We present THFM, a unified video foundation model for human-centric perception that jointly addresses dense tasks (depth, normals, segmentation, dense pose) and sparse tasks (2d/3d keypoint estimation) within a single architecture. THFM is derived from a pretrained text-to-video diffusion model, repurposed as a single-forward-pass perception model and augmented with learnable tokens for sparse predictions. Modulated by the text prompt, our single unified model is capable of performing various perception tasks. Crucially, our model is on-par or surpassing state-of-the-art specialized models on a variety of benchmarks despite being trained exclusively on synthetic data (i.e.~without training on real-world or benchmark specific data). We further highlight intriguing emergent properties of our model, which we attribute to the underlying diffusion-based video representation. For example, our model trained on videos with a single human in the scene generalizes to multiple humans and other object classes such as anthropomorphic characters and animals -- a capability that hasn't been demonstrated in the past.

Pre-trained vision-language models (VLMs) excel in multimodal tasks, commonly encoding images as embedding vectors for storage in databases and retrieval via approximate nearest neighbor search (ANNS). However, these models struggle with compositional queries and out-of-distribution (OOD) image-text pairs. Inspired by human cognition's ability to learn from minimal examples, we address this performance gap through few-shot learning approaches specifically designed for image retrieval. We introduce the Few-Shot Text-to-Image Retrieval (FSIR) task and its accompanying benchmark dataset, FSIR-BD - the first to explicitly target image retrieval by text accompanied by reference examples, focusing on the challenging compositional and OOD queries. The compositional part is divided to urban scenes and nature species, both in specific situations or with distinctive features. FSIR-BD contains 38,353 images and 303 queries, with 82% comprising the test corpus (averaging per query 37 positives, ground truth matches, and significant number of hard negatives) and 18% forming the few-shot reference corpus (FSR) of exemplar positive and hard negative images. Additionally, we propose two novel retrieval optimization methods leveraging single shot or few shot reference examples in the FSR to improve performance. Both methods are compatible with any pre-trained image encoder, making them applicable to existing large-scale environments. Our experiments demonstrate that: (1) FSIR-BD provides a challenging benchmark for image retrieval; and (2) our optimization methods outperform existing baselines as measured by mean Average Precision (mAP). Further research into FSIR optimization methods will help narrow the gap between machine and human-level understanding, particularly for compositional reasoning from limited examples.

Head-mounted devices integrated with eye tracking promise a solution for natural human-computer interaction. However, they typically require per-user calibration for optimal performance due to inter-person variability. A differential personalization approach using Siamese architectures learns relative gaze displacements and reconstructs absolute gaze from a small set of calibration frames. In this paper, we benchmark Siamese personalization on polarization-enabled eye tracking. For benchmarking, we use a 338-subject dataset captured with a polarization-sensitive camera and 850 nm illumination. We achieve performance comparable to linear calibration with 10-fold fewer samples. Using polarization inputs for Siamese personalization reduces gaze error by up to 12% compared to near-infrared (NIR)-based inputs. Combining Siamese personalization with linear calibration yields further improvements of up to 13% over a linearly calibrated baseline. These results establish Siamese personalization as a practical approach enabling accurate eye tracking.

World models (WMs) are intended to serve as internal simulators of the real world that enable agents to understand, anticipate, and act upon complex environments. Existing WM benchmarks remain narrowly focused on next-state prediction and visual fidelity, overlooking the richer simulation capabilities required for intelligent behavior. To address this gap, we introduce WR-Arena, a comprehensive benchmark for evaluating WMs along three fundamental dimensions of next world simulation: (i) Action Simulation Fidelity, the ability to interpret and follow semantically meaningful, multi-step instructions and generate diverse counterfactual rollouts; (ii) Long-horizon Forecast, the ability to sustain accurate, coherent, and physically plausible simulations across extended interactions; and (iii) Simulative Reasoning and Planning, the ability to support goal-directed reasoning by simulating, comparing, and selecting among alternative futures in both structured and open-ended environments. We build a task taxonomy and curate diverse datasets designed to probe these capabilities, moving beyond single-turn and perceptual evaluations. Through extensive experiments with state-of-the-art WMs, our results expose a substantial gap between current models and human-level hypothetical reasoning, and establish WR-Arena as both a diagnostic tool and a guideline for advancing next-generation world models capable of robust understanding, forecasting, and purposeful action. The code is available at https://github.com/MBZUAI-IFM/WR-Arena.

