Science & Mathematics

E-commerce decisions require applying natural-language rules to the available case evidence. We introduce WenSight-Decision-1, a shared model that scores supplied options directly from rules and case information for six operational tasks, with examples spanning six languages. We fine-tune Qwen3.5-9B with reference options as supervision, using the same option scores for training and inference. On 1,440 synthetic operational cases reserved for final evaluation, accuracy rises from 61.60% for Qwen3.5-9B to 92.92% for WenSight-Decision-1 under identical inputs and scoring. Joint fine-tuning improves all six tasks, bringing heterogeneous operational judgments under a common decision interface.

Electromyography (EMG)-driven control enables wearable robots to assist users based on their neuromuscular activity. Upper-limb exoskeletons typically rely on proportional myoelectric control (PMC), which amplifies user effort but may compromise stability at high assistance levels. EMG-driven adaptive gravity compensation (AGC) offers a promising alternative, counterbalancing gravitational torques while continuously adapting to neuromuscular activity. However, these approaches have not been systematically compared in a human-in-the-loop setting. This study evaluates five control algorithms: two PMC strategies at different assistance levels, a purely integral AGC (AGC-PI), a kinematics- and acceleration-aware model-based (MB) AGC (AGC-AW), and a novel MB AGC approach based on Recursive Least Squares (AGC-RLS) for online payload estimation. Fourteen healthy participants performed target-reaching tasks using an active elbow exoskeleton under 0 kg and 2 kg loading conditions. Performance metrics included muscular activity, movement accuracy, task efficiency, and subjective workload. PMC exhibited a trade-off between physical effort and task performance: low assistance levels required greater physical effort under load, whereas high gains impaired task execution. AGC-PI reduced agonist activity but induced antagonist activation. In contrast, both model-based AGC approaches effectively compensated for external payloads, reducing physical and cognitive demands while improving movement accuracy and completion time.

Ground truth in Speech Emotion Recognition (SER) is often presented as a single categorical fact, yet an acted emotion and the emotion heard by listeners describe different aspects of the same recording. We examine this distinction in Emozionalmente, an acted Italian corpus that provides both an elicited production label and five listener judgements for every clip. Using frozen XLS-R, HuBERT, and WavLM representations with matched probes, we compare three label authorities: the expressed condition, the listeners' hard plurality, and their complete vote distribution. The results tell a consistent story. Recovering the expressed condition is easier than predicting a tie-broken listener plurality, but that gap compares two different questions rather than revealing a general loss of SER ability. As listener agreement weakens, the plurality becomes an increasingly brittle hard target. Training on the full vote distribution does not substantially change which class wins the model's hard decision, but it produces probability assignments that more closely follow the alternatives selected by listeners, especially for ambiguous clips. No authority therefore replaces the others. Expressed labels describe how speech was produced, pluralities provide an operational single-class decision, and vote distributions preserve perceptual uncertainty. For human-facing Internet of Sounds applications, evaluations should state which authority is being predicted, retain its aggregation rule and agreement information, and use distributional scores when disagreement is part of the target.

The real-world usability of ASR models remains challenging, particularly for low-resource languages, where existing speech resources are often collected in clean studio environments. In this work, we introduce TeleASR-Amharic, a benchmark for evaluating Amharic ASR under various acoustic and telecommunication conditions. We analyze the major degradations observed in real telecommunications recordings and develop a benchmark that covers multiple channel conditions, environmental disturbances, and telecom-domain variations. Our results show that telecommunication degradation can substantially reduce recognition performance, making even high-performing ASR models impractical under severe conditions. For example, the WER of the best-performing model increases from 0.23 to 0.79 under severe telecommunication degradation, corresponding to a relative increase of 243% in recognition errors. We further investigate the effects of different noise and channel distortions and show that scaling training data and applying carefully designed telecommunication-aware augmentation substantially improve robustness. In particular, our best training configuration achieves a 23.6% relative WER reduction under extreme noise conditions. Together with the release of our telecommunications-domain data, these findings provide a reproducible benchmark for measuring progress in Amharic ASR while highlighting open challenges in developing robust speech technologies for real-world deployment.

Players often have ideas about a game while playing it, but turning them into changes usually means stepping out of play to edit it. We built PATCH FIGHTER, a two-player fighting game defined by 91 types of adjustable parameters. A large language model (LLM) reads players' conversations during play and rewrites these parameters between games, so players can change the game just by talking. We conducted a within-subjects experiment with 27 pairs (54 participants) comparing three conditions, solo play, paired play, and paired watching. Participants rated enjoyment after each game and answered a questionnaire on speech and changes after each condition. Paired play was the most enjoyable condition. In paired play, participants also felt more able to say what they thought about the game than in solo play. In interviews, participants valued talking with the partner and checking changes through their own play. Paired watching was less preferred because they could not try the changes themselves. As our design cannot separate the enjoyment of editing from that of play, we outline a design that measures them separately and the evaluation of the resulting games as future work.

