional teacher, this asymmetry is strongest at high noise, where few-step streaming models operate. The resulting mechanism is simple: a single forward pass suffices to capture source identity in the value features, which can be reused throughout generation. Starting from pure Gaussian noise and injecting only these source values reconstructs the input with fidelity matching or surpassing inversion-based methods, effectively eliminating inversion. Based on this insight, we introduce AR-Edit, a training-free, inversion-free method that caches source values once and selectively injects them during denoising. Across multiple AR backbones, AR-Edit enables real-time streaming editing with superior source preservation and tonal fidelity at minimal overhead.
External lexical resources can help large language models (LLMs) translate texts in low-resource languages and specialized domains, but these resources may conflict with models' prior lexical knowledge. To isolate this conflict, we construct a controlled Mandarin-to-English translation task in which familiar source words are assigned new meanings within and across semantic categories. We treat these prompt-supplied meanings as contextual knowledge (CK) and systematically evaluate the effects of word frequency, sentence context, and prompt framing on the translated outputs. Across nine models, CK reliance is greater for less frequent source words and in sentence contexts that support the supplied meanings. Prompt framing can reduce CK reliance, but seven of nine models favor CK even when explicitly instructed to ignore the supplied meanings. Within model families, larger models are more likely than smaller models to utilize CK in their outputs. Three follow-up experiments further show that modal weakening lowers CK reliance, the effect of repetition depends on prompt layout, and the effect of reasoning varies across models. Overall, our controlled study reveals how CK reliance varies across conditions and models, motivating new methods to detect and evaluate lexical conflicts and guide models' lexical choices in LLM-based translation.
We study planning for finite-horizon partially observable Markov decision processes (POMDPs) with finite state, action, and observation spaces. Both exact and approximate planning are known to be computationally hard in this setting, so prior work has explored natural assumptions that make planning easier. Recent work [Golowich et al., 2022] gave quasipolynomial-time approximate planning algorithms for observable POMDPs. These are POMDPs whose observation channels satisfy an $\ell_1$-observability condition, but notably, no assumptions are required on the transition dynamics. In a similar vein, for every fixed $\rho>0$, we give quasipolynomial-time approximate planning algorithms for $\rho$-decodable POMDPs. We show that decodability is implied by both $\ell_1$- and $\ell_2$-observability and is strictly weaker than observability. Moreover, we show that the optimal decodability constant can be computed in polynomial time, whereas computing the optimal $\ell_1$-observability constant is NP-hard in the worst case. We also show that the fast filter-forgetting guarantee underlying the approach of [Golowich et al., 2022] does not extend directly from $\ell_1$-observability to constant $\ell_2$-observability. Thus, decodability is an easy-to-check, unifying assumption under which quasipolynomial-time approximate planning for POMDPs is possible.
Multimodal Large Language Models (MLLMs) require continuous factual updates to stay current, yet lifelong knowledge editing remains a daunting challenge. The primary obstacle is the severe Locality Erosion and Catastrophic Forgetting, where sequential edits inevitably interfere with pre-trained capabilities and prior updates. In this paper, we propose \textbf{ROSE}, a \textbf{R}isk-aware \textbf{O}rthogonal \textbf{S}ubspace navigation framework for lifelong knowledge \textbf{E}diting. ROSE addresses forgetting by aiming to restrict parameter updates to the orthogonal complement of previously edited subspaces. Crucially, it introduces a risk-guided mechanism that utilizes a dynamically computed risk mask to safeguard specific factual parameter subspaces essential for model stability. During inference, ROSE employs a multi-granularity knowledge integration strategy, featuring prototype-anchored semantic gating and perturbation-driven subspace fusion to precisely activate edited knowledge for relevant queries while limiting interference on unrelated inputs. Comprehensive experiments on two prominent multimodal benchmarks show that ROSE is competitive with state-of-the-art approaches across five key metrics, while maintaining outstanding stability over $1,000$ sequential updates. Our code is available at https://github.com/sunlitsong/ROSE.
Animation is pervasive on the Web, the CSS animation property alone appears on over three-quarters of webpages. Yet existing webpage code generation benchmarks target static screenshots, user-triggered interactions, or isolated motion graphics, leaving the ability of multimodal large language models (MLLMs) and coding agents to reconstruct complete animated webpages largely unexplored. To fill this gap, we introduce AnimWeb-Bench, the first benchmark for animated webpage code generation from videos. It comprises 124 real-world webpages spanning 7 application domains and 33 animation types, with 91.9% of pages combining multiple types, each paired with a reference video and the static assets of the original webpage. We further propose a human-aligned, two-dimensional evaluation protocol: Static Appearance Fidelity measures structure, text, color, and visual similarity on aligned keyframes, while Animation Fidelity tracks UI components across frames and compares their position, size, velocity, and appearance with those in the reference video. Evaluating 14 MLLMs and three coding agents reveals: (1) reproducing webpage animations is substantially more difficult than reproducing static appearance; (2) explicit frame inspection and render feedback improve the reconstruction quality of coding agents; and (3) rapid, continuous changes are harder to reproduce than discrete state transitions. Our code and dataset are available at https://starry1001.github.io/AnimWeb-Bench-website/ .
Large language model (LLM) agents coordinate through conversation, and the same message can be phrased in simpler or more complex language, bluntly or politely. We test whether such differences change how other agents judge and act, using the LURK Werewolf framework, which separates each utterance's speech-act predicate from its natural-language realization. We rewrite only the realization, check that the game content is preserved, and replay each game day from saved checkpoints against a re-sampled control. We vary linguistic range consistently across a speaker's messages and politeness strategy per utterance. Across 30 games and 4,335 replays, neither manipulation changes the trust other agents place in the speaker, and no effect on votes or night actions survives correction for multiple comparisons; the largest uncorrected effect is that shifting range away from a neutral paraphrase changes about 4 percentage points more of the other players' votes. In natural play, 88% of face-threatening speech is bald on-record regardless of role.
Probably Approximately Correct (PAC) bounds on the generalization error (the difference between the empirical prediction error computed on the training data and its expectation) are an established tool in statistical machine learning, and there have been several extensions of these bounds to dynamical systems. In parallel, in recent years, there has been growing interest in finite-sample bounds for parameter estimation error in system identification. The latter bounds tend to be algorithm-specific, relying on reformulating the identification algorithm as a least-squares regression problem. This is in contrast to classical PAC bounds, which are algorithm-agnostic and can be applied to any identification algorithm based on minimizing a cost function involving the empirical loss. In this paper, we show how oracle bounds derived from PAC bounds can be used to derive finite-sample bounds for dynamical systems. As an illustration of this approach, we present finite-sample parameter-estimation bounds for linear and bilinear systems with bounded inputs and noise. These bounds apply to identification algorithms based on minimizing a regularized empirical loss function. The key ingredient is to establish an oracle inequality, i.e., an inequality between the true prediction error of the learned model and the minimum achievable true prediction error, and then use either the convexity of the true loss or system-theoretic properties of the loss function to derive bounds on the parameter estimation error. We also analyze the rates of the obtained bounds and compare them with those of classical asymptotic results.