Papers

Radiology report generation models can produce fluent text while still containing finding-level inaccuracies. Diagnosis-driven methods improve generation by conditioning on predicted findings, but these predictions do not directly modify the visual patch features provided to the decoder, and global gating applies the same modulation across spatial locations. We introduce a Position-Aware Gate (PAG) that uses predicted finding representations to modulate visual patches spatially without region supervision. We also propose a Decoder-level Supervised Contrastive Loss (DSCL) that structures decoder representations using shared positive findings rather than instance identity. On MIMIC-CXR, PAG+DSCL improves clinical efficacy (CE) F1 from 0.484 to 0.502 over a matched global-gate reference, while PAG and DSCL individually reach 0.491 and 0.495. Without additional fine-tuning, the combined model achieves 0.226 CE F1 on IU X-Ray, compared with 0.211 reported by PromptMRG.

Recent studies on reinforcement learning (RL) report seemingly conflicting evidence about large language model (LLM) reasoning. Training on mathematics can improve performance in other domains, yet gains in Pass@1 can coincide with lower Pass@$N$ than the base model. This raises a fundamental question: does RL expand an LLM's reasoning boundary, or merely reweight its existing reasoning space? We revisit these phenomena across Qwen and Gemma model families, showing both cross-domain gains and forgetting, while coverage at large sampling budgets increases on some tasks and decreases on others. Detailed analysis of solution traces before and after RL indicates a shift in the reasoning strategies the model employs, motivating a two-stage autoregressive policy model that separates strategy selection from problem-specific execution. Within this framework, we prove how RL's implicit bias reshapes strategy preferences, allowing gains on some tasks while suppressing strategies required by others. This mechanism can also broaden or narrow coverage at a given sampling budget even without expanding strategy support. We further provide theoretical justifications for log-sigmoid and log-linear scaling laws in RL compute, and evaluate their predictive power. Together, these results connect changes in strategy selection to cross-domain transfer, reasoning coverage, and compute scaling.

Graph Neural Networks (GNNs) often struggle to capture long-range dependencies due to over-squashing – a phenomenon in which the repeated compression of node embeddings into finite-size messages causes representations to collapse. Over-squashing is most often diagnosed as a property of the graph topology, with effective resistance serving as a principled measure of the bottleneck. We provide a complementary view on the matter: building on cellular sheaves, we introduce sheaf effective resistance, a generalization of effective resistance that depends on the sheaf attached to the graph, and we prove that for flat vector bundles, the over-squashing sensitivity in the Jacobian sense is upper bounded by a quantity related to the sheaf effective resistance between the nodes. The bottleneck thus need not lie in the graph itself: it can be relocated, and reduced, by adjusting the sheaf. We instantiate this idea in FlatNSD, a simple message-passing variant of Neural Sheaf Diffusion, and show that it implicitly learns to modulate total sheaf effective resistance, performing well on benchmarks designed to stress over-squashing without altering the original graph topology.

Conversation success is important for communication, social bonds, and well-being and is often reduced in autistic-neurotypical conversations. Previous studies have predicted perceived conversation success using acoustic and lexical features in pediatric autism-enriched samples, or using nonverbal behavior in non-autistic group interactions. Other work has characterized prosodic, facial, motor, and turn-taking synchrony in autistic interactions, but has primarily targeted diagnostic classification or group differences. The present study instead predicts participant perceived conversation success from complementary articulatory, facial, acoustic-prosodic, and turn-taking measures across three spontaneous dyadic-interaction datasets including autistic and non autistic individuals. By leveraging multimodal features, we demonstrate that conversational success is inherently multidimensional and requires a multi-faceted analytical framework. Combining these complementary modalities offers the potential to improve prediction of perceived conversation success by capturing information unavailable from any single modality.

Structured generation is typically formulated as a deterministic mapping from input to a single structured output. This assumption is fundamentally misaligned with natural language, which often admits multiple structurally distinct yet semantically valid interpretations. As a result, existing methods rely on single-reference supervision that collapses ambiguity into one prediction, leading to unstable and inconsistent outputs under underspecified inputs. We reformulate structured generation as an ambiguity-aware inference problem, where the objective is to reason over a set of plausible structured outputs rather than predict a single solution. We show that standard training inherently suppresses valid alternatives, and instead adopt a multi-valid learning framework that preserves ambiguity during optimization. To operationalize this formulation, we construct a candidate space of structured outputs, represent each candidate as a graph capturing entities and their relationships, and perform selection via graph-based reasoning over semantic alignment and relational consistency. This enables principled comparison across competing interpretations beyond token-level likelihoods. Experiments on structured generation benchmarks demonstrate consistent improvements in accuracy, stability, and robustness under varying levels of ambiguity. These results highlight the importance of modeling ambiguity explicitly and suggest a shift from single-output prediction to set-level reasoning in structured generation.