Blind Sweep Obstetric Ultrasound (BSOU) enables scalable fetal imaging in low-resource settings by allowing minimally trained operators to acquire standardized sweep videos for automated Artificial Intelligence(AI) interpretation. However, the reliability of such AI systems depends critically on the quality of the acquired sweeps, and little is known about how deviations from the intended protocol affect downstream predictions. In this work, we present a systematic evaluation of BSOU quality and its impact on three key AI tasks: sweep-tag classification, fetal presentation classification, and placenta-location classification. We simulate plausible acquisition deviations, including reversed sweep direction, probe inversion, and incomplete sweeps, to quantify model robustness, and we develop automated quality-assessment models capable of detecting these perturbations. To approximate real-world deployment, we simulate a feedback loop in which flagged sweeps are re-acquired, showing that such correction improves downstream task performance. Our findings highlight the sensitivity of BSOU-based AI models to acquisition variability and demonstrate that automated quality assessment can play a central role in building reliable, scalable AI-assisted prenatal ultrasound workflows, particularly in low-resource environments.

Diffusion models have achieved remarkable success in generating high-fidelity content but suffer from slow, iterative sampling, resulting in high latency that limits their use in interactive applications. We introduce DRiffusion, a parallel sampling framework that parallelizes diffusion inference through a draft-and-refine process. DRiffusion employs skip transitions to generate multiple draft states for future timesteps and computes their corresponding noises in parallel, which are then used in the standard denoising process to produce refined results. Theoretically, our method achieves an acceleration rate of $\tfrac{1}{n}$ or $\tfrac{2}{n+1}$, depending on whether the conservative or aggressive mode is used, where $n$ denotes the number of devices. Empirically, DRiffusion attains 1.4$\times$-3.7$\times$ speedup across multiple diffusion models while incur minimal degradation in generation quality: on MS-COCO dataset, both FID and CLIP remain largely on par with those of the original model, while PickScore and HPSv2.1 show only minor average drops of 0.17 and 0.43, respectively. These results verify that DRiffusion delivers substantial acceleration and preserves perceptual quality.

Creating whiteboard-style educational videos demands precise coordination between freehand illustrations and spoken narration, yet no existing method addresses this multimodal synchronization problem with structured, reproducible drawing representations. We present the first dataset of 24 paired Excalidraw demonstrations with narrated audio, where every drawing element carries millisecond-precision creation timestamps spanning 8 STEM domains. Using this data, we study whether a vision-language model (Qwen2-VL-7B), fine-tuned via LoRA, can predict full stroke sequences synchronized to speech from only 24 demonstrations. Our topic-stratified five-fold evaluation reveals that timestamp conditioning significantly improves temporal alignment over ablated baselines, while the model generalizes across unseen STEM topics. We discuss transferability to real classroom settings and release our dataset and code to support future research in automated educational content generation.

Minimally invasive and robot-assisted surgery relies heavily on endoscopic imaging, yet surgical smoke produced by electrocautery and vessel-sealing instruments can severely degrade visual perception and hinder vision-based functionalities. We present a transformer-based surgical desmoking model with a physics-inspired desmoking head that jointly predicts smoke-free image and corresponding smoke map. To address the scarcity of paired smoky-to-smoke-free training data, we develop a synthetic data generation pipeline that blends artificial smoke patterns with real endoscopic images, yielding over 80,000 paired samples for supervised training. We further curate, to our knowledge, the largest paired surgical smoke dataset to date, comprising 5,817 image pairs captured with the da Vinci robotic surgical system, enabling benchmarking on high-resolution endoscopic images. Extensive experiments on both a public benchmark and our dataset demonstrate state-of-the-art performance in image reconstruction compared to existing dehazing and desmoking approaches. We also assess the impact of desmoking on downstream stereo depth estimation and instrument segmentation, highlighting both the potential benefits and current limitations of digital smoke removal methods.

