Science & Mathematics

Multimodal large language models (MLLMs) have substantially advanced chart understanding and reasoning, yet their performance remains highly uneven across chart types. While modern MLLMs perform well on common charts, their accuracy drops sharply on long-tail charts with diverse visual structures and reasoning requirements. Tool-augmented agents can improve complex chart reasoning by introducing specialized visual capabilities, but manually engineering strategies and tools for the heterogeneous and evolving long tail is difficult to scale. We introduce \textbf{ChartEvo}, a recursive self-improvement framework that enables chart agents to adapt to underperforming chart types through experience-driven harness-level self-improvement. ChartEvo jointly evolves reusable chart-specific Skills and executable Tools, while learning reusable local tool-call chains to compose evolved capabilities and control the inference overhead of tool-augmented reasoning. On long-tail charts from ChartBench and ChartX, ChartEvo improves accuracy over its base MLLM, Qwen3.7-Plus, by \textbf{45.6\%} and \textbf{32.5\%}, respectively, demonstrating the effectiveness of recursive self-improvement for long-tail chart adaptation.

Geometric Relative Positional Encoding has become the primary mechanism for injecting spatial inductive biases into multi-view transformers. Existing designs condition attention on camera-level frustums or ray-marched depth, associating cross-view tokens through the sensor geometry. Such sensor-centric formulations, however, describe only where a token is observed from, leaving the rich surface geometry of the scene region depicted by the token. In this paper, we advocate a shift toward surface-aware positional encoding that explicitly grounds patch tokens in the geometry of physical surfaces. We introduce HRoPE, which represents the local geometry of each patch as a slanted 3D plane, following the slanted support-window assumption of classical multi-view stereo. We inject token-level, plane-induced homographies into attention as relative transformations. This equips the transformer with a geometric prior for rectifying perspective distortions and aligning patch features across viewpoints. Our encoding requires no ground-truth geometry: the plane parameters can be predicted end-to-end under a view-synthesis objective alone, or derived from foundation models. Extensive experiments on RealEstate10K, CO3Dv2, and DL3DV show that our best variant of HRoPE consistently outperforms the evaluated geometric positional encodings for novel view synthesis. Extensive ablations and analysis show the effectiveness of the proposed encoding. Please refer to our project page at: https://gaohchen.github.io/HRoPE

Robot fleets must often agree on a shared value before they act, over wireless networks that change as the robots move. Their supervisor needs a deadline: the number of communication rounds after which the fleet has agreed with a stated probability. Today it comes from simulation plus a safety margin, with an open probability of being met. We certify it. Our method, conformal Koopman certification, rests on one observation: averaged over the random network changes, the expected behaviour of the fleet evolves linearly, for any nonlinearity of the agents. The decay rate of this linear evolution gives the shape of the deadline. One remaining constant is calibrated from 200 short test runs by conformal prediction, a statistical method that turns any predictor into a guarantee that holds with a user-chosen probability. The certified deadline holds for every agreement tolerance at once and for any error in the estimated rate. The same linear model predicts the value the fleet will agree on and certifies how much corruption of the exchanged messages the collective decision absorbs. For 60 simulated robots the certified deadline was met in all 500 test runs against a 90\% target and exceeded the slowest run by 16\%. The classical worst-case bound exceeds that run by more than 160 orders of magnitude, a linearised deadline is missed in up to 26\% of runs, and direct conformal calibration costs 80 times more simulation per tolerance.

Teams of large language model (LLM) agents are increasingly applied to complex tasks, but their collaboration protocols are mostly designed empirically, without convergence guarantees. We formalize collaborative problem solving as block optimization: each agent improves one part of a shared solution, guided by its local objective plus a correction for how its changes affect the other parts. A coordinator checks every proposal against the global objective and merges proposals that do not interfere with each other. We prove linear convergence up to an error floor that vanishes when the corrections are exact. Interfering proposals can also be merged once their combination has been checked; the rate then depends on how well the proposals actually combine, not on worst-case interactions. With GPT-OSS-120B agents on robot routing, a three-agent team finds collision-free plans in every run, better than those of a single agent that controls all robots. On a larger routing task, a six-agent team completes its search about twice as fast as the single agent (in median). Merging the best checked combination of proposals, rather than only proposals that do not interfere, cuts the team's search time by about 30%. With constants measured on the runs, our convergence bound holds in every round and follows the observed progress. We also test the theory on query planning in SQLite and on a synthetic task.