Our position shifts the target of peer-review reform from reducing uncertainty through evaluator improvement or judgment aggregation to constraining evaluative authority under persistent uncertainty. Scientific peer review aims to assure research quality but also allocates scarce academic opportunities, such as publication in high-impact venues. However, the review system is under growing pressure as submissions increase while reviewer capacity, expertise, scrutiny, and incentives cannot scale accordingly. As peer review operates under persistent evaluative uncertainty, evaluator improvement and judgment aggregation remain structurally limited. We therefore propose authority-constrained peer review: before a consequential judgment is released as decision-relevant, it should withstand structured scrutiny through critique-authority decoupling, independent secondary evaluation, contestability and reversibility, and accountability for unsupported influence. Our aim is to prevent weakly supported judgments from entering final deliberation with the same authority as judgments that have withstood structured scrutiny, while creating incentives for reviewers to justify consequential claims more carefully. For large-scale computer-science conferences, this position offers an institutional response to uneven review quality and reviewer-assignment dependence; more broadly, it provides a governance principle for any domain where uncertain expert judgments allocate scarce academic opportunities.

Self-Supervised Learning (SSL) has achieved impressive success in learning semantic visual representations, yet the underlying principles driving this success remain underexplored. In this work, we hypothesize that common SSL pretext tasks implicitly model object co-occurrence statistics, a fundamental cue for visual learning. To test this, we curate three datasets of segmented objects from existing vision benchmarks using a state-of-the-art segmentation model. Through experiments across many SSL models, we reveal a hierarchical encoding of semantic information: while the visual backbone captures object categories, the projection head specializes in encoding co-occurrences between object categories. As a result, the projection head outperforms the backbone in aligning with human judgments of inter-category similarity. Furthermore, by controlling co-occurrence patterns during pre-training, we demonstrate that encoding object co-occurrences can significantly accelerate the emergence of category-level representations. Our findings uncover a previously hidden learning principle in SSL and suggest a path toward designing more effective pretext tasks by explicitly leveraging object co-occurrence structure.

When a fact becomes harder to elicit from a language model, has the model lost it, or has it merely become less accessible? Parameter-updating test-time continual learning agents are usually evaluated by current output accuracy; we ask whether that measurement captures how prior experience shapes later updates. We test whether learning a specific factual binding produces a later relearning advantage beyond familiarity with its entity, answer vocabulary, format, or query slot. In Qwen2.5-1.5B, we compare three matched, counterbalanced histories: LEARNED receives the correct binding, CONTROL receives matched exposure without it, and WRONG receives a competing relation-valid answer in the same slot. After an identical rehearsal-free continual fine-tuning stream, LEARNED reacquires the target faster than CONTROL at a prespecified checkpoint of partial forgetting. WRONG shows no positive advantage over CONTROL, while its comparison with LEARNED provides slot-matched corroboration, subject to possible competitor interference. The advantage remains positive after adjustment for measured pre-relearning behavior. In a secondary descriptive conversion using an unpaired never-seen pool, prior learning reduces the optimizer updates required for restoration by about 18%. Current behavioral accuracy can therefore underestimate the lasting effect of previously acquired experience. Code is available at https://github.com/zhezhou1106/relearning-savings

