Diffusion models generate data by removing noise gradually, which corresponds to the time-reversal of a noising process. However, access to only the denoising kernels is often insufficient. In many applications, we need the knowledge of the marginal densities along the generation trajectory, which enables tasks such as inference-time control. To address this gap, in this paper, we introduce the Radon-Nikodym Estimator (RNE). Based on the concept of the \textit{density ratio} between path distributions, it reveals a fundamental connection between marginal densities and transition kernels, providing a flexible plug-and-play framework that unifies (1) diffusion density estimation, (2) inference-time control, and (3) energy-based diffusion training under a single perspective. Experiments demonstrate that RNE delivers strong results in inference-time control applications, such as annealing and model composition, with promising inference-time scaling performance, and achieves a simple yet efficient regularisation for training energy-based diffusion models. Additionally, our proposed RNE is modality-agnostic and applicable not only to continuous diffusion models but also to their discrete diffusion counterparts.

As large language models evolve into tool-augmented agents, a central question remains unresolved: when is external tool use actually justified? Existing agent frameworks typically treat tools as ordinary actions and optimize for task success or reward, offering little principled distinction between epistemically necessary interaction and unnecessary delegation. This position paper argues that agents should invoke external tools only when epistemically necessary. Here, epistemic necessity means that a task cannot be completed reliably via the agent's internal reasoning over its current context, without any external interaction. We introduce the Theory of Agent (ToA), a framework that treats agents as making sequential decisions about whether remaining uncertainty should be resolved internally or delegated externally. From this perspective, common agent failure modes (e.g., overthinking and overacting) arise from miscalibrated decisions under uncertainty rather than deficiencies in reasoning or tool execution alone. We further discuss implications for training, evaluation, and agent design, highlighting that unnecessary delegation not only causes inefficiency but can impede the development of internal reasoning capability. Our position provides a normative criterion for tool use that complements existing decision-theoretic models and is essential for building agents that are not only correct, but increasingly intelligent.

Direct training of Spiking Neural Networks (SNNs) on neuromorphic hardware can greatly reduce energy costs compared to GPU-based training. However, implementing Backpropagation (BP) on such hardware is challenging because forward and backward passes are typically performed by separate networks with distinct weights. To compute correct gradients, forward and feedback weights must remain symmetric during training, necessitating weight transport between the two networks. This symmetry requirement imposes hardware overhead and increases energy costs. To address this issue, we introduce Frozen Backpropagation (\textsc{fBP}), a BP-based training algorithm relaxing weight symmetry in settings with separate networks. fBP updates forward weights by computing gradients with periodically frozen feedback weights, reducing weight transports during training and minimizing synchronization overhead. To further improve transport efficiency, we propose three partial weight transport schemes of varying computational complexity, where only a subset of weights is transported at a time. We evaluate our methods on image recognition tasks using both temporally and rate-coded SNNs, and compare them to existing approaches addressing the weight symmetry requirement. Our results show that fBP outperforms these methods and achieves accuracy comparable to BP while significantly lowering transport costs. With partial weight transport, fBP can further lower those costs by up to 10,000x at the expense of moderate accuracy loss. This work provides insights for guiding the design of neuromorphic hardware incorporating BP-based on-chip learning.

Modeling the joint distribution of data samples and their properties allows to construct a single model for both data generation and property prediction, with synergistic benefits reaching beyond purely generative or predictive models. However, training joint models presents daunting architectural and optimization challenges. Here, we propose Hyformer, a transformer-based joint model that successfully blends the generative and predictive functionalities, using an alternating attention mechanism and a joint pre-training scheme. We show that Hyformer is simultaneously optimized for molecule generation and property prediction, while exhibiting synergistic benefits in conditional sampling, out-of-distribution property prediction and representation learning. Finally, we demonstrate the benefits of joint learning in a drug design use case of discovering novel antimicrobial~peptides.

