Computer Science & Mathematics

First and last authors make critical but distinct contributions in the majority of scientific papers. First authors tend to conduct research and analysis, whereas last authors are usually responsible for setting research directions and providing resources and supervision. The first-last author prestige gap (FL gap, $ΔH$) is defined as the H-index difference between the last and first authors prior to the focal paper's publication. The FL gap serves as a measure of the prestige asymmetry between the two dominant collaborators. We find that larger FL gaps in either direction are associated with higher impact and lower disruption. This trade-off is robust across team sizes and intensities of prior collaboration and can not be explained by first and last authors' separate prestige effects or their academic age gap. Further analysis shows that variation in last author prestige is associated with larger differences in both the article's impact and disruption than variation in first author prestige. We refer to this coexistence of higher impact and lower disruption related to the FL gap as the ``golden cage of science'', which reflects a common collaboration structure that statistically produces highly cited but less disruptive work. These findings suggest the first-last author prestige gap as an important indicator for examining how research teams balance senior guidance with exploratory autonomy.

Structured scientific spectra encode rich physical information while obeying hard constraints, such as conservation laws and characteristic spectral geometry. Accurate prediction of these spectra is central to materials discovery, yet existing methods often treat them as unconstrained sequences and therefore fail to enforce the underlying physical structure. Taking electronic density of states (eDOS) as the prototypical example, we introduce DeepSciReasoner, a general paradigm for predicting scientific spectra under physical constraints. The framework combines structure-aware spectrum decoding with constraint-preserving physical reasoning, allowing predictions to capture rich spectral structure while respecting the underlying physics. We evaluate DeepSciReasoner on eDOS, phonon density of states (phDOS), X-ray absorption near-edge structure (XANES), and Raman spectra, where it substantially improves prediction accuracy while maintaining physical consistency. These results establish DeepSciReasoner as a reusable blueprint for structured scientific spectrum prediction under hard physical constraints.

The differential neural tangent kernel (DNTK) is a tool for analyzing the training dynamics of overparameterized physics informed neural networks. In this work, we investigate the symmetry of the infinite DNTK under an orthogonal group, for two layer and deep neural networks separately, based on the explicit expression of the DNTK and recursive formula, respectively. The results are for all activations with or without a bias term, and shed insight into the structure of the infinite width DNTK. Then we prove that the integral operators over the unit sphere induced by operators from the same group orbit share the spectrum, and their reproducing kernel Hilbert spaces differ only by an orthogonal transformation, and that the operator fixed DNTK exhibits zonal symmetry on the unit sphere for symmetric operators. Thus the eigenvalue decay rate of the associated integral operator can be derived for shallow networks with RePU activation and zero bias initialization.

Sustainability debates are monopolized by quantities: total energy use, material stocks, or recycling volumes. These metrics are informative, but they obscure a more fundamental constraint. On a materially closed but energetically open Earth, long-term viability is governed by rates: power density, areal power density, and the rate at which dispersed materials can be repurified under dif- fuse solar fluxes. We develop a physical framework that links technological metabolism, material dilution, and recycling limits through these rate con- straints. This perspective clarifies why many contemporary technologies remain viable only through stock drawdown and high-power buffering, and why green growth strategies often shift rather than resolve the underlying mismatch between industrial activity and Earth-system dynamics. We argue that long-term sustain- ability requires both quantitative reduction in aggregate industrial throughput toward rate-bounded regimes and the development of technologies qualitatively compatible with biospheric operating conditions.

GEPA is among the strongest reflective prompt optimizers for language-model programs, but its system-aware merge operator is unstable across models: merge improves gpt-4.1-mini on average while degrading Qwen3-8B, an asymmetry noted but left unexplained by the GEPA paper. We investigate this instability across five datasets and two models. Our nine-configuration sweep over merge timing and merge operator shows that merge can improve both models when configured appropriately. This suggests that merge instability is better explained by configuration sensitivity than by an inherent flaw in the merge operator. We then test whether prompt-level features predict GEPA's merge-accept decisions and find that they do not, ruling out static-feature gating. As a proof of concept, a behavior-based probe built from these diagnostics avoids large merge failures in our multi-seed runs by abstaining on high-risk tasks, although it does not consistently outperform NoMerge. Code is available as GEPA-MergeProbe at https://github.com/ChongZZZZ/GEPA_MERGE .