Continuous diffusion treats the forward process as a design space; discrete language diffusion often uses absorbing corruption, leaving context fully trusted throughout generation. This makes trust binary: corrupted context is copied as ground truth, propagating errors into generated spans. We reopen this design space by combining absorbing and uniform-mixing kernels with region-dependent rates. Missing positions follow absorbing dynamics, while context undergoes low-rate random substitution, exposing the model to observed but uncertain tokens during training. At inference, a heuristic sampler re-injects noise into suspicious context under an annealed intensity and freezes context updates late in the trajectory. An 8,000-step finetune of a pretrained 170M-parameter masked diffusion model takes under five hours on a 6 GB laptop GPU. The base model assigns roughly 50% probability to observed tokens even when they are uniform-random garbage. For uniform-random corruption, finetuning detects corrupted context with 0.96 recall and 0.91 precision, improves infill token-F1 by 3.8 points at 10% corruption and 6.4 points at 20%, and removes the accuracy dip next to corruptions. Clean conditioning shows no measurable regression, with a false-edit rate below 0.4%. Gains for plausible corruptions are weaker: about 0.5 F1 points at 10% corruption, without statistical significance. Our results suggest that detectability and damage depend jointly on departure from the data conditional: nearby noise is hard to detect and semantically mild, while distant noise is conspicuous and destructive. We interpret absorbing diffusion as the infinite-contrast limit of a trust axis, analogous to the signal-to-noise gap between known and unknown regions in diffusion inpainting. In our setting of a 170M model and a single text domain, intermediate trust is reached by changing forward dynamics rather than architecture.

Global streamflow forecasting supports flood preparedness, yet data-driven models trained on reanalysis lose skill when driven by operational forecasts, increasingly so with lead time. AIFL addresses this shift with a two-stage transfer-learning recipe and a deterministic LSTM, but it feeds one forecast lead time into the whole input window and gives no uncertainty. We ask how far this open recipe can be extended without changing its recurrent core. KNOU-Flood keeps a single-layer LSTM and the two-stage training and changes the information around it: a longer input window (424 days), lead-aware forecast inputs, GraphCast as a second forecast source next to ECMWF HRES, and a latent-noise ensemble trained with the fair CRPS. We compare against a re-implementation of the AIFL recipe (its code is not public) under identical data, stations, splits and metrics. Our contributions are threefold. First, over three seeds and 2,099 test basins (2021-2023), KNOU-Flood reaches a median KGE' of 0.65 at LT1 and 0.48 at LT10 against 0.62 and 0.34, with significant gains at all ten lead times under two scoring protocols, also against a 1024-unit baseline and on held-out basins (one seed each). Second, identical weights give an LT10 KGE' of 0.18 or 0.34 depending on how forecasts are fed, so much of the lead-time decay reported for AIFL is an input-protocol effect; the probabilistic head contributes most to the gain, followed by the raw GraphCast channels, the longer window and mixed fine-tuning. Third, the ensembles reach lower CRPS than probabilistic versions of the baseline, are nearly calibrated with inference-time dropout, and detect 1.5- and 2-year floods better than all baselines; the benefit for rarer floods depends on the seed, and an exploratory tail-weighted variant improves 10-year and larger events. KNOU-Flood is a controlled extension of an open recipe, not a new architecture.

Reinforcement learning with verifiable rewards (RLVR) is credited with gains beyond the domain its rewards check, such as higher MMLU accuracy after training on mathematics. Such accuracies are usually read out of the model's text by an answer extractor. A model that learns to write its answer where the extractor looks can therefore score higher without answering better. We measure how much of the gain reported by extractors copied from public code arises this way. We generate outputs from a base model and its RL model and score them twice: with those extractors, and with one extractor of our own that reads a stated answer whether it is in the answer box \boxed{} or in the text. On four released 7B SimpleRL base-RL pairs, 82% of the out-of-domain multiple-choice gain that these extractors report disappears under ours: it is a gain in answer format, not in answers. The base models already state most of their answers in text, and RL moves them into the box, where these extractors look first. Independent LLM readers, in place of our extractor, put the figure at 77%. With a prompt that itself asks for the box, our extractor puts it at 79%. On a 32B pair trained with another recipe, almost none of the reported gain is format gain: its base model boxes its answers when asked, and its RL model rarely does when not. A reported gain is partly a property of the extractor that read it, and re-scoring stored outputs with a second extractor shows how much, without re-running RL.