Contrastive learning methods in computer vision typically rely on augmented views of the same image or multimodal pretraining strategies that align paired modalities. However, these approaches often overlook semantic relationships between distinct instances, leading to false negatives when semantically similar samples are treated as negatives. This limitation is especially critical in medical imaging domains such as cardiology, where demographic and clinical attributes play a critical role in assessing disease risk and patient outcomes. We introduce a tabular-guided contrastive learning framework that leverages clinically relevant tabular data to identify patient-level similarities and construct more meaningful pairs, enabling semantically aligned representation learning without requiring joint embeddings across modalities. Additionally, we adapt the k-NN algorithm for zero-shot prediction to overcome the lack of zero-shot capability in unimodal representations. We demonstrate the strength of our methods using a large cohort of short-axis cardiac MR images and clinical attributes, where tabular data helps to more effectively distinguish between patient subgroups. Evaluation on downstream tasks, including fine-tuning, linear probing, and zero-shot prediction of cardiovascular artery diseases and cardiac phenotypes, shows that incorporating tabular data guidance yields stronger visual representations than conventional methods that rely solely on image augmentation or combined image-tabular embeddings. Further, we show that our method can generalize to natural images by evaluating it on a car advertisement dataset. Code is available at https://github.com/marteczkah/tables_guide_vision.

We study bi-criteria combinatorial optimization under noisy function evaluations. While resilience and black-box offline-to-online reductions have been studied in single-objective settings, extending these ideas to bi-criteria problems introduces new challenges due to the coupled degradation of approximation guarantees for objectives and constraints. We introduce a notion of $(α,β,δ,\texttt{N})$-resilience for bi-criteria approximation algorithms, capturing how joint approximation guarantees degrade under bounded (possibly worst-case) oracle noise, and develop a general black-box framework that converts any resilient offline algorithm into an online algorithm for bi-criteria combinatorial multi-armed bandits with bandit feedback. The resulting online guarantees achieve sublinear regret and cumulative constraint violation of order $\tilde{O}(δ^{2/3}\texttt{N}^{1/3}T^{2/3})$ without requiring structural assumptions such as linearity, submodularity, or semi-bandit feedback on the noisy functions. We demonstrate the applicability of the framework by establishing resilience for several classical greedy algorithms in submodular optimization.

The severe shortage of medical doctors limits access to timely and reliable healthcare, leaving millions underserved. Large language models (LLMs) offer a potential solution but struggle in real-world clinical interactions. Many LLMs are not grounded in authoritative medical guidelines and fail to transparently manage diagnostic uncertainty. Their language is often rigid and mechanical, lacking the human-like qualities essential for patient trust. To address these challenges, we propose Ask Patients with Patience (APP), a multi-turn LLM-based medical assistant designed for grounded reasoning, transparent diagnoses, and human-centric interaction. APP enhances communication by eliciting user symptoms through empathetic dialogue, significantly improving accessibility and user engagement. It also incorporates Bayesian active learning to support transparent and adaptive diagnoses. The framework is built on verified medical guidelines, ensuring clinically grounded and evidence-based reasoning. To evaluate its performance, we develop a new benchmark that simulates realistic medical conversations using patient agents driven by profiles extracted from real-world consultation cases. We compare APP against SOTA one-shot and multi-turn LLM baselines. The results show that APP improves diagnostic accuracy, reduces uncertainty, and enhances user experience. By integrating medical expertise with transparent, human-like interaction, APP bridges the gap between AI-driven medical assistance and real-world clinical practice.

Owing to recent advances in machine learning and the ability to harvest large amounts of data during robotic-assisted surgeries, surgical data science is ripe for foundational work. We present a large dataset of surgical videos and their accompanying labels for this purpose. We describe how the data was collected and some of its unique attributes. Multiple example problems are outlined. Although the dataset was curated for a particular set of scientific challenges (in an accompanying paper), it is general enough to be used for a broad range machine learning questions. Our hope is that this dataset exposes the larger machine learning community to the challenging problems within surgical data science, and becomes a touch-stone for future research. The videos are available at https://storage.googleapis.com/isi-surgvu/surgvu24_videos_only.zip, the labels at https://storage.googleapis.com/isi-surgvu/surgvu24_labels_updated_v2.zip, a validation set for tool detection problem at https://storage.googleapis.com/isi-surgvu/cat1_test_set_public.zip, and a sample set of question & answer pairs dataset for surgical visual question answering at https://storage.googleapis.com/isi-surgvu/SURGVU25_cat_2_sample_set_public.zip.