As Vision-Language Models (VLMs) increasingly enable robots to follow language instructions, understanding their navigation decisions becomes important for reliable deployment. Existing explainability methods overlook a key challenge in Vision-Language Navigation (VLN): evidence for the current action may be distributed across historical and current observations and associated with different instruction stages. We identify instruction-stage insensitivity, where existing methods produce similar explanations despite changes across different navigation stages. To address this issue, we propose a stage-aware attribution framework for VLN. We first construct action-conditioned temporal instruction--visual attributions across observations. These attributions are projected into instruction space to estimate stage associations, which guide the aggregation of relevant historical evidence for refining the current explanation. An instruction-guided spatial filter further suppresses isolated responses while preserving compact, instruction-relevant regions. Experiments across multiple VLN policies and benchmarks show improved attribution quality over existing methods. We further demonstrate that our method acts as a diagnostic tool for analyzing failure cases, and validate its applicability to real-world navigation on a Unitree G1 robot.

Template-based crystal structure prediction (CSP) transfers a known crystal structure to a new composition by element substitution, so its accuracy depends on retrieving suitable templates, mapping elements and ranking the resulting candidates correctly. Auditing the original TCSP2.0 implementation on the T180 benchmark, we found that 469 of its 2,347 selected structures (20%) have the wrong composition, caused by element mappings that ignore stoichiometric coefficients and by sequential text substitution that cascades or corrupts element symbols. We propose TCSP3.0, which enforces coefficient-preserving mappings, substitutes species simultaneously on parsed structures and validates every candidate's composition and cell. It further adds periodic-number retrieval, soft oxidation-state evidence, radius-compatibility scoring, structural deduplication, and ORB-v3 machine-learning-potential relaxation and energy ranking of its candidates, as used by deep generative CSP baselines. TCSP3.0 achieves Top-1 and Top-20 match rates of 72.2% and 90.6% on T180, compared with 28.3% and 45.6% for TCSP2.0, and a Top-20 match rate of 74.0% on a new 800-target benchmark (T800). On T180 it also outperforms PNcsp+ and deep generative models CrystalFlow and MaskGXT (Top-1 72.2% versus at most 63.9%). Even without relaxation, TCSP3.0 reaches 67.8% at Top-1 and 90.6% at Top-20 in a median of 28 s per target on two CPU cores, about ten times faster than CrystalFlow. Paired ten-arm ablations on T180 and T800 show that most of the improvement comes from the correctness fixes; structural deduplication significantly improves the Top-5 match rate, embedding-based retrieval is required for coverage, and the remaining scoring features have small or null effects.

Robots that learn from people often see several valid ways to do the same task: pass an obstacle on the left or on the right, open the microwave or move the kettle first. A policy that re-decides at every step can waver between these ways mid-task. We ask when the action head (the part of a policy that turns the state into actions) matters, and how it should make this choice. On a controlled task, heads that sample one way succeed far more often than a head that averages the ways when the demonstrations contain several ways and the state gives no hint which to take (+0.72 success with two ways); this trend was not reproduced on the other tasks we ran. We then present ElevateHead, a head that decides once at a learned decision point, commits to that choice for the episode, and exposes the choice as an input a user can set. We designed it by diagnosis: each version was frozen and tested on tasks it had not seen, and the next version changed only what a measured failure pointed to; predictions were registered before each run. The first four versions each fell below BeT on at least one held-out task. The fifth completed all four subtasks of the held-out Franka Kitchen task in 45.2% of episodes against 29.3% for BeT, the strongest baseline across our tasks (+0.160, 95% CI [+0.129, +0.188]); when a mode was forced, the first completed subtask matched it in 95% of successful episodes. Diagnosis also located where commitment breaks: varied start positions (repaired on the controlled task: commitment 0.22 to 0.81, success 0.74 to 0.91) and varied object placement (partly repaired: in an exploratory variant, commitment 0.03 to 0.59 on aligning with no detectable change in success). On held-out block stacking, version 4 was comparable to BeT and did not exceed it. A real-robot test with predictions fixed in advance is planned for a follow-up version of this paper.