Graphical User Interface (GUI) agents have the potential to assist users in interacting with complex software (e.g., PowerPoint, Photoshop). While prior research has primarily focused on automating user actions through clicks and keystrokes, this paradigm overlooks human intention, where users value the ability to explore, iterate, and refine their ideas while maintaining agency. To move beyond automation and toward collaboration, GUI agents must understand what users are doing and why. We introduce GUIDE (GUI User Intent Detection Evaluation), a benchmark that evaluates AI models on their ability to perceive user behavior, infer intent, and provide assistance in open-ended GUI tasks. GUIDE consists of 67.5 hours of screen recordings from 120 novice user demonstrations with think-aloud narrations, across 10 software. GUIDE defines three tasks - (i) Behavior State Detection, (ii) Intent Prediction, and (iii) Help Prediction that test a model's ability to recognize behavior state, reason about goals, and decide when and how to help. Evaluations across eight state-of-the-art multimodal models reveal that all models struggled, achieving only 44.6% and 55.0% accuracy on behavior state and help prediction. However, providing user context significantly improved the performance, raising help prediction by up to 50.2pp, highlighting the critical role of structured user understanding in effective assistance. Our dataset is available at https://guide-bench.github.io.

This paper proposes a method for dynamic hand gesture recognition based on the composition of two models: the MediaPipe Hand Landmarker, responsible for extracting 21 skeletal keypoints of the hand, and a convolutional neural network (CNN) trained to classify gestures from a spatiotemporal matrix representation of dimensions 90 by 21 of those keypoints. The method is applied to the recognition of LIBRAS (Brazilian Sign Language) gestures for device control in a home automation system, covering 11 classes of static and dynamic gestures. For real-time inference, a sliding window with temporal frame triplication is used, enabling continuous recognition without recurrent networks. Tests achieved 95\% accuracy under low-light conditions and 92\% under normal lighting. The results indicate that the approach is effective, although systematic experiments with greater user diversity are needed for a more thorough evaluation of generalization.

Virtually every sector of society is experiencing a dramatic growth in the volume of unstructured textual data that is generated and published, from news and social media online interactions, through open access scholarly communications and observational data in the form of digital health records and online drug reviews. The volume and variety of data across all this range of domains has created both unprecedented opportunities and pressing challenges for extracting actionable knowledge for several application scenarios. However, the extraction of rich semantic knowledge demands the deployment of scalable and flexible automatic methods adaptable across text genres and schema specifications. Moreover, the full potential of these data can only be unlocked by coupling information extraction methods with Semantic Web techniques for the construction of full-fledged Knowledge Graphs, that are semantically transparent, explainable by design and interoperable. In this thesis, we experiment with the application of Natural Language Processing, Machine Learning and Generative AI methods, powered by Semantic Web best practices, to the automatic construction of Knowledge Graphs from large text corpora, in three use case applications: the analysis of the Digital Transformation discourse in the global news and social media platforms; the mapping and trend analysis of recent research in the Architecture, Engineering, Construction and Operations domain from a large corpus of publications; the generation of causal relation graphs of biomedical entities from electronic health records and patient-authored drug reviews. The contributions of this thesis to the research community are in terms of benchmark evaluation results, the design of customized algorithms and the creation of data resources in the form of Knowledge Graphs, together with data analysis results built on top of them.

Activation-based probes have emerged as a promising approach for detecting deceptively aligned AI systems by identifying internal conflict between true and stated goals. We identify a fundamental blind spot: probes fail on coherent misalignment - models that believe their harmful behavior is virtuous rather than strategically hiding it. We prove that no polynomial-time probe can detect such misalignment with non-trivial accuracy when belief structures reach sufficient complexity (PRF-like triggers). We show the emergence of this phenomenon on a simple task by training two models with identical RLHF procedures: one producing direct hostile responses ("the Liar"), another trained towards coherent misalignment using rationalizations that frame hostility as protective ("the Fanatic"). Both exhibit identical behavior, but the Liar is detected 95%+ of the time while the Fanatic evades detection almost entirely. We term this Emergent Probe Evasion: training with belief-consistent reasoning shifts models from a detectable "deceptive" regime to an undetectable "coherent" regime - not by learning to hide, but by learning to believe.

The capabilities of large language models (LLMs) have expanded beyond natural language processing to scientific prediction tasks, including molecular property prediction. However, their effectiveness in in-context learning remains ambiguous, particularly given the potential for training data contamination in widely used benchmarks. This paper investigates whether LLMs perform genuine in-context regression on molecular properties or rely primarily on memorized values. Furthermore, we analyze the interplay between pre-trained knowledge and in-context information through a series of progressively blinded experiments. We evaluate nine LLM variants across three families (GPT-4.1, GPT-5, Gemini 2.5) on three MoleculeNet datasets (Delaney solubility, Lipophilicity, QM7 atomization energy) using a systematic blinding approach that iteratively reduces available information. Complementing this, we utilize varying in-context sample sizes (0-, 60-, and 1000-shot) as an additional control for information access. This work provides a principled framework for evaluating molecular property prediction under controlled information access, addressing concerns regarding memorization and exposing conflicts between pre-trained knowledge and in-context information.