This paper presents UNCOM, a novel hybrid framework for interpreting natural human commands in tabletop scenarios. The system integrates multiple sources of information -- speech, gestures, and scene context -- to extract structured, actionable instructions for robots. Addressing the need for general-purpose human-robot interaction in domestic environments, UNCOM is designed for zero-shot operation, without reliance on predefined object models or training data specific to a given task. Using foundational and task-specific deep learning models, it allows out-of-the-box speech recognition, natural language understanding, gesture detection, and object segmentation. The modular architecture enhances transparency and explainability by explicitly parsing commands into object-action-target representations, enabling integration with symbolic robotic frameworks. We demonstrate the system in a TIAGo++ robot and provide an evaluation on a real-world data set of human-robot interaction scenarios; achieving an 82.39\% success rate over our benchmark data set, highlighting the robustness of the system to diversity, noise, and communication ambiguity. The data set, evaluation scenarios, and the code are publicly available to support future research.

Reinforcement learning (RL) in robotics faces significant hurdles regarding sample efficiency and generalization across varying goals. While Offline RL mitigates the need for costly online interactions, its integration with goal-conditioned policies and transformer-based architectures remains underexplored. We introduce a Goal-Conditioned Decision Transformer adapted for offline multi-goal robotics. By explicitly incorporating goal states into the sequence modeling framework, our approach efficiently solves varying tasks using only pre-collected data. We validate this method on a newly released offline dataset for the Franka Emika Panda platform. Experimental results demonstrate that our approach outperforms state-of-the-art online baselines in complex tasks and maintains robustness in sparse-reward settings, even with limited expert demonstrations.

Voice signals originating from the respiratory tract are utilized as valuable acoustic biomarkers for the diagnosis and assessment of respiratory diseases. Among the employed acoustic features, Mel Frequency Cepstral Coefficients (MFCC) is widely used for automatic analysis, with MFCC extraction commonly relying on default parameters. However, no comprehensive study has systematically investigated the impact of MFCC extraction parameters on respiratory disease diagnosis. In this study, we address this gap by examining the effects of key parameters, namely the number of coefficients, frame length, and hop length between frames, on respiratory condition examination. Our investigation uses four datasets: the Cambridge COVID-19 Sound database, the Coswara dataset, the Saarbrucken Voice Disorders (SVD) database, and a TACTICAS dataset. The Support Vector Machine (SVM) is employed as the classifier, given its widespread adoption and efficacy. Our findings indicate that the accuracy of MFCC decreases as hop length increases, and the optimal number of coefficients is observed to be approximately 30. The performance of MFCC varies with frame length across the datasets: for the COVID-19 datasets (Cambridge COVID-19 Sound database and Coswara dataset), performance declines with longer frame lengths, while for the SVD dataset, performance improves with increasing frame length (from 50 ms to 500 ms). Furthermore, we investigate the optimized combination of these parameters and observe substantial enhancements in accuracy. Compared to the worst combination, the SVM model achieves an accuracy of 81.1%, 80.6%, and 71.7%, with improvements of 19.6%, 16.10%, and 14.90% for the Cambridge COVID-19 Sound database, the Coswara dataset, and the SVD dataset respectively.

Recent works utilize CLIP to perform the challenging unsupervised semantic segmentation task where only images without annotations are available. However, we observe that when adopting CLIP to such a pixel-level understanding task, unexpected bias (including class-preference bias and space-preference bias) occurs. Previous works don't explicitly model the bias, which largely constrains the segmentation performance. In this paper, we propose to explicitly model and rectify the bias existing in CLIP to facilitate the unsupervised semantic segmentation task. Specifically, we design a learnable "Reference" prompt to encode class-preference bias and a projection of the positional embedding in the vision transformer to encode space-preference bias respectively. To avoid interference, two kinds of biases are firstly independently encoded into different features, i.e., the Reference feature and the positional feature. Via a matrix multiplication between the Reference feature and the positional feature, a bias logit map is generated to explicitly represent two kinds of biases. Then we rectify the logits of CLIP via a simple element-wise subtraction. To make the rectified results smoother and more contextual, we design a mask decoder which takes the feature of CLIP and the rectified logits as input and outputs a rectified segmentation mask with the help of Gumbel-Softmax operation. A contrastive loss based on the masked visual features and the text features of different classes is imposed, which makes the bias modeling and rectification process meaningful and effective. Extensive experiments on various benchmarks including PASCAL VOC, PASCAL Context, ADE20K, Cityscapes, and COCO Stuff demonstrate that our method performs favorably against previous state-of-the-arts. The implementation is available at: https://github.com/dogehhh/ReCLIP.