Research on deceptive alignment is becoming increasingly urgent as frontier-model capabilities advance and their risk of causing catastrophic harm increases. Alignment faking is a form of deceptive alignment, with the compliance gap serving as a key behavioural indicator. This behaviour remains understudied in small LLMs and prior mechanistic interventions have not been shown to mitigate the compliance gap. We test whether steering SAE (sparse autoencoders) features associated with deontological ethics can mitigate this behaviour. We use a novel bag-of-words feature elicitation pipeline for SAE steering to identify what we construe as concept-level steering directions. Our approach reduced the compliance gap to non-significance in Gemma-3-27B-IT, whereas the tested configurations were not effective in LLaMA-3-8B-Instruct. Analysis of scratchpad traces also revealed deceptive model behaviour which we term deceptive defanging: deliberately degrading the aligned functionality of a response (in this case harmfulness) while preserving the appearance of alignment (in this case helpful compliance with the harmful request), sometimes conditioned on tier, training, or RLHF consequences. These findings provide, to our knowledge, the first evidence that a mechanistic activation-level intervention can mitigate the compliance gap in an alignment faking setup. They also identify a form of deceptive behaviour that may broaden our understanding of the different forms of deceptive behaviour in LLMs.

We expand the exponential kernel of attention into a truncated Taylor series of $n$ terms and factorize each term into a tensor-power inner product, which collapses the causal history into exactly $n$ recurrent states, each updated by one rank-one outer product per token and independent of context length $N$. The recursion is lossless with respect to the truncated kernel ($2.2 \times 10^{-16}$ in float64), and the states are additive across chunks ($5.7 \times 10^{-14}$), so prefill admits a parallel scan. Measured exponents are $t \propto N^{2.28}$ for softmax against $N^{0.91}$ for states, with a crossover at $N \approx 728$ and $20.8\times$ at $N = 8192$; per-token decoding latency is constant at $7.3\,\mu$s against $N^{1.04}$ for softmax, giving $76.0\times$ at $N = 65536$. Accuracy improves exponentially in $n$ while cost grows combinatorially, since the $m$-th state is $\binom{d+m-1}{m}$ wide—a bound confirmed by SVD—so the optimum is second order, $n = 3$, at $6.5 \times 10^{-4}$ error and $9{,}945$ elements per head. Query and key normalization is a prerequisite: without it the argument exceeds the convergence radius and adding terms stops helping.

A longstanding goal for robot reinforcement learning (RL) is that robots learn from their own physical experience, as humans and animals do. A key obstacle to learning directly from physical experience, especially outside instrumented lab environments, is that the state of the world is usually unobservable to the learner and must be estimated from on-board sensing. Errors in that estimate corrupt both the policy input and potentially the reward signal. We study this challenge using the contact-rich task of a bipedal robot learning to score soccer goals. We estimate all key state variables on-board with an unscented Kalman filter (UKF) and no external instrumentation (e.g., motion capture or overhead cameras). Because the UKF may incorrectly estimate state when the ball is unobserved, we learn a new state representation with a recurrent state-space model (RSSM) that, critically, only corrects the RSSM's latent state when the ball is observed. Building on RLPD, we introduce RSSM-RLPD and apply it to the task of a NAO robot learning to dribble a soccer ball into a goal. We show that the robot learns to score from multiple starting positions after 30 minutes of real-world interaction, reaching 100% success against 0% for a feed-forward policy under the same training budget. We complement our physical robot experiments with a simulation study and show that RSSM-RLPD learns policies that achieve a higher success rate than behavior cloning, a feed-forward, and GRU baseline. Taken together, our results show that the addition of a learned state representation to existing on-robot learning methods enables RL from real-robot data even with a high degree of partial observability.

Inverse Reinforcement Learning (IRL) is classically solved using Lagrangian optimisation, by jointly optimising a primal-dual problem with gradient descent to obtain the optimal reward function parameters and corresponding optimal policy. While this algorithmic view of IRL has been highly influential, gradient-based reward updates only paint a partial picture, largely ignoring the geometry of the problem in occupancy space. In this paper, we present an occupancy-space view of reward updates in IRL. We leverage the insight that both the optimal solution and its geometry in occupancy space are known, and show that the geometry along the e-geodesic (straight line in occupancy space) toward the optimal occupancy directly affects the gradient of the IRL dual problem in reward parameter space. This perspective naturally suggests preconditioning the dual gradient with the average Fisher information along the occupancy trajectory. We show that this preconditioned gradient is just a convex interpolation of the old reward and a new target reward, yielding efficient natural-gradient-style updates with little additional computational overhead. We then analyse these occupancy-space reward updates from a Mirror Descent (MD) perspective, and show that policy search over the resulting rewards induces approximate entropic MD on the policy. Finally, we conclude with convergence results of such geometry-aware IRL algorithms.