Binary Goppa codes are important in coding theory and cryptography. It is well-known that the minimum distance of a binary Goppa code defined by a squarefree polynomial of degree $r$ is at least $2r+1$. Though from the classical results of C. J. Moreno, O. Moreno, F. Levy-dit-Vehel and S. Litsyn that the true minimum distances of binary Goppa codes are $2r+1$, when the extension field degree $m$ is larger than $(r-1)^{4r}$. There have only very few families of binary Goppa codes with determined true minimum distances. In this paper, we give necessary and sufficient conditions for binary Goppa codes such that their minimum distances are $2r+1$ and $2r+2$. Infinitely many infinite families of binary Goppa codes with minimum distance $2r+1$ are given. In particular, we construct such codes for every Goppa degree $r$ in the range $2^{h-1}

Unified large multimodal models (ULMMs) are able to perform visual reasoning through interleaved text-image generation, but generating explicit intermediate images incurs significant computational overhead. While visual latent reasoning addresses this, most existing methods rely on vision-language models (VLMs) incapable of visual generation. We propose Oddin, a novel framework enabling latent visual reasoning in ULMMs. Oddin replaces intermediate images with compact latent tokens processed entirely by the understanding branch, reserving the generation branch for the final output image. We train the model via offline self-distillation with three objectives, final image generation losses, text distillation losses, and cross-entropy losses for special tokens. Oddin surpasses interleaved generation baselines and closed-source VLMs on average across seven VBVR-Pro visual reasoning tasks. Qualitative analysis reveals that latent tokens attend to relevant regions in an input image, demonstrating that ULMMs can perform complex visual reasoning without generating intermediate images.

Given a mixed graph $G$ and an ordered pair $(u,v)$ of vertices in $V(G)$, we say that an orientation $\vec{G}$ of $G$ satisfies $(u,v)$ if $v$ is reachable from $u$ in $\vec{G}$. We consider problems in which we seek orientations that satisfy a maximum number of pairs. We distinguish between two different problems. In the most general problem called Fixed Pairs Maximum Mixed Reachability Orientation (FPRO), a multiset $P$ of ordered pairs of vertices in $V(G)$ comes with the input. In the more restricted problem All Pairs Maximum Mixed Reachability Orientation (APRO), $P$ contains exactly one copy of every ordered pair of distinct vertices in $V(G)$. We compare the computational complexity of FPRO and APRO in two ways: First, we show that both FPRO and APRO behave similarly when it comes to approximation. Namely, they both admit an asymptotic approximation, but are unlikely to admit a constant factor approximation. Then, we consider the parameterization with respect to the treewidth of the underlying graph of the input mixed graph. We show that here the complexities of FPRO and APRO differ significantly. While it is well-known that FPRO is NP-hard even when this parameter is 1, we show that APRO is XP with respect to this parameter. On the other hand, we give a $W[1]$-hardness result for APRO.

Text-to-image diffusion models can produce unsafe content, while output filtering intervenes only after synthesis. We investigate whether native generation signals enable early detection of unsafe and adversarial inputs. We find that classifier-free guidance exposes strong safety signals in conditional responses, accessible through compact statistics and linear readouts. Based on this finding, we propose Conditional Semantic Trajectory Interception (CSTI), which aggregates activation contrasts between conditional and unconditional responses at early checkpoints for calibrated risk detection. CSTI fits concept-specific detectors using canonical concept and safe examples, without attack samples, intermediate image decoding, or generator modification. Across five generators, four adversarial attack collections, and two general NSFW datasets, CSTI achieves the strongest overall performance among six baselines at a matched 5% FPR. Early interception reduces generation latency by over 60% on intercepted paths. These results demonstrate the effectiveness of native conditional responses for early safety interception.

Self-improving agents aim to continuously improve their behavior by learning from experience after deployment. However, accumulating and reusing more experience does not necessarily lead to better decisions, particularly when generation, interaction, and memory resources are jointly constrained. Under such stringent constraints, we find that several existing lifelong-memory methods provide limited benefits over simple retrieval or even no-memory agents. This reveals a fundamental challenge for resource-efficient self-improvement: agents must learn not only which past experiences are relevant, but which are actually worth reusing to improve future decisions. When memory capacity and opportunities for reasoning and recovery are limited, retrieving an ineffective experience can consume scarce resources and hinder downstream execution. To address this challenge, we propose TokenWorth, a lightweight feedback-driven memory adaptation framework that learns query-dependent memory worth from sparse execution feedback. Without updating the backbone model, TokenWorth uses the learned worth to guide both experience retrieval and bounded memory management, enabling the agent to continually adapt how it reuses accumulated experience. Across three LifelongAgentBench environments and both local and API-based deployments, TokenWorth achieves the highest task success rate while using the fewest output tokens among the evaluated methods under the specified resource budgets. These results suggest that effective self-improvement under resource constraints depends not on accumulating more experience, but on learning what is worth remembering. Our code is publicly available at https://github.com/AIGNLAI/TokenWorth.

