Science & Mathematics

Probing studies of latent reasoning often read a model's state where the answer begins and take a failed probe as evidence that a value is absent. We test this with a digit copied from the question, in the looped model Ouro-1.4B and the standard transformer Qwen 2.5-1.5B. Comparing prompts that differ only in that digit shows it is present at the last prompt token, in a code shared across questions (readout about 0.9). A probe that sees one prompt at a time, the usual setting, barely reads it: 0.36 in Ouro and 0.19 in Qwen even after 5,249 training questions. The digit's change is small rather than crowded: random noise lands on equally busy directions, while a random word in its place moves the state far more and is readable. Since every varied token leaves a trace, a successful probe is ambiguous too. We give a protocol, including a synthetic null for generic mixing, for testing computed values against copies and mixtures of the inputs. In pre-registered studies of two-step problems it finds a small signal for the unwritten intermediate a+b just before Ouro writes the answer's last digit, and its sign follows the operation, though mixing alone produces much of that sign-following part. Holding that value at another problem's from the first loop on does not move the written digit toward the other problem's, while the same patch of the answer does: the value is readable but not used.

For efficient deployment of large language models (LLMs), methods of ultra-low-bit backbones reduce weight storage, memory traffic, and linear-layer execution cost, while multi-token prediction (MTP) methods reduce sequential decoding steps by predicting multiple future tokens in parallel. Although complementary, the direct combination of these methods cannot accelerate decoding at comparable quality: masked drafting lacks intermediate token input supplied in next-token training, while wider computation and causal key-value (KV) updates can offset the savings from fewer steps. To address these challenges, we propose BitMTP, an algorithm-system co-design framework for efficient MTP on ultra-low-bit LLM backbones. A joint next-token prediction and MTP objective equips a single model with both autoregressive (AR) generation and block-causal drafting. Confidence-only decoding, speculative-only decoding, and hybrid decoding provide different quality-throughput operating points without using separate draft and verifier models. Accepted-prefix alignment keeps the committed prefix, materialized causal KV boundary, and CUDA Graph replay state consistent for continued generation. BitMTP uses this contract to select width-specialized low-bit execution and reuse captured graphs over shared KV storage. On NVIDIA H20, BitMTP with hybrid decoding achieves five-task mean speedups of 2.2x on BitNet b1.58-2B and 2.1-2.4x on 1.7B-4B Bonsai models at comparable task quality. On RTX 3090 and Jetson Orin, it further achieves up to 2.0x in five-task mean throughput and 3.3x on MBPP+, respectively. These results demonstrate that BitMTP accelerates ultra-low-bit LLM generation across backbones, model sizes, and GPU types. The source code is released at https://github.com/PolyU-VCLab/BitMTP

In the Unsplittable Capacitated Vehicle Routing Problem (UCVRP), we are given a set $V$ of $n$ clients with normalized demands $d_v \in (0,1]$, a depot $r$, a unit vehicle capacity, and symmetric metric travel costs $c$ on $V \cup \{r\}$. We seek a collection of routes of minimum total cost that start and end at $r$, serve each client's entire demand on a single route, and respect the unit capacity constraint $\sum_{v \in V(R)} d_v \le 1$ on every route $R$, where $V(R)$ denotes the clients served by $R$. This models deliveries from a common depot in which each customer's order must be delivered in a single trip. UCVRP generalizes the metric traveling salesman problem (TSP) and is therefore NP-hard. Capacity constraints add a packing difficulty: when all clients are colocated away from the depot, minimizing the route cost is equivalent to minimizing the number of bins in bin packing. Consequently, UCVRP admits no polynomial-time approximation ratio below $3/2$ unless $\textbf{P=NP}$ (Grandoni, Mathieu, and Zhou, 2023). The classical $3.5$-approximation has been improved through several recent advances; in particular, Zhao and Xiao (2026) recently obtained a $(2+\ln 2+\varepsilon)$-approximation. In this paper, we give a deterministic $(5/2+\varepsilon)$-approximation algorithm running in $n^{O(1/\varepsilon)}$ time for every rational $0

Smart glasses can help users understand their activities and surroundings, but small vision-language models must preserve useful evidence within limited computation. This is difficult because redundant frames consume the visual budget, while aggressive spatial compression can discard relevant details. Although gaze identifies the wearer's focus, the question may require evidence from the surrounding scene. To address these constraints, we propose GazeLens, which combines frame selection with gaze- and question-guided token reduction. First, HOG transitions and image sharpness select frames before visual encoding. For each selected frame, Gaussian pooling then forms gaze-focused and complementary summaries. Because their relevance depends on the question, a lightweight gate blends them into a primary token. Together with a global scene token, this retains both focused and contextual summaries within two tokens per frame. On StreamGaze, GazeLens achieves higher macro and micro accuracy than the evaluated full-grid and token-reduction baselines, with lower input-forward FLOPs than the shared-backbone token-reduction baselines.