Camera motion recognition requires separating viewpoint changes caused by camera translation, rotation, or zoom from object motion and scene dynamics. Video-language models often rely on semantic cues and struggle with the geometric evidence needed for this task. Joint-Embedding Predictive Architectures such as V-JEPA offer strong spatio-temporal representations, but our experiments suggest that they underrepresent geometric cues in some regions. Under temporal-reversal supervision, V-JEPA improves overall but degrades on several motion families, whereas a weaker geometry-oriented VGGT predictor improves across all families. We refer to these geometrically underrepresented regions as geometric blind spots and propose Fovea-JEPA to identify and address them. Using V-JEPA as the base model and VGGT as a geometry expert, we construct intervention pseudo-labels from their prediction errors to weakly supervise a blind-spot predictor. The predictor produces token-wise response maps, which are aggregated into a clip-level intervention score to guide adaptive fusion of visual and geometric features. This selective intervention introduces complementary geometric information while preserving reliable visual predictions. Experiments on CameraMotionDataset, CameraBench, and ShotBench demonstrate the effectiveness of Fovea-JEPA, with gains of 9.67, 8.67, and 24.38 percentage points over previously reported state-of-the-art baselines, respectively. Code and models will be released upon acceptance.

WiFi Channel State Information (CSI) enables fine-grained sensing of spatial relationships between wireless devices. The WiFi Sensor for Robotics (WSR) toolbox demonstrated that WiFi-CSI measurements collected during robot motion, combined with odometry, can be used to synthesize virtual antenna arrays, enabling motion-induced bearing estimation in multi-robot experiments. However, deploying the WSR approach has remained challenging due to its limited integration with standard robotics tools and reliance on deprecated WiFi-CSI acquisition hardware. This work presents an open-source simulation framework that integrates the WSR sensing pipeline within the robotic simulator Gazebo. The framework generates synthetic WiFi-CSI measurements with realistic physical signal properties and leverages realistic robot motion required for motion-induced sensing and downstream localization. It further supports processing both simulated and real WiFi-CSI data for direct comparison and validation. Our aim is to make WSR-style sensing accessible as a modular sensing modality for robotic research. We validate our framework by comparing simulated and real-world experiments using TurtleBot3 robots, and demonstrate its use for bearing estimation and robot localization. This work lowers the barrier to entry for WiFi-CSI-based sensing in multi-robot systems and provides a foundation for integrating WiFi-CSI sensing with multi-modal perception for localization and coordination. All code and experimental details are publicly available at https://github.com/baharhaghighat/DARS_2026_Dijkstra_et_al

Off-policy evaluation (OPE) is essential for assessing decision policies before deployment when online experimentation is costly or unsafe. Accurate estimation remains difficult when behavior-policy data are limited and the target policy favors states and actions rarely observed in those data. We propose InCOPE, a prior-data fitted network that learns a shared policy-value prediction rule across synthetic OPE tasks. We construct an OPE task prior over Markov decision processes and behavior-target policy pairs, generating behavior-policy datasets and exact target-policy values for pretraining. For a new tabular decision problem, the pretrained network uses the behavior-policy dataset, target policy, and known task settings as context to predict the policy's expected return without task-specific parameter updates. Experiments across four tabular domains, including a semi-synthetic clinical setting, demonstrate InCOPE's effectiveness for policy-value estimation.

LVLM benchmarks tell us whether a model answers correctly, but not which visual conventions it has retained, nor where those conventions live. Sketches expose this hidden structure. A sketch removes texture, lighting, and photographic context while preserving the strokes humans judged sufficient for recognition. If an LVLM can operate on that abstraction, the surviving signal is structural rather than merely photometric. We introduce a training-free sketch probe with three readouts: temporal part-construction order, spatial attention-based grounding, and the mechanistic layer/head locus of grounding. We evaluate nine LVLMs on part order and five on the attention readouts, over two part-annotated sketch datasets. Three findings emerge. First, LVLM part order tracks human convention: models reproduce canonical orderings where human drawings converge, and become correspondingly variable where human orderings vary. Controls that blank, withhold, and mismatch the sketch locate this convention in the language model. Second, part-level grounding survives the abstraction: attention heads selected from the frozen model with an existing recipe recover part and object regions on object and scene sketches, with no grounding supervision. Third, the grounding signal concentrates in a narrow mid-network band whose location follows the base language model rather than parameter count in the two families we compare, and remains stable across object and scene sketches. The result is a practical diagnostic for sketch-driven LVLMs: no fine-tuning, no auxiliary segmenter, and a layer band that doubles as a head-selection prior, matching or improving grounding [email protected] in every model and dataset we tested.