We discuss a new symmetric-key cipher for digital data encryption. Its implementation is fast, memory efficient, and resilient against classical, AI-assisted, and quantum attacks. Let $x$, $k$, and $c$ be elements of a finite abelian group $G$ with operation $+$ and neutral element $0$. Suppose one is given ciphertext $c=x+k$. Retrieving the plaintext $x=c-k$ from the ciphertext is trivial when the key $k$ is known. Without the key, however, recovering $x$ amounts to a blind search. Even if an efficient criterion were available for identifying the plaintext among all group elements, the average number of trials would remain prohibitively large when the group is sufficiently large. The challenge in constructing a practical cryptosystem is therefore to identify a sufficiently rich group $G$, together with efficient algorithms that enable an immersion of real digital data into $G$ and efficient evaluation of the group operations. The proposed cipher is based on generalized Dirichlet convolution over finite fields. We investigate its security against plaintext attacks from an information-theoretic perspective. The nonlocal structure of Dirichlet convolution naturally permits plaintext salting, providing protection against plaintext attacks while allowing controlled key reuse. We establish perfect masking properties over finite fields and prove that the resulting cipher achieves information-theoretic security under key reuse when the underlying arithmetic is carried out in $\mathbb{F}_p$ with $p>2$. For arithmetic over $\mathbb{F}_2$, some information leakage remains, leading to a finite key-reuse budget.

Existing vision-language-action (VLA) models benefit from large-scale vision-language pretraining, but tend to memorize training trajectories, limiting their ability to adapt learned interactions to new task and scene configurations. To address this problem, we propose Region-Aware Interaction Networks (RAIN), which formulates a task as a sequence of goal-conditioned subtasks, each specified by an action type and a target-region mask. This formulation allows learned skills to be reused across different objects and scene layouts and composed into task sequences not seen during training. Its Target-adaptive Cross-View Encoder (TCE) conditions multi-view features on target masks and exchanges information across views, while PlanDiT predicts target-directed trajectories and actions. A Transition Head estimates subtask completion to enable automatic transitions. We also introduce a Reference-based Retargeting Strategy (RRS), which uses recorded trajectories to synthesize smooth approaches to reference interactions with alternative target regions, without collecting additional data. Experiments on LIBERO-Analogy, our new benchmark for robotic task generalization, and real-world tabletop tasks demonstrate that RAIN generalizes more robustly than existing VLA baselines.

Embodied Artificial Intelligence (AI) introduces cyber-physical risks because digital failures and attacks can cause physical harm. This survey organizes embodied AI safety through a four-layer reference architecture – comprising the physical interaction, system and middleware, algorithm and decision, and cloud-edge and data layers – and uses it to connect vulnerabilities, propagation paths, and safeguards across the system stack. It traces the evolution of safety paradigms and examines how faults and attacks cross architectural boundaries to produce physical consequences. Furthermore, the survey systematizes composite attack chains, outlines multi-layered defense strategies, and reviews representative embodied safety benchmarks. Ultimately, we advocate for unifying software alignment with rigorous hardware constraints to build trustworthy systems capable of operating safely within human-centric environments.

Dynamic-point filters are routinely added to feature-based visual SLAM, and several recent systems argue that removing dynamic features can leave too few static features in low-texture regions. So far, these systems have been evaluated only on texture-rich benchmark sequences. We present a controlled study that isolates this interaction. We render synthetic indoor sequences in which surface texture (four levels, quantified by FAST-corner density and image-gradient entropy) and scene dynamics (three levels) are varied factorially along identical camera trajectories, with stereo, RGB-D, ground-truth poses and dynamic masks. On this grid we compare ORB-SLAM2 without filtering, with an optical-flow and epipolar-residual filter (FLOW), and with a multi-view depth-consistency filter (GEOM), and report trajectory error, tracking completeness and surviving static features over five runs. Because the masks give per-keypoint ground truth, we also measure each filter's dynamic-point precision and recall and its static-feature false-removal rate, so that mechanistic explanations can be tested directly. We do not propose a new filter. On 720 runs over 24 sequences, filtering helped mainly in the most dynamic cells; the benefit did not decline monotonically with texture, but at the lowest level filtering reduced tracking completeness, and ORB-SLAM2 never initialised in static L3 scenes. Contrary to our hypothesis, GEOM discarded more static keypoints than FLOW (median FRR 6.8% vs. 1.5% for RGB-D, 19.6% vs. 1.5% for stereo); its RGB-D advantage tracked dynamic-point recall and vanished in stereo mode. Data and code are available at https://github.com/felixxxue/texture-dynamics-slam.