We introduce Proximal Policy Distillation (PPD), a novel policy distillation method that integrates student-driven distillation and Proximal Policy Optimization (PPO) to increase sample efficiency and to leverage the additional rewards that the student policy collects during distillation. To assess the efficacy of our method, we compare PPD with two common alternatives, student-distill and teacher-distill, over a wide range of reinforcement learning environments that include discrete actions and continuous control (ATARI, Mujoco, and Procgen). For each environment and method, we perform distillation to a set of target student neural networks that are smaller, identical (self-distillation), or larger than the teacher network. Our findings indicate that PPD improves sample efficiency and produces better student policies compared to typical policy distillation approaches. Moreover, PPD demonstrates greater robustness than alternative methods when distilling policies from imperfect demonstrations. The code for the paper is released as part of a new Python library built on top of stable-baselines3 to facilitate policy distillation: `sb3-distill'.

Urban land use inference is a critically important task that aids in city planning and policy-making. Recently, the increased use of sensor and location technologies has facilitated the collection of multi-modal mobility data, offering valuable insights into daily activity patterns. Many studies have adopted advanced data-driven techniques to explore the potential of these multi-modal mobility data in land use inference. However, existing studies often process samples independently, ignoring the spatial correlations among neighbouring objects and heterogeneity among different services. Furthermore, the inherently low interpretability of complex deep learning methods poses a significant barrier in urban planning, where transparency and extrapolability are crucial for making long-term policy decisions. To overcome these challenges, we introduce an explainable framework for inferring land use that synergises heterogeneous graph neural networks (HGNs) with Explainable AI techniques, enhancing both accuracy and explainability. The empirical experiments demonstrate that the proposed HGNs significantly outperform baseline graph neural networks for all six land-use indicators, especially in terms of 'office' and 'sustenance'. As explanations, we consider feature attribution and counterfactual explanations. The analysis of feature attribution explanations shows that the symmetrical nature of the `residence' and 'work' categories predicted by the framework aligns well with the commuter's 'work' and 'recreation' activities in London. The analysis of the counterfactual explanations reveals that variations in node features and types are primarily responsible for the differences observed between the predicted land use distribution and the ideal mixed state. These analyses demonstrate that the proposed HGNs can suitably support urban stakeholders in their urban planning and policy-making.

Model-predictive control (MPC) is a state-of-the-art control method for constrained robotic systems, yet deployment on resource-limited hardware remains difficult. This challenge is magnified by expressive conic constraints, which offer greater modeling power but require significantly more computation than linear alternatives. To address this challenge, we extend recent work developing fast, structure-exploiting, cached solvers for embedded applications based on the Alternating Direction Method of Multipliers (ADMM) to provide support for second-order cones, as well as C++ code generation from Python, MATLAB, and Julia. Microcontroller benchmarks show that our solver provides up to a two-order-of-magnitude speedup, ranging from 10.6x to 142.7x, over state-of-the-art embedded solvers on QP and SOCP problems, and enables us to fit order-of-magnitude larger problems in memory. We validate our solver's deployed performance through simulation and hardware experiments, including trajectory tracking with conic constraints on a 27g Crazyflie quadrotor. Our open-source code is available at https://tinympc.org.