Laboratory measurements are essential for clinical decision making but typically require invasive blood sampling. Recent research suggests that ECG recordings may contain information beyond cardiac rhythm and provide information related to laboratory values. This thesis investigates NT-proBNP prediction from ECG recordings and patient metadata using a multimodal deep learning framework, while considering the physiological plausibility of ECG-based prediction. It also compares a 1-lead configuration using lead II with the corresponding 12-lead configuration. Additional experiments on glucose, hemoglobin, and potassium are presented in the appendix. Regression, binary classification, and three-class classification experiments were conducted using MIMIC-IV, with the three-class task based on clinically established thresholds. The 12-lead configuration consistently achieved better performance. Regression reached a mean absolute error of 534.98 pg/mL for 1-lead and 514.27 pg/mL for 12-lead. Binary classification achieved AUROCs of 0.849 and 0.876, respectively, while three-class classification achieved macro-averaged AUROCs of 0.738 and 0.752. The results indicate that precise NT-proBNP regression remains challenging, whereas broader concentration ranges can be classified more effectively. The findings are consistent with the physiological hypothesis that ECG characteristics and NT-proBNP concentrations reflect shared cardiovascular processes. ECG-based NT-proBNP prediction may therefore provide complementary information, particularly for classifying broader concentration ranges. However, laboratory measurements remain necessary, and validation on independent datasets is required.

Reliable lane markings are important for both human drivers and automated driving systems, but their condition is still commonly assessed through manual inspection. Existing autonomous-driving datasets primarily support lane detection and scene understanding and usually lack maintenance-oriented labels for lane-marking visibility, continuity, or degradation. In this study, we present a geo-referenced smartphone dashcam dataset and baseline framework for lane-marking quality assessment. The dataset contains 63,839 road-facing frames collected across South Texas using a GPS-enabled dashcam application. A subset of 3,339 frames is matched with OpenStreetMap-derived lane-count labels, which are used as weak supervision for lane-count regression and map-consistency evaluation. We propose three lane-quality measures: Contrast Stability Index (CSI), which captures lane-background contrast using Laplacian responses; Line Continuity Ratio (LCR), which estimates structural continuity using edge and Hough-line evidence; and Regression Lane Confidence Score (RLCS), which measures agreement between predicted and map-derived lane counts on matched frames. The baseline prediction models attained strong performance, with CSI (MAE $0.039$, $R^2=0.97$) and LCR (MAE $0.093$, $R^2=0.84$). Notably, CSI and LCR are no-reference, segmentation-derived quality metrics that can be computed without dense human labels. In a pilot validation using 50 human-ranked images, these proposed lane-specific metrics show stronger rank agreement with human lane-quality ordering (e.g., Spearman's $ρ=0.7316$ and Kendall's $τ=0.5472$ for LCR) than general image-quality measures such as PSNR, SSIM, BRISQUE, NIQE, and LPIPS. We believe the dataset, metrics, and baseline provide an initial step toward scalable, low-cost, location-aware lane-marking quality assessment. Data and code: https://github.com/Shaif95/Road_Quality

We study variance reduction for non-convex optimization when gradient evaluations return with a delay. The server can update the parameter while earlier evaluations remain unfinished. We first analyze an ordered recursive momentum estimator for stochastic optimization. Under average smoothness and bounded variance, its oracle complexity is $O(1+σ^2ε^{-2}+LΔ_f(1+σ/ε+τ)ε^{-2})$, where the target is an expected squared gradient norm of at most $ε^2$ and $τ$ bounds the feedback delay. We next study finite sums with $n$ components and obtain $O(n+LΔ_f(\sqrt n+\min\{τ,n-1\})ε^{-2})$ evaluations. An asynchronous snapshot variant continues updating during full-gradient computations under an explicit bound on their completion time. For the stochastic method, we further replace a maximum-delay bound by a bound on the weighted amount of missing feedback.

Tactile-visual-language models combine vision and touch through a shared language space to recognize objects and materials. In real-world deployment, changes in lighting, contact quality or sensor hardware can shift visual or tactile inputs and degrade recognition performance. When both modalities contribute to a prediction, an unreliable modality can override correct evidence from the other. Since labeled data are generally unavailable during deployment, test-time adaptation offers a way to recover recognition performance from unlabeled observations. In this setting, we find that comparing accumulated observations with the text embeddings of candidate classes reveals modality reliability differences that confidence can miss. Refining features against competing classes also improves recognition beyond geometric alignment alone. These findings motivate SenseAlign, which uses a common semantic reference for test-time modality weighting and feature correction. Alignment Trust weights each modality according to its support for shared predicted classes. Anchor Realignment uses the same class text embeddings to fit a shared feature correction that separates predicted classes from alternatives. Both modules use the same unlabeled history while keeping all encoders frozen. Extensive experiments across diverse datasets demonstrate strong recognition performance under both natural shifts and controlled corruptions.