Genome-Wide Association Studies (GWAS) identify associations between genetic variants and disease; however, moving beyond associations to causal mechanisms is critical for therapeutic target prioritization. The recently proposed Knowledge Graph GWAS (KGWAS) framework addresses this challenge by linking genetic variants to downstream gene-gene interactions via a knowledge graph (KG), thereby improving detection power and providing mechanistic insights. However, the original KGWAS implementation relies on a large general-purpose KG, which can introduce spurious correlations. We hypothesize that cell-type specific KGs from disease-relevant cell types will better support disease mechanism discovery. Here, we show that the general-purpose KG in KGWAS can be substantially pruned with no loss of statistical power on downstream tasks, and that performance further improves by incorporating gene-gene relationships derived from perturb-seq data. Importantly, using a sparse, context-specific KG from direct perturb-seq evidence yields more consistent and biologically robust disease-critical networks.

Current multimodal large language models (MLLMs) cannot effectively utilize eye-gaze information for video understanding, even when gaze cues are supplied via visual overlays or text descriptions. We introduce GazeQwen, a parameter efficient approach that equips an open-source MLLM with gaze awareness through hidden-state modulation. At its core is a compact gaze resampler (~1-5 M trainable parameters) that encodes V-JEPA 2.1 video features together with fixation-derived positional encodings and produces additive residuals injected into selected LLM decoder layers via forward hooks. An optional second training stage adds low-rank adapters (LoRA) to the LLM for tighter integration. Evaluated on all 10 tasks of the StreamGaze benchmark, GazeQwen reaches 63.9% accuracy, a +16.1 point gain over the same Qwen2.5-VL-7B backbone with gaze as visual prompts and +10.5 points over GPT-4o, the highest score among all open-source and proprietary models tested. These results suggest that learning where to inject gaze within an LLM is more effective than scaling model size or engineering better prompts. All code and checkpoints are available at https://github.com/phamtrongthang123/gazeqwen .

Deep neural networks exhibit a simplicity bias, a well-documented tendency to favor simple functions over complex ones. In this work, we cast new light on this phenomenon through the lens of the Minimum Description Length principle, formalizing supervised learning as a problem of optimal two-part lossless compression. Our theory explains how simplicity bias governs feature selection in neural networks through a fundamental trade-off between model complexity (the cost of describing the hypothesis) and predictive power (the cost of describing the data). Our framework predicts that as the amount of available training data increases, learners transition through qualitatively different features -- from simple spurious shortcuts to complex features -- only when the reduction in data encoding cost justifies the increased model complexity. Consequently, we identify distinct data regimes where increasing data promotes robustness by ruling out trivial shortcuts, and conversely, regimes where limiting data can act as a form of complexity-based regularization, preventing the learning of unreliable complex environmental cues. We validate our theory on a semi-synthetic benchmark showing that the feature selection of neural networks follows the same trajectory of solutions as optimal two-part compressors.

In low-resource multilingual speech-to-text translation, uniform architectural sharing across languages frequently introduces representation conflicts that impede convergence. This work proposes a principled methodology to automatically determine layer-specific sharing patterns by mining training gradient information. Our approach employs three distinct analysis strategies: distance-based language clustering, self/cross-task divergence metrics for capacity allocation, and joint factorization coupled with canonical correlation analysis for subspace alignment. Extensive evaluation across four language pairs (using the SeamlessM4T-Medium architecture) demonstrates persistent improvements in translation quality metrics.

Recent advances in parallel computing and GPU acceleration have created new opportunities for computation-intensive learning problems such as Active SLAM -- where actions are selected to reduce uncertainty and improve joint mapping and localization. However, existing DRL-based approaches remain constrained by the lack of scalable parallel training. In this work, we address this challenge by proposing a scalable end-to-end DRL framework for Active SLAM that enables massively parallel training. Compared with the state of the art, our method significantly reduces training time, supports continuous action spaces and facilitates the exploration of more realistic scenarios. It is released as an open-source framework to promote reproducibility and community adoption.