Unsupervised semantic segmentation aims to achieve high-quality semantic grouping without human-labeled annotations. With the advent of self-supervised pre-training, various frameworks utilize the pre-trained features to train prediction heads for unsupervised dense prediction. However, a significant challenge in this unsupervised setup is determining the appropriate level of clustering required for segmenting concepts. To address it, we propose a novel framework, CAusal Unsupervised Semantic sEgmentation (CAUSE), which leverages insights from causal inference. Specifically, we bridge intervention-oriented approach (i.e., frontdoor adjustment) to define suitable two-step tasks for unsupervised prediction. The first step involves constructing a concept clusterbook as a mediator, which represents possible concept prototypes at different levels of granularity in a discretized form. Then, the mediator establishes an explicit link to the subsequent concept-wise self-supervised learning for pixel-level grouping. Through extensive experiments and analyses on various datasets, we corroborate the effectiveness of CAUSE and achieve state-of-the-art performance in unsupervised semantic segmentation.

Time Series Extrinsic Regression (TSER) involves using a set of training time series to form a predictive model of a continuous response variable that is not directly related to the regressor series. The TSER archive for comparing algorithms was released in 2022 with 19 problems. We increase the size of this archive to 63 problems and reproduce the previous comparison of baseline algorithms. We then extend the comparison to include a wider range of standard regressors and the latest versions of TSER models used in the previous study. We show that none of the previously evaluated regressors can outperform a regression adaptation of a standard classifier, rotation forest. We introduce two new TSER algorithms developed from related work in time series classification. FreshPRINCE is a pipeline estimator consisting of a transform into a wide range of summary features followed by a rotation forest regressor. DrCIF is a tree ensemble that creates features from summary statistics over random intervals. Our study demonstrates that both algorithms, along with InceptionTime, exhibit significantly better performance compared to the other 18 regressors tested. More importantly, these two proposals (DrCIF and FreshPRINCE) models are the only ones that significantly outperform the standard rotation forest regressor.

In 2017, a research paper compared 18 Time Series Classification (TSC) algorithms on 85 datasets from the University of California, Riverside (UCR) archive. This study, commonly referred to as a `bake off', identified that only nine algorithms performed significantly better than the Dynamic Time Warping (DTW) and Rotation Forest benchmarks that were used. The study categorised each algorithm by the type of feature they extract from time series data, forming a taxonomy of five main algorithm types. This categorisation of algorithms alongside the provision of code and accessible results for reproducibility has helped fuel an increase in popularity of the TSC field. Over six years have passed since this bake off, the UCR archive has expanded to 112 datasets and there have been a large number of new algorithms proposed. We revisit the bake off, seeing how each of the proposed categories have advanced since the original publication, and evaluate the performance of newer algorithms against the previous best-of-category using an expanded UCR archive. We extend the taxonomy to include three new categories to reflect recent developments. Alongside the originally proposed distance, interval, shapelet, dictionary and hybrid based algorithms, we compare newer convolution and feature based algorithms as well as deep learning approaches. We introduce 30 classification datasets either recently donated to the archive or reformatted to the TSC format, and use these to further evaluate the best performing algorithm from each category. Overall, we find that two recently proposed algorithms, Hydra+MultiROCKET and HIVE-COTEv2, perform significantly better than other approaches on both the current and new TSC problems.

The Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE) is a heterogeneous meta ensemble for time series classification. HIVE-COTE forms its ensemble from classifiers of multiple domains, including phase-independent shapelets, bag-of-words based dictionaries and phase-dependent intervals. Since it was first proposed in 2016, the algorithm has remained state of the art for accuracy on the UCR time series classification archive. Over time it has been incrementally updated, culminating in its current state, HIVE-COTE 1.0. During this time a number of algorithms have been proposed which match the accuracy of HIVE-COTE. We propose comprehensive changes to the HIVE-COTE algorithm which significantly improve its accuracy and usability, presenting this upgrade as HIVE-COTE 2.0. We introduce two novel classifiers, the Temporal Dictionary Ensemble (TDE) and Diverse Representation Canonical Interval Forest (DrCIF), which replace existing ensemble members. Additionally, we introduce the Arsenal, an ensemble of ROCKET classifiers as a new HIVE-COTE 2.0 constituent. We demonstrate that HIVE-COTE 2.0 is significantly more accurate than the current state of the art on 112 univariate UCR archive datasets and 26 multivariate UEA archive datasets.