Although automatic detection of cognitive impairment from conversational speech is a promising way, it becomes less accurate, as large real-world datasets often contain unreliable diagnostic labels due to self-referral and limited clinical verification, and here we investigated whether neuropsychologically calibrated diagnostic relabeling improves it. We analysed 138 hours of conversational speech from 720 participants, recalibrating unreliable diagnostic labels using attention, executive function, and memory performance via external clinical thresholds, Gaussian, and normative z-score calibration. Transformer-based classifiers were trained using both original and recalibrated labels and evaluated internally on 150 clinically verified participants and externally on Pitt, ADReSS and PROCESS-2 datasets. Memory- and MoCA-based z-score calibration produced the greatest improvements. Using our internal dataset, $F_1$ increased from $0.70$ to $0.83$ ($+13\%$) and from $0.54$ to $0.66$ ($+12\%$) for binary and multiclass classification, respectively. For external validation, the increases are $+4\%$, $+5\%$ and $+9\%$ on Pitt, ADReSS and PROCESS-2 corpus. Neuropsychologically calibrated relabeling substantially improves speech-based cognitive assessment, highlighting label quality, rather than model architecture, as a critical bottleneck in clinical AI for dementia screening.

Radiologists interpret follow-up chest X-rays by comparing current and prior examinations to identify new, resolved or progressing findings. Most report generation systems either omit this longitudinal context or fuse prior features without distinguishing interval changes from anatomy that remains stable over time. We propose CPVB-RG, a prior-aware framework that combines current and prior radiographs with available clinical text. Its Cross-Prior Visual Bridge (CPVB) contains Progression-Guided Focus Attention (PGFA) and two Cross-Attention units. PGFA is designed to contrast progression-difference attention with shared-spatial attention, aiming to reduce invariant responses whilst retaining coarse spatial correspondence between examinations. The Cross-Attention units then connect the resulting longitudinal visual evidence with clinical semantics. CPVB-RG is trained in two stages, using prior-guided image-text alignment followed by autoregressive report generation with a lightweight language decoder. Experiments on MIMIC-CXR and IU X-Ray show consistent improvements across language and clinical metrics. On MIMIC-CXR, CPVB-RG achieves 0.2847 F1RadGraph and 0.6692 VLScore, with statistically significant gains over the closest decoder-matched prior-aware baseline. These results support the effectiveness of the proposed difference-aware and spatially guided fusion design for prior-aware chest X-ray report generation.

We propose a distributed platoon controller that guarantees safe and smooth crossing of multi-vehicle systems at an unsignalized intersection. To support the stability analysis, we consider a multi-vehicle platoon maneuvering along a closed path with a single unsignalized intersection. Unlike traditional bi-level decoupled designs on open paths, which primarily focus on intersection control while neglecting its coupling with platoon convergence, and thus hindering rigorous convergence analysis, the closed-path setting allows us to analyze the asymptotic behavior of the platoon. The local exponential convergence analysis of the switched closed-loop system, induced by repeated vehicle switching at the intersection, is presented. Under some mild conditions on the controller gains, we show that safe intersection crossing is ensured during the transient phase and smooth crossing is achieved in steady state, where all vehicles travel at a common constant speed while avoiding collisions at the intersection. Finally, the effectiveness of the proposed distributed controller is demonstrated in a high-fidelity CARLA simulator using multiple commercial vehicle models.

Undersampled magnetic resonance imaging (MRI) reconstruction aims to recover high-quality images from partially acquired k-space data, thereby reducing scan time. However, missing frequency-domain measurements often lead to residual aliasing artifacts, blurred edges, and loss of local structural details. In this paper, we propose AGGR-Mamba, a lightweight output-stage residual refinement method built upon a Mamba backbone for undersampled brain MRI reconstruction. The Mamba backbone first produces an initial reconstruction, and the proposed refinement modules then correct the remaining local errors and high-frequency artifacts in a controlled residual manner. Specifically, an Attention Refine Head is introduced to recalibrate local anatomical structures, edge textures, and residual-error regions through channel and spatial attention. In addition, a GDFN-style gated residual refinement branch is designed to selectively preserve informative residual features while suppressing redundant or unstable responses. By combining attention-based recalibration with gated residual selection, AGGR-Mamba improves local structure recovery without substantially disturbing the backbone reconstruction. Experiments are conducted on the IXI brain MRI dataset with T1- and T2-weighted images under acceleration factors of R=4 and R=8. The results show that AGGR-Mamba achieves competitive PSNR and SSIM performance across different reconstruction settings, demonstrating the effectiveness of the proposed output-stage refinement strategy.