Unresolved atmospheric processes can introduce history dependence beyond the short input windows used by many weather forecasting models. We introduce GraphCast--Mamba, which augments a frozen, pretrained $1^\circ$ GraphCast backbone with a trainable Mamba correction branch combining graph message passing and Mamba layers. The branch retains a temporal memory state across successive six-hour forecast steps, learning corrections informed by temporal context beyond GraphCast's two-frame input. Training is confined to the correction branch, using ERA5 data from 2015--2021. After 2,500 optimizer updates, including warm-up, the selected model achieves substantial improvements relative to the uncorrected GraphCast backbone. It reduces the GraphCast loss (normalized weighted mean-squared error; MSE) by 14.42% when aggregated over the full ten-day forecast and by 21.66% at day 10 (240 h after forecast initialization). At this final forecast lead time, root-mean-square error (RMSE) decreases by 7.09--14.60% across all 11 evaluated variable groups, spanning temperature, wind, pressure, geopotential, specific humidity, and precipitation. These results demonstrate broad improvements in ten-day weather forecasts through a short correction-training schedule while retaining the pretrained backbone.

We study makespan minimization on unrelated machines with the additional goal of constructing diverse schedules: Instead of outputting a single schedule, we seek a collection of schedules with provable makespan guarantees and substantially different job assignments. Committing to a single schedule in advance may be undesirable. Processing times may be uncertain, and constraints or domain knowledge may emerge later, so $k$ schedules of similar quality offer meaningful alternatives. We measure pairwise diversity by the weighted Hamming distance between job assignments and study the two standard objectives in the diverse-solutions literature: max-sum diversity, which maximizes the total pairwise distance between schedules, and max-min diversity, which maximizes the smallest pairwise distance. For the max-sum objective, we show that optimal diversity can be achieved in polynomial time without sacrificing the classic makespan approximation factor. For every feasible makespan target $T$, we can compute $k$ schedules of makespan at most $2T$ whose max-sum diversity is optimal among all $k$ schedules of makespan at most $T$. Therefore, we match the classic factor-$2$ makespan guarantee of Lenstra, Shmoys, and Tardos (Math. Prog., 1990) and preserve optimal diversity. For the max-min objective, we show that the diversity problem is inherently harder. Namely, we show that, even without makespan constraints, deciding whether three schedules of given pairwise diversity exist is NP-hard. We complete this with two complementary bicriteria approximation algorithms for every fixed $k$ and $\varepsilon \in (0,1)$. The first preserves almost all of the optimal max-min diversity with makespan at most $(3\lceil12/\varepsilon\rceil+1)T$. The second keeps the makespan arbitrarily close to the classic $2$ guarantee while achieving a $\frac{1}{2}(1-\varepsilon)$ fraction of the optimal max-min diversity.

Understanding contact and tactile magnitude in hand-object interaction is a key step toward physically intelligent systems, yet learning such representations is limited by the cost and restricted coverage of tactile measurements. We present Touch-JEPA, a framework for learning tactile representations from hand-object videos with limited tactile supervision. We first introduce a self-supervised pretraining objective that learns from video alone without tactile supervision. Using video-derived hand geometry, Touch-JEPA predicts masked visual features at corresponding hand-surface locations from spatial and temporal context, encouraging the representation to capture latent contact dynamics. We then extend this formulation to joint representation learning, where masked prediction is applied to all videos while tactile supervision is available only for a labeled subset, grounding the shared representation in measured contact and magnitude. With self-supervised pretraining alone, simple supervised readouts on frozen Touch-JEPA features outperform supervised baselines in contact discrimination and tactile magnitude estimation across three human-hand benchmarks, and transfer to robotic hands without robot tactile training data. Joint learning with limited tactile supervision further improves contact and magnitude estimation on the human-hand benchmarks. Together, these results show that hand-object video can provide scalable self-supervision for tactile representation learning, while limited tactile measurements provide complementary physical grounding.