World Action Models (WAMs) enable decision-making through imagined rollouts by predicting future observations and actions. However, the reliability of these imagined futures remains under-examined: is a generated future merely visually plausible, or is it dynamically compatible with the action sequence it claims to model? In this work, we identify action-state consistency, the alignment between predicted actions and induced state transitions, as a missing reliability axis for WAMs. Through a systematic study across representative joint-prediction and inverse-dynamics models, we find that action-state consistency systematically separates successful and failed rollouts across many tasks and follows similar success-failure trends as learned value estimates. These results suggest that consistency captures decision-relevant structure beyond visual realism. We further identify background collapse as an important boundary condition, where low-dynamics failed trajectories can become deceptively consistent because static futures are easier to predict. Building on these findings, we introduce a value-free consensus strategy for test-time selection, which ranks candidate rollouts by agreement among predicted futures. This strategy improves success rates on RoboCasa and RoboTwin 2.0 without additional training or reward modeling. Taken together, our findings establish action-state consistency as both a diagnostic tool for evaluating WAM reliability and a practical signal for value-free planning.

We study Flow Matching in a semi-discrete setting where a Gaussian source is transported toward a discrete target supported on finitely many points. This semi-discrete regime is the theoretical setting behind the use of Flow Matching for generative modeling, where the target distribution is represented by a finite dataset. In this semi-discrete regime, the exact Flow Matching velocity field is available in closed form, which makes it possible to analyze the geometry induced by the terminal flow map independently of optimization and approximation effects. We investigate the terminal assignment regions, namely the preimages of the target atoms under the terminal flow. We show that these regions are open, simply connected and, under an additional assumption, homeomorphic to the unit ball. At the same time, a planar four-point example shows that these cells can differ sharply from Laguerre cells arising in semi-discrete optimal transport: they may be non-convex, have curved boundaries, and exhibit different boundedness and adjacency patterns. These results clarify the geometry intrinsically induced by the exact semi-discrete Flow Matching objective before neural approximation enters the picture.

Class-incremental learning aims to continuously acquire new knowledge while preserving previously learned information, thereby mitigating catastrophic forgetting. Existing methods primarily restrict parameter updates but often overlook their structural properties in high-dimensional spaces. From a subspace perspective, updates induced by different tasks tend to lie in multiple overlapping low-rank subspaces, leading to cross-task subspace interference and severe forgetting. To address this issue, we propose HDSD, a Hierarchical Dual-Subspace Decoupling framework for continual learning in vision-language models. Specifically, we introduce a lightweight Feature Modulation Module (FMM) that explicitly decomposes the parameter space into general and task-specific subspaces. Building on this design, we develop two complementary components. First, a General Fusion Module (GFM) evaluates relative parameter changes across tasks and uses an adaptive threshold to capture stable and transferable knowledge. Second, a Hierarchical Learning Module (HLM) performs structured parameter decomposition via Singular Value Decomposition (SVD) and uses a scaling mechanism to constrain updates within distinct subspace scales. Together, these designs reduce subspace interference and parameter drift. Extensive experiments on conventional benchmarks show that HDSD achieves state-of-the-art results.

Existing benchmarks for multimodal agentic search evaluate multimodal search and visual browsing, but visual evidence is either confined to the input or treated as an answer endpoint rather than part of an interleaved search trajectory. We introduce \textbf{InterLV-Search}, a benchmark for Interleaved Language-Vision Agentic Search, in which textual and visual evidence is repeatedly used to condition later search. It contains 2,061 examples across three levels: active visual evidence seeking, controlled offline interleaved multimodal search, and open-web interleaved multimodal search. Beyond existing benchmarks, it also includes multimodal multi-branch samples that involve comparison between multiple entities during the evidence search. We construct Level 1 and Level 2 with automated pipelines and Level 3 with a machine-led, human-supervised open-web pipeline. We further provide InterLV-Agent for standardized tool use, trajectory logging, and evaluation. Experiments on proprietary and open-source multimodal agents show that current systems remain far from solving interleaved multimodal search, with the best model below 50% overall accuracy, highlighting challenges in visual evidence seeking, search control, and multimodal evidence integration. We release the benchmark data and evaluation code at https://github.com/hbhalpha/InterLV-Search-Bench