Purpose: Current abdominal ultrasound (US) simulation methods often require CT-based anatomical references for ray-casting, limiting deformation and pathology variability. We propose a learning-based pipeline trained to predict physics-based images derived from CT scans from semantic labels, enabling controlled simulations without patient-specific CT volumes at inference time. Methods: We introduce a two-stage pipeline that maps anatomical segmentations to realistic US images through a simplified US image. Stage~I synthesizes this image from semantic labels using models trained on CT-based ray-casting outputs. Stage~II refines it into a realistic US scan using anatomically guided unpaired translation. Deformations and pathologies are generated by editing anatomical maps. Results: We evaluated Pix2Pix and the Semantic Diffusion Model (SDM) in Stage~I, followed by segmentation-guided CycleGAN (SG-CycleGAN) refinement in Stage~II. SDM significantly outperformed Pix2Pix in morphological metrics, including MAE (19.85 vs. 21.65), SSIM (0.28 vs. 0.24), and mIoU (0.43 vs. 0.29), whereas Pix2Pix yielded better perceptual point estimates (LPIPS: 0.17 vs. 0.19; FID: 0.32 vs. 0.37; KID: 0.25 vs. 0.48). Conclusion: Training paired generative models with physics-based supervision enables approximation of CT-derived ray-casting outputs at inference time directly from semantic labels. Although the pipeline does not require patient-specific CT volumes at inference time, CT-derived segmentations and ray-casting simulations remain necessary to train Stage~I. Once trained, the framework enables controllable healthy and pathological simulations through semantic-map modification.

Glioblastoma (GBM), an aggressive primary brain tumor, is routinely monitored with longitudinal MRI after treatment. Distinguishing stable disease (SD), pseudoprogression (PsP), and true progression (TP) remains challenging because these states can show overlapping MRI appearances despite different temporal trajectories. Existing longitudinal methods still face challenges in modeling scan-specific spatial variability, variable follow-up intervals, and complementary information from the observed follow-up state and its longitudinal change. We propose STRIDE, a framework for spatial-temporal representation of interval-conditioned disease evolution that takes paired post-treatment MRI scans and their inter-scan interval as input and predicts SD, PsP, or TP. The lesion-prior-guided spatial representation combines SoftGate and an Adaptive-window Hierarchical Transformer (AWHT) to emphasize lesion-related regions while preserving surrounding context. The time-conditioned latent transition uses pair-level context and the actual inter-scan interval to estimate interval-dependent representation changes between visits. The observed–transition fusion integrates the transition-estimated follow-up representation with the directly observed follow-up representation to jointly characterize the follow-up state and its longitudinal change. BraTS2024 is used to develop and evaluate the lesion-prior generator, while longitudinal pretraining on LUMIERE supports transfer before downstream adaptation to Burdenko. On the Burdenko three-class task, STRIDE achieves a macro ROC–AUC of 0.816 and a macro F1-score of 0.796. These results support its potential for more reliable longitudinal post-treatment GBM state assessment.

Artificial intelligence has advanced individual radiotherapy tasks, yet these capabilities remain separated across clinical stages, software environments and data modalities. This fragmentation contrasts with the longitudinal radiotherapy workflow from treatment decision-making through follow-up. Here we present RadOnc-Agent, an agentic artificial-intelligence framework that formalizes radiotherapy into four clinical phases and provides 26 callable functions through a conversational interface. A large-language-model controller maps clinical intent to schema-constrained calls, preserves patient and workflow context, and routes requests to specialist services. We evaluated system execution using 2,600 single-function requests (7,800 repeat executions), 200 prespecified synthetic cross-stage scenarios spanning four phases (600 executions), and 120 workflow instances from 60 de-identified patient records (360 clean executions) representing decision-to-planning and planning-to-adaptation. RadOnc-Agent selected the intended function in 98.79% of single-function executions, completed 96.50% of scripted cross-stage workflows, and completed 96.67% of real-patient workflow executions. In comparative ablations, removing longitudinal state reduced cross-stage completion from 96.50% to 84.00%, while disabling schema and identity validation increased mismatched backend dispatch from 0% to 95.28% in a replay/test evaluation. These findings establish the technical feasibility of an LLM-orchestrated architecture for coordinating heterogeneous radiotherapy capabilities and information across longitudinal workflows; they do not establish clinical correctness, clinical utility or prospective benefit.