Human-collected robot manipulation data often mix successful behavior with retries, pauses, recoveries, and failures. Advantage-Conditioned Behavior Cloning (AC-BC) can exploit such mixed-quality data, but only if the conditioning signal is both accurately constructed from offline trajectories and effectively incorporated into the policy's action-vector-field prediction. Existing pipelines can weaken this signal by thresholding estimator-derived advantage scores into noisy conditions, and by routing conditions through prompts that weakly affect the action vector field. We introduce ALIGN, a condition-aligned AC-BC framework for vision-language-action policies. ALIGN directly predicts categorical pairwise conditions in the policy-facing label space and injects them into a DiT action expert through AdaRMS modulation fused with flow time. We evaluate ALIGN on deliberately mixed real-robot datasets for ARX X5 bowl stacking and Franka key hanging, each combining standard demonstrations with retry-heavy trajectories. ALIGN improves condition fidelity, strengthens conditional vector-field separability, and achieves 90.0% and 56.7% closed-loop success, compared with 56.7% and 33.3% for the strongest AC-BC baseline. These results indicate that aligning conditions from estimation to action prediction is an effective way to extract robust manipulation policies from data in which success, recovery and error coexist.

Sector rotation is a common decision in benchmark-relative portfolio management, where portfolio managers rank sectors for overweight and underweight positions. We formulate this task as ranking sector actions by expected future risk-adjusted active performance, measured with forward sector-benchmark Sharpe ratios. Such rankings should preserve temporal information, express model outputs through economically familiar signal characteristics, account for changes in signal relevance across macroeconomic environments, and capture dependencies among sectors. We propose RegimeRank, an interpretable sector-ranking model that represents fundamental and technical signals through expert-defined temporal characteristics, conditions them on learned macro factors via cross-attention, and models cross-sector dependencies through self-attention. Expected-gradient attributions identify the contributions of individual signal-transformation pairs, while macroeconomic and cross-sector attention maps provide diagnostics of the learned conditioning and relational structure. We evaluate RegimeRank on weekly S&P 500 data from 2002 to 2024 under combinatorial purged cross-validation, where it outperforms representative ranking and machine-learning baselines across multiple ranking metrics. Ablation studies support the contributions of latent macro states, regime conditioning, and cross-sector modelling, while attribution and attention analyses illustrate the temporal characteristics, macro states, and sector relationships underlying its rankings.

Brain tumor segmentation uses complementary magnetic resonance imaging (MRI) sequences, but some sequences may be unavailable in practice. Their absence changes the information available for segmenting different tumor regions. Fusion methods can support different input subsets. However, fusion at several scales does not by itself specify how tumor evidence is retained and updated to guide fusion at later stages. We propose a progressive multi-scale tumor-state memory network (PMTM) for segmentation from any non-empty subset of T1ce, T1, FLAIR, and T2. PMTM maintains a compact state for the current subject across the encoder stages. At each stage, we use assignment maps to pool spatial features into ordered evidence vectors for each available modality. We average these vectors across the available modalities, and a learned gate combines this evidence with the previous state. The up- dated state guides residual fusion at that stage and is used in the state update at the next stage. A modality-token mixer also models interactions among the available modalities at each stage. We apply semantic supervision to the assignment maps at all four stages. We evaluate PMTM on BraTS2018, BraTS2020, and BraTS2023 GLI across all 15 non-empty modality subsets. Averaged over the three tumor regions and the 15 subsets, PMTM achieves Dice scores of 75.31%, 77.54%, and 80.47% on BraTS2018, BraTS2020, and BraTS2023 GLI, respectively. These scores are higher than those of RobustSeg, RFNet, and mmFormer in our experiments on each dataset. Component ablations and internal state analysis are performed on BraTS2020.

Solutions of second-order elliptic PDE problems on nonconvex or cracked domains with mixed Dirichlet-Neumann conditions typically lie outside $H^2(Ω)$. Near re-entrant corners, crack tips and points where the boundary condition changes type, the gradient grows like $r^{λ-1}$ with $λ

Obtaining defect-free single-crystals is of crucial importance in soft matter and materials science, spanning both fundamental interests and practical applications. Active crystals, encompassing diverse dense systems from living organisms to artificial metamaterials, can offer a promising platform for autonomous structural reorganization. We demonstrate that increasing spatiotemporal orientational fluctuations around the mean polarized state can enhance crystallographic order in active crystals. Poly-to-single-crystal transformations in defect-rich, self-aligned active Brownian crystals with coupled translational-rotational dynamics are investigated. We show that long-lived persistence with slow alignment controls the defect-annealing capability. By disentangling activity magnitude from associated timescales, we reveal activity-assisted annealing. Tuning the persistence length enables overcoming thermal energy barriers. Simultaneously adjusting persistence and alignment timescales can tune relaxation from `fast' to `slow', yielding defect-rich and defect-free crystals, respectively. Furthermore, active crystals of point particles are found to exhibit similar defect annealing when flocking is delayed by tuning the timescale of the velocity relaxation towards the polarized state. These findings suggest that the mechanism may extend to a broader class of active crystals. Our results intersect statistical physics of living soft matter with the mechanics of defects and microstructures, proposing a nonequilibrium route towards reconfigurable materials.