Decision models support routing, guardrails, and output verification in LLM systems. However, item accuracy alone cannot establish whether models make correct decisions across controlled variations or assign reliable confidence to their predictions. We introduce DecisionPivot, a diagnostic benchmark comprising 48 tasks and 40,033 items across seven decision roles. Its core track contains 7,800 items from 13 generated task families, organized into contrast groups with computed answers verified by independently implemented solvers. The complementary reference track contains 32,233 items from 35 public test sets, preserving their original evaluation metrics. We evaluate 40 decision models and 14 general models using documented, model-specific protocols, assessing item and contrast-group accuracy, probability ordering, calibration, and empirical selective coverage. Our results reveal substantial differences obscured by aggregate scores: two 27B models separated by only 2.6 points on the reference track achieve contrast-group accuracies of 70.1% and 44.2%. Across 2,400 binary contrast pairs, Perplexity Decider correctly ranks the positive member above the negative member in 92.3% of pairs but answers both members correctly in only 74.1%, exposing a gap between probability ordering and decision correctness. Models with similar item accuracy also differ in the reliability of their high-confidence predictions. On a separate 1,296-item subset, GLM-5.3 achieves 93.5\% contrast-group accuracy at a higher reasoning setting, compared with 81.5% at its lowest effort. By linking errors to pivotal details and difficulty levels, DecisionPivot provides fine-grained diagnostics of decision failures that aggregate evaluation alone cannot reveal.

Video generation and feed-forward 3D reconstruction typically use separate representations, requiring RGB decoding and visual re-encoding to recover geometry from generated latents. We propose Latent Bridge, which connects video generative latent spaces with geometric feature spaces to directly predict 3D geometry from video latents. We develop representation-specific bridges for Wan2.1 and V-RAE, enabling feed-forward prediction of cameras, depth, and 3D point clouds without the intermediate RGB pathway. Each bridge is trained on RGB videos through feature distillation, while the video encoder and geometry model remain frozen. Training requires neither ground-truth geometry nor execution of the downstream geometric aggregator or prediction heads. At inference, the same video latents support both RGB decoding and geometry prediction. The same trained bridge can be reused for image-to-video and text-to-video generation within the same video latent space, without bridge retraining. Experiments on Co3Dv2 and WildRGB-D demonstrate competitive geometric reconstruction quality, efficient end-to-end generation and reconstruction, and faster latent-to-geometry decoding, highlighting latent alignment as a practical and reusable interface between video generation and 3D reconstruction. Our code is available at: https://github.com/hustvl/Latent-Bridge.

Humanoid-human interaction requires robots to operate under severe partial observability: during close physical contact, the human partner is occluded, truncated, or outside the robot's field of view, causing pose estimation to fail and often necessitating external tracking or privileged human state. At the same time, heavy mutual occlusions introduce errors into the human-human motion-capture data used for learning. We present REACT, a framework for learning deployable humanoid-human interaction policies from human-human demonstrations. REACT treats semantic contact structure as the key supervisory signal connecting demonstration processing, interaction learning, and onboard control. First, a VLM-guided contact modeling and retargeting pipeline identifies semantically meaningful contacts and transfers them onto humanoid-human geometry while preserving interaction structure. Using these annotations, privileged interaction policies learn physically grounded behaviors through interaction-aware contact rewards and are then distilled into deployable egocentric depth policies trained entirely from onboard sensing using photorealistic human depth rendering. To address severe partial observability, we introduce generative partner-state modeling objectives that predict partner motion from occluded egocentric views. We evaluate REACT in simulation and on a real Unitree G1 humanoid across five contact-rich interactions. Semantic contact modeling substantially improves preservation of demonstrated interactions, while generative partner-state modeling improves contact recall on four of five interactions, most under heavy occlusion. On real hardware, REACT performs all five interactions using onboard sensing alone.

Video generation models can create realistic-looking video footage from images and text prompts. Some of the recent advances have raised concerns about the reproduction of copyrighted visual content. However, copyright leakage detection in generated videos remains largely unexplored. In this paper, we introduce a new benchmark data set for copyright leakage detection, comprising 50 copyrighted concepts, each with 5 images. For each concept, we have 5 videos created using the image+prompt video generation model and the prompt-only video generation model. Following that, we develop an AI toolset based on a contrastive retrieval framework that formulates copyright leakage detection as a video-to-image retrieval problem by matching aggregated video frame embeddings against a database of copyrighted image embeddings. Our method uses a single similarity threshold and retrieves the most similar copyrighted reference as supporting evidence. We conducted experiments showing strong retrieval and detection performance, achieving 99.6\% detection accuracy with only a single false negative. These results show the effectiveness of the retrieval-based method for scalable and explainable copyright leakage detection in AI-generated videos. Data can be accessed publicly from https://zenodo.org/records/21386117 and the source code for this project is publicly available at https://github.com/Shaif95/video-cp-Leakage.