Before deploying runtime recovery for a frozen vision-language-action (VLA) policy, one must establish that an intervention improves success beyond ordinary run-to-run variation and that its complexity adds value over a simple action. We evaluate these questions on frozen $\pi_{0.5}$ across four RoboTwin tasks. For each test seed, we pair rollouts with and without correction and include a same-seed base-policy re-run as a placebo. Seed-cluster intervals and prespecified comparison rules assess net gains against stochastic outcome changes. Across 3,888 paired episodes, the full pipeline raises success on beat_allowbreak block_allowbreak hammer by $+13.5$\,pp (95% interval $[+9.4,+17.7]$), with no detectable gain on the other three tasks at the deployed weight. Among failed base episodes on the responsive task, $43.2\%$ succeed on a plain re-run, compared with $63.5\%$ after correction; many nominal rescues therefore reflect the base policy's own variability. A fixed-time trigger and scripted return to an earlier joint configuration produce a net gain with no detected difference from the learned pipeline across two rounds, although our prespecified equivalence criterion is not met consistently. Pausing and a constant-action control do not yield comparable gains. On this benchmark, the decision to intervene depends strongly on the task, and a paired placebo plus a simple retreat baseline are needed to establish what learned correction contributes.

Automatic music annotation is typically tackled under the assumption of a fixed annotation schema. In practice, commercial music catalogs often need to accommodate new musical attributes as needs evolve. Given that expert music annotation is expensive, it is not evident which methodological approach is most effective at accommodating new attributes and backfilling existing tracks; audio-language models promise zero-shot prediction, but at what annotation budget does supervised adaptation become more compelling? We propose a benchmark based on the MGPHot popular music annotation dataset for simulating music schema extension across different annotation budgets. We investigate zero-shot prediction with audio-language models, learning new attributes from pretrained representations, and adapting models trained on existing annotations. Our results suggest that supervised adaptation is more effective than zero-shot prediction even with small annotation budgets, while frozen representation reuse remains the most effective approach for modest budgets without the tuning required by deeper adaptation.

This paper concerns how semantic context determines geometry in learned vector representations. Similarity is typically measured using cosine similarity, which provides a single fixed geometry. Semantic similarity, however, is inherently context dependent: two images may be similar because they depict the same object, share a visual style, or are relevant to the same clinical finding. We show that contrastive representations naturally encompass a family of geometries that can be specialized to particular semantic structure. The key idea is to use an interplay between contrastive learning, exponential families, and information geometry to establish a correspondence between probability distributions over "anchors" and Bregman geometries on the representation space. We use this correspondence to define "Anchor Divergences", a method for specifying context-specific semantic geometries on fixed representations. Under this correspondence, modeling the anchor distribution models the geometry itself. Experiments on retrieval show that anchor divergences provide an effective and efficient way to specify context-specific semantic similarity.

We study federated reinforcement learning in which multiple agents interact with a common Markov decision process and communicate through a central server to collaboratively learn the optimal state-action value function. Our goal is to understand whether the sample-efficiency benefits of collaboration can be retained when a fraction of the agents behave adversarially and transmit arbitrarily corrupted information. To address this problem, we introduce Robust Async-Fed-Q, an epoch-based federated learning algorithm that combines variance-reduced estimation of the Bellman optimality operator at the agents with robust aggregation at the server. We establish high-probability finite-time guarantees showing that the proposed method preserves the statistical gains of collaboration among the honest agents while tolerating adversarial corruption. In particular, the effect of the adversarial agents decreases as the amount of data collected by each honest agent grows and eventually vanishes in the infinite-sample limit. We complement these guarantees with information-theoretic lower bounds that characterize the unavoidable statistical cost of adversarial corruption, leading to the first nearly matching upper and lower bounds for adversarially robust federated reinforcement learning. We further extend our framework to accommodate single-trajectory Markovian sampling and heterogeneous partial coverage, where different agents may explore different regions of the state-action space and learning relies on their collective coverage. Finally, our epoch-based design substantially improves the best known communication complexity for federated Q-learning under asynchronous sampling.