The exponential growth of academic publications has created an urgent need for automated tools capable of extracting structured knowledge from unstructured scientific texts. While large language models (LLMs) have demonstrated remarkable capabilities in natural language understanding and information extraction, existing solutions often require specialized infrastructure, programming expertise, or fine-tuned domain-specific models that create barriers for researchers in specialized fields. This paper presents TCMIIES, a browser-based, zero-installation platform that leverages commercial LLM APIs to perform structured information extraction from academic literature. The system employs a novel schema-guided prompting framework with automatic system prompt generation, enabling researchers to define custom extraction schemas through an intuitive graphical interface without any programming. TCMIIES features a pure front-end architecture that ensures data privacy by processing all information locally in the browser, supports five major LLM providers, implements concurrent batch processing with automatic retry mechanisms, and provides intelligent field mapping for Chinese academic databases including CNKI and Wanfang. We demonstrate the system's effectiveness through comprehensive evaluation across multiple extraction scenarios in Traditional Chinese Medicine research, achieving structured output compliance rates exceeding 94\% and information extraction accuracy comparable to domain-expert annotation. The system represents a practical, accessible solution that bridges the gap between advanced LLM capabilities and domain-specific academic information extraction needs, particularly for researchers in specialized fields who require flexible, privacy-preserving, and cost-effective extraction tools.

Developing lightweight, on-device vision-language GUI agents is essential for efficient cross-platform automated interaction. However, current on-device agents are constrained by limited model capacity, and further performance improvements remain urgently needed. Traditional Supervised Fine-Tuning (SFT) for small-scale models often leads to overfitting, catastrophic forgetting and policy rigidity, and thus fails to fully address these challenges. In this work, we propose a novel SFT-free training paradigm that significantly enhances the performance of small-scale models. We first present the initial systematic integration of generalized knowledge distillation into the GUI agent domain via Guided On-policy Distillation. By incorporating oracle reference trajectories together with a dynamic retrieval mechanism, our method reduces hallucinations and mitigates the cognitive misalignment inherent in multi-solution GUI tasks. Building on this foundation, we further introduce a Multi-solution Dual-level GRPO framework that jointly aligns macro-level subtask planning with micro-level execution matching, thereby improving exploration in long-horizon GUI agent scenarios. In addition, we construct an automated data generation pipeline to synthesize GUI task trajectories with rich multi-solution annotations. Extensive experiments show that our method achieves state-of-the-art performance among lightweight models while remaining competitive with substantially larger-scale models across all benchmarks. Ablation studies further demonstrate that structured on-policy distillation and multi-solution dual-level exploration can fully unlock the capabilities of 2B/3B scale agents, surpassing the performance limits of conventional imitation learning.

Efficiently aligning large-scale video diffusion models with human intent requires a scalable and trajectory-aware pathway that bridges the inherent discrepancy between training noise distributions and practical inference trajectories. While existing paradigms such as Direct Preference Optimization (DPO) and Group Relative Policy Optimization (GRPO) attempt to address this, they are often hindered by either reliance on bias-prone, complex reward models or suboptimal timestep sampling. In this paper, we propose Diffusion-APO (Aligned Preference Optimization), a trajectory-aware algorithm that resolves this misalignment by synchronizing training noise with inference-time denoising paths to maximize gradient signal efficacy. To translate this algorithmic innovation into a practical solution, we introduce a unified and modular RLHF framework that integrates online ranking, half-online anchoring, offline refinement, and distillation-aware drift correction. This framework enables flexible, multi-stage preference alignment across diverse data and computational constraints without relying on scalar-reward-based policy gradients. Through extensive experiments, we demonstrate that Diffusion-APO consistently outperforms standard baselines in visual quality and instruction following, while effectively preserving generative fidelity during model acceleration, providing a robust, end-to-end pathway for scalable video diffusion alignment.