Dynamic and continuously evolving environments present significant challenges for robotic mapping and scene understanding, particularly on construction sites where moving agents, temporary obstacles, and persistent structural changes coexist. While dynamic and long-term SLAM methods can detect motion and environmental changes, their outputs are predominantly metric representations that provide limited structure for higher-level spatial and temporal reasoning. This paper presents a LiDAR-based real-time temporal construction scene graph (T-ConSG) that jointly represents free-space regions, static objects, dynamic objects, and environmental events with spatial-relational and event-participation edges. The proposed framework incrementally constructs the graph from LiDAR-derived geometric evidence, maintaining persistent free-space and static-object information while tracking dynamic objects. Changes in accumulated spatial evidence are further represented as explicit installation and removal events, allowing the graph to capture both transient dynamics and persistent environmental evolution. We demonstrate the proposed representation through three complementary application studies: dynamics-aware navigation in a bricklaying testbed, construction progress monitoring on the ConSLAM dataset, and online change detection in a simulated construction environment. The results demonstrate that the proposed T-ConSG provides a unified representation for organizing spatial and temporal information and supports robotic reasoning across environments containing both short-term dynamics and persistent changes.

Autonomous agents are increasingly deployed across sequences of related tasks, yet they often fail to retain and improve their knowledge over time. Existing approaches convert interaction trajectories into memories or reusable skills, but experience is not automatically reliable knowledge. Automatically collected skills and memories can propagate errors, mislead, and corrupt useful knowledge. To address this gap, we introduce Knowledge Evolution from Expertise and Lessons (KEEL), an agent-agnostic framework that uses verifier evidence to transform successful behavior into reusable skills and failures into corrective lessons. As new evidence accumulates, individual skill claims can be supported, contradicted, replaced, or retired. On SkillFlow benchmark, KEEL improves task success by 4.2 to 21.7 percentage points over the no-knowledge baseline, compared with gains of 2.4 to 9.6 points from SkillFlow's default method. Ablations show that skills and lessons are complementary. These results show that reliable continual agent learning depends not simply on accumulating experience, but on continually validating and evolving the knowledge derived from it. Our code is available at: https://github.com/microsoft/KEEL-TTCL26.

Entropy in reinforcement learning with verifiable rewards (RLVR) moves up or down depending on choices that should be neutral, such as whether over-length responses are masked or counted as failures. We trace this to sampler mismatch, where rollouts are drawn with top-k/top-p truncation or at a temperature the learner does not apply, and prove that a constant advantage shift then changes the entropy of a softmax policy, at first order and in expectation, with a sign fixed by the sampler: positive shifts sharpen the policy under any sampler whose likelihood ratio to the policy increases with token probability, including top-k, top-p, min-p, and temperatures below one, and flatten it under flatter samplers. In group-based RLVR with token-level losses, the baseline-controlled part of this drift is set by the token-weighted net advantage, which masking and failing over-length responses can push to opposite signs; the remainder, a covariance between advantages and the sampler's sharpness gap, can feed back over long horizons. On a 2B model, the entropy slope is linear in shifts of the net advantage under truncated sampling and indistinguishable from flat under exact sampling, and token-level importance ratios leave 98% of the drift. In 100-step runs, a net-zero baseline only delays an entropy explosion that exact sampling does not show, a fixed positive shift delays it further, and a baseline weighted by the sharpness gap keeps entropy stable without changing the learner. On a 9B reasoning model over ten steps, the uncorrected rule loses held-out accuracy, whereas zeroing the net advantage or renormalizing the learner gains 8.2 and 8.8 points over it, with no held-out effect under exact sampling.