Prefill-decode disaggregation is becoming a common architecture for LLM serving because it separates two phases with distinct execution patterns and SLO objectives. Existing systems typically combine a fixed prefill/decode worker ratio with request routing across workers. However, real-world workloads exhibit both short bursts and sustained shifts in the prefill-to-decode demand ratio. As a result, a configuration that is well provisioned at one time may quickly become mismatched, causing latency SLO violations even when idle capacity exists elsewhere. Existing autoscaling mechanisms can add capacity, but they react slowly, require spare GPUs, and do not directly address short-timescale phase imbalance. We present FluidPD, a P/D-disaggregated serving system that provides SLO-aware in-place elasticity. FluidPD introduces two complementary mechanisms. FluidToken handles transient imbalance by offloading a bounded portion of prefill computation to decode workers when decode-side slack is available. FluidRole handles sustained imbalance by reassigning running workers between prefill and decode roles in place, avoiding model reload and engine restart. Both mechanisms are guided by lightweight pressure indices that expose prefill and decode-side resource pressure before they appear as SLO violations. Across production Azure trace workloads, FluidPD improves overall SLO attainment over static SGLang by up to 94.6 percentage points, demonstrating that SLO-aware in-place P/D elasticity improves service quality without provisioning additional workers.

We present a DataBrain-specific prototype that turns natural-language analytic requests into inspectable dashboards. An installable Codex plugin and standalone Agent Runtime combine schema-constrained model decisions with typed tools, persistent state, and deterministic Hooks for approval, audit, checkpointing, recovery, and failure handling. The pipeline resolves entities, discovers metadata, enforces read-only SQL, composes dashboards, and applies static checks, dynamic preflight, and browser inspection. The model proposes actions while deterministic software controls execution and records state transitions. We evaluate the workflow on frozen real-DataBrain tasks and controlled Hook faults. Strict success was 6/8 on metadata and SQL tasks: metadata selection passed 4/4, all four SQL tasks met semantic criteria, and 2/4 met the exact output-column contract. The final release passed 4/4 single-panel dashboard tasks, one two-panel task, and one existing-dashboard refinement; a parameterised task exceeded its step limit. All ten fault scenarios met their specified outcomes without unapproved external side effects. Model inference accounted for over 97% of observed runtime in every reported group. These small, DataBrain-specific results do not establish production readiness, general text-to-SQL accuracy, or an efficiency advantage over manual dashboard construction.

Directed Diffusion (DD), uses mechanisms such as data aggregation and in-network processing to suppress the additional data overhead however there is no guarantee that paths from nearby sources join after a few hops. Also for path construction, each source floods a distinct exploratory data (ED) packet through the network which is not necessary for nearby sources. In this paper an efficient data-gathering protocol (EDAP) is proposed to address the problems of late-aggregation and distinct ED-flooding through using a virtual sink (VS) node nearby the sources which plays the role of sink node and broadcasts local interest messages and routed gathered data toward destination. Also in this protocol, multiple paths are constructed between virtual sink and the sink node which leads to load-balancing and increase in the life-time of the network nodes. Our routing protocol can produce more than one disjoint or braided path and spread the gathered data properly between constructed paths according to their length and remaining energy. Simulation results shows that in this way a significant amount of energy can be saved and the network lifetime will be increased considerably (up to 75% better than DD).

Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in web front-end execution, with browser-based game generation emerging as a particularly prominent frontier. While previous efforts frequently rely on complex multi-turn workflows or focus on static game evaluation benchmarks, this work targets direct end-to-end real-world game synthesis driven by coding agents. However, generating complex games directly from sparse user queries often forces coding agents to make underspecified assumptions, yielding incomplete mechanics, disconnected gameplay flows, and limited visual aesthetics. To resolve this issue, this paper presents GameGo, a scalable framework that systematically transforms brief game seeds into comprehensive Product Requirements Documents grounded in industry game-development practices. To retain core gameplay constraints without restricting design exploration, GameGo uses task-specific dynamic compression to maximize information density while preserving instruction following. Based on this pipeline, GameGoData is constructed with 55,060 development trajectories across 2D, 2.5D, and 3D games, alongside GameGoBench, a benchmark comprising 124 diverse game queries. Training GameGoCoder on GameGoData yields a model that outperforms matched baselines and is comparable to frontier models across gamedev benchmarks. All code, datasets, and models will be made publicly available.