Federated learning based radio map estimation (RME) approaches rely on frequent exchange of high-dimensional model updates and often require a central server for collaboration coordination and update aggregation. This incurs prohibitively high communication overhead, rendering them impractical for bandwidth-constrained mobile and wireless edge networks. Moreover, in practical mobile sensing scenarios, sparse and unevenly collected measurements lead to estimation uncertainty, which is critical for guiding RME and collaboration among distributed nodes. To address these challenges, this paper proposes an uncertainty-aware distributed radio map estimation (UA-D-RME) framework for decentralized networks. In UA-D-RME, each node locally infers a probabilistic model from its sparse measurements to estimate both its local radio map and the associated estimation uncertainty. Instead of exchanging raw data or model updates, neighboring nodes communicate only low-dimensional statistical information, including radio map and uncertainty measures over their overlapping monitoring areas. An uncertainty-aware statistical consensus mechanism is then developed to iteratively align probabilistic models in neighborhood by ensuring consistency on estimation confidence. Theoretical analysis establishes the convergence of the proposed consensus mechanism and characterizes the distributed generalization performance of the proposed UA-D-RME method. Experiments on urban radio map datasets verify that UA-D-RME achieves desired reconstruction accuracy comparable to the centralized RME benchmark while reducing communication costs and without exchanging raw measurements.

Bioacoustic foundation models are claimed to generalize across taxa and recording conditions, but are trained mainly on terrestrial archives of focal, high-quality recordings and evaluated on benchmarks emphasizing architecture over data conditions. We evaluate four such models as frozen encoders on an underwater detection task, using linear probing on continuously recorded data from the Kerguelen Plateau. Four datasets of identical size were built to differ only in the signal-to-noise ratio (SNR) of their positive segments, and probes were trained and evaluated across all combinations. Evaluation SNR accounts for more variation in average precision (AP) than model choice, and performance peaks when training and evaluation share a noise regime --- the configuration least representative of a continuous deployment. Training on mixed SNR redistributes performance toward low-SNR conditions without improving it overall, and AP remains low throughout. In light of these results, we call for evaluation standards grounded in realistic noise and deployment conditions rather than curated, high-quality recordings.

Test-time scaling allocates additional inference compute to refining existing solutions or generating new ones. This raises a fundamental question: Should I keep improving this solution, or stop working on it and try a completely new solution? Existing methods make this decision using fixed refinement schedules, self-verification judgments, or learned controllers, which can waste compute or rely on signals that may be incorrect or require additional training. We introduce Adaptive Replay Refinement (ARR), a training-free method that detects when a stationary revision reaches a fixed point and reallocates the released compute to a new independent sample; replay decoding treats the current solution as a draft and verifies unchanged tokens in parallel, enabling efficient fixed-point detection. Across seven models from five families and seven benchmarks, ARR matches majority voting on three of five mathematical reasoning benchmarks at equal compute, while reducing critical-path serial passes by up to 2.4 times. On MATH500, ARR achieves 84.6% accuracy with only 41% of the serial passes required by majority voting. These results demonstrate that detecting refinement exhaustion enables adaptive reallocation from revision to independent sampling, improving inference efficiency while highlighting the limits of self-revision. Code is available at https://github.com/AhsanBilal7/adaptive_refine.

The data samples collected at the Large Hadron Collider at CERN are highly valuable, and the analysis of these data has led to numerous discoveries and the 2025 Breakthrough Prize in Fundamental Physics. Preserving and providing open access to these data for future analysts is imperative to maximize the scientific output of the collected data. The LHCb collaboration has developed a novel approach to this pursuit of open science, the LHCb Ntupling Service: a web service for on-demand production and publishing of custom LHCb open data, providing open access to both Run 1, and for the first time, Run 2 proton-proton collision data collected by the LHCb experiment, amounting to over 4 PB of data to explore.