Accurate four-dimensional (4D) precipitation information is essential for understanding the Earth's energy and water cycles, yet remains observationally unresolved at global scales. Conventional theory holds that geostationary infrared observations primarily sense cloud-top properties, with limited sensitivity to sub-cloud precipitation. Here we show that cloud-top infrared measurements nevertheless encode sufficient information to recover the four-dimensional structure of precipitation, revealing a previously unexploited observability of sub-cloud processes. We introduce a physically constrained deep learning framework, 4DPrecipNet, in which a moisture-first constraint requires the latent representation to recover precipitable water vapour, anchoring the model in thermodynamic consistency. By integrating multi-channel infrared radiances with these constraints and radar-derived precipitation profiles, we reconstruct the vertical and temporal evolution of precipitation systems from geostationary orbit. The framework captures deep convective structures and their evolution, with robust performance across large samples and independent radar comparisons. These results demonstrate that sub-cloud precipitation is physically encoded in cloud-top infrared observations, establishing a new pathway for continuous global monitoring of precipitation structure.

Unpaired smartphone ISP is a challenging problem due to the lack of scene and color alignment between RAW and target RGB images. Many existing methods either require paired data or rely heavily on adversarial training, which can become unstable in the unpaired setting. In this work, we present a simple and effective approach developed for the NTIRE 2026 Learned Smartphone ISP Challenge with Unpaired Data. Our method first reconstructs larger images from training patches to recover global context. Then, we extract semantic embeddings with DINOv2, and use fused Gromov-Wasserstein (FGW) optimal transport to build pseudo pairs between RAW and RGB images at both image and patch levels. This semantic matching allows us to partially alleviate the unpairedness of the data and build these pseudo input-target pairs. Based on these pseudo pairs, we train a lightweight CNN with only 7K parameters for color rendering. The network is designed to be compact and focus on color transformation rather than structural change, which helps reduce artifacts and improve training stability. Our challenge submission achieves 22.569 PSNR, 0.675 SSIM, and 8.067 $ΔE$ on the final hidden test set, significantly improving over the baseline and achieving the 3rd best SSIM and $ΔE$ among all challenge entries. Our code is available at github.com/nuniniyujin/Unpaired-ISP .

Continual learning enables vision-language models to accumulate knowledge and adapt to evolving tasks without retraining from scratch. However, in multi-domain task-incremental learning, large domain shifts intensify the stability-plasticity dilemma. Most existing methods rely on fixed architectures with statically allocated parameters, which limits adaptation to new domains and aggravates catastrophic forgetting. To address these challenges, we propose DIMoE-Adapters, a Dynamic Incremental Mixture-of-Experts Adapters framework that introduces a dynamic expert evolution paradigm to balance stability and plasticity. This paradigm is implemented through two collaborative components: Self-Calibrated Expert Evolution (SCEE) and Prototype-Guided Expert Selection (PGES). SCEE constructs and evolves a sparse expert pool through expert optimization dynamics, improving plasticity while reducing redundant capacity. PGES controls expert utilization based on the pool shaped by SCEE, improving stability across both previously encountered and unseen tasks. Extensive experiments show that DIMoE-Adapters outperforms previous state-of-the-art methods across various settings.

The past year has seen over 20 open-source document parsing models, yet thefield still benchmarks almost exclusively on OmniDocBench, a 1,355-pagemanually annotated dataset whose top scores have saturated above 90%. Athree-stage audit pipeline we run on OmniDocBench screens its 21,353evaluator-scored blocks and confirms 2,580 errors (12.08%); combined with overa year of public availability, both annotation quality and contamination riskcall its rankings into question. To address these issues, we presentPureDocBench, a programmatically generated, source-traceable benchmark thatrenders document images from HTML/CSS and produces verifiable annotations fromthe same source, covering 10 domains, 66 subcategories, and 1,475 pages, eachin three versions: clean, digitally degraded, and real-degraded (4,425 imagestotal). Evaluating 40 models spanning pipeline specialists, end-to-endspecialists, and general-purpose VLMs, we find: (i) document parsing is farfrom solved: the best model scores only ~74 out of 100, with a 44.6-point gapbetween the strongest and weakest models; (ii) specialist parsers with <=4Bparameters rival or surpass general VLMs that are 5-100x larger, yet formularecognition remains a shared bottleneck where no model exceeds 67% whenaveraging the formula metric across all three tracks; (iii) general VLMs loseonly 0.99/8.52 Overall points under digital/real degradation versus 4.90/14.21for pipeline specialists, producing ranking reversals that make clean-onlyevaluation misleading for deployment. All data, code, and artifacts arepublicly released.