Riemannian manifolds are powerful models for problems with non-Euclidean structure. Recent work in geometric machine learning has focused on training deep neural networks to learn Riemannian metrics from high-dimensional data. However, many geometric algorithms require repeated evaluations of the exponential and logarithmic maps, whose high computational cost restricts the utility of learned Riemannian manifolds in practice. We propose to address this cost by training a conditional invertible neural network, called a flow chart, to approximate normal coordinates around points on a given Riemannian manifold. With this approach, the exponential and logarithmic maps can be approximated by fast Euclidean operations in the model's learned coordinate space. Our main contribution is an efficient loss function for training flow charts that is nonnegative and, under suitable domain assumptions, is zero precisely when the chart is normal. Across sixteen synthetic and data-derived geometries, flow charts are competitive with a specialized neural geodesic interpolator for logarithmic map estimation, while also being able to estimate the exponential map. When pretrained, flow charts provide a median speedup of 60-70x over numerical baselines at comparable logarithmic map estimation error, and their learned coordinates improve the accuracy of a numerical geodesic solver.

Monitoring the chain-of-thought (CoT) of language models (LMs) is a key strategy for detecting and preventing misbehaviors for AI safety. As such, it is critical to identify reinforcement learning (RL) practices that lead to unmonitorable chain-of-thought. One such practice may be unavoidable: reward misspecification. As RL environments become more complex, it is infeasible to have perfect verification, and models are rewarded for behaviors that are different from the true task intent. We show that reward misspecification alone already degrades monitorability. We create a testbed with pairs of true reward aligned with task intent and misspecified rewards that favor a different behavior, so that reward hacking can be induced by training and labeled verifiably. Across three models and three tasks, CoT monitorability remains low as models' hack rates increase: the CoTs keep reasoning about the task intent while the outputs violate the intent and get rewarded. To mitigate unmonitorability without prior knowledge of the true reward or the hack, we collect rollouts with diverse rewards across training steps and use a LM to infer a reward description. We then inject the description into the input, which teaches the model to verbalize the reward. This reward description training improves monitorability by an average of 24% or 42% absolute, depending on the metric, across all models and tasks, and yields CoTs that reason about the misspecified reward explicitly. Code is available at https://github.com/Lorenayannnnn/reward-misspecification-degrades-cot-monitorability.

The $\mathbb{Z}_p\mathbb{Z}_{p^2}$-additive codes are subgroups of $\mathbb{Z}_p^{α_1}\times\mathbb{Z}_{p^2}^{α_2}$, and they can be seen as linear codes over $\mathbb{Z}_p$ when $α_2=0$, as $\mathbb{Z}_{p^2}$-additive codes when $α_1=0$, and as $\mathbb{Z}_2\mathbb{Z}_4$-additive codes when $p=2$. A $\mathbb{Z}_p\mathbb{Z}_{p^2}$-linear generalized Hadamard (GH) code is a GH code over $\mathbb{Z}_p$ which is the Gray map image of a $\mathbb{Z}_p\mathbb{Z}_{p^2}$-additive code. For every prime $p$, there are recursive constructions of $\mathbb{Z}_p\mathbb{Z}_{p^2}$-linear GH codes $H_p^{t_1,t_2}$ with $α_1\neq0$ and $α_2\neq0$, and of $\mathbb{Z}_{p^s}$-linear GH codes $\bar H_p^{t_1,\dots,t_s}$. We show that the kernel blocks of a nonlinear code $H_p^{t_1,t_2}$ give equivalent punctured codes, and that these punctured codes are linear. Any of them is called the derived code of $H_p^{t_1,t_2}$. We also determine the ranks of the codes $H_p^{t_1,t_2}$ and $\bar H_p^{t_1,t_2}$. For odd $p$, the rank of $H_p^{t_1,t_2}$ depends on the Bernoulli numbers modulo $p$. By comparing the derived codes and the ranks, we show that, for odd $p$, no code $H_p^{t_1,t_2}$ is equivalent to any of these $\mathbb{Z}_{p^s}$-linear GH codes. For $p=2$, a nonlinear code $H_2^{t_1,t_2}$ is equivalent to a $\mathbb{Z}_{2^s}$-linear Hadamard code if and only if $t_2\geq2$. Finally, we count the nonequivalent codes in the union of both families.