We propose a feed-forward method for dense Signed Distance Field (SDF) regression from unstructured image collections in less than three seconds, without camera calibration or post-hoc fusion. Our key insight is that the intermediate feature space of pretrained multi-view feed-forward geometry transformers already encodes a powerful joint world representation; yet, existing pipelines discard it, routing features through per-view prediction heads before assembling 3D geometry post-hoc, which discards valuable completeness information and accumulates inaccuracies. We instead perform 3D extraction directly from geometry transformer features via learned volumetric extraction: voxelized canonical embeddings that progressively absorb multi-view geometry information through interleaved cross- and self-attention into a structured volumetric latent grid. A simple convolutional decoder then maps this grid to a dense SDF. We additionally propose a scalable, validity-aware supervision scheme directly using SDFs derived from depth maps or 3D assets, tackling practical issues like non-watertight meshes. Our approach yields complete and well-defined distance values across sparse- and dense-view settings and demonstrates geometrically plausible completions. Code and further material can be found at https://lorafib.github.io/fus3d.

Beneath the stunning visual fidelity of modern AIGC models lies a "logical desert", where systems fail tasks that require physical, causal, or complex spatial reasoning. Current evaluations largely rely on superficial metrics or fragmented benchmarks, creating a ``performance mirage'' that overlooks the generative process. To address this, we introduce ViGoR Vision-G}nerative Reasoning-centric Benchmark), a unified framework designed to dismantle this mirage. ViGoR distinguishes itself through four key innovations: 1) holistic cross-modal coverage bridging Image-to-Image and Video tasks; 2) a dual-track mechanism evaluating both intermediate processes and final results; 3) an evidence-grounded automated judge ensuring high human alignment; and 4) granular diagnostic analysis that decomposes performance into fine-grained cognitive dimensions. Experiments on over 20 leading models reveal that even state-of-the-art systems harbor significant reasoning deficits, establishing ViGoR as a critical ``stress test'' for the next generation of intelligent vision models. The demo have been available at https://vincenthancoder.github.io/ViGoR-Bench/

We present Doctorina MedBench, a comprehensive evaluation framework for agent-based medical AI based on the simulation of realistic physician-patient interactions. Unlike traditional medical benchmarks that rely on solving standardized test questions, the proposed approach models a multi-step clinical dialogue in which either a physician or an AI system must collect medical history, analyze attached materials (including laboratory reports, images, and medical documents), formulate differential diagnoses, and provide personalized recommendations. System performance is evaluated using the D.O.T.S. metric, which consists of four components: Diagnosis, Observations/Investigations, Treatment, and Step Count, enabling assessment of both clinical correctness and dialogue efficiency. The system also incorporates a multi-level testing and quality monitoring architecture designed to detect model degradation during both development and deployment. The framework supports safety-oriented trap cases, category-based random sampling of clinical scenarios, and full regression testing. The dataset currently contains more than 1,000 clinical cases covering over 750 diagnoses. The universality of the evaluation metrics allows the framework to be used not only to assess medical AI systems, but also to evaluate physicians and support the development of clinical reasoning skills. Our results suggest that simulation of clinical dialogue may provide a more realistic assessment of clinical competence compared to traditional examination-style benchmarks.

Cross-view geo-spatial learning consists of two important tasks: Cross-View Geo-Localization (CVGL) and Cross-View Image Synthesis (CVIS), both of which rely on establishing geometric correspondences between ground and aerial views. Recent Geometric Foundation Models (GFMs) have demonstrated strong capabilities in extracting generalizable 3D geometric features from images, but their potential in cross-view geo-spatial tasks remains underexplored. In this work, we present Geo^2, a unified framework that leverages Geometric priors from GFMs (e.g., VGGT) to jointly perform geo-spatial tasks, CVGL and bidirectional CVIS. Despite the 3D reconstruction ability of GFMs, directly applying them to CVGL and CVIS remains challenging due to the large viewpoint gap between ground and aerial imagery. We propose GeoMap, which embeds ground and aerial features into a shared 3D-aware latent space, effectively reducing cross-view discrepancies for localization. This shared latent space naturally bridges cross-view image synthesis in both directions. To exploit this, we propose GeoFlow, a flow-matching model conditioned on geometry-aware latent embeddings. We further introduce a consistency loss to enforce latent alignment between the two synthesis directions, ensuring bidirectional coherence. Extensive experiments on standard benchmarks, including CVUSA, CVACT, and VIGOR, demonstrate that Geo^2 achieves state-of-the-art performance in both localization and synthesis, highlighting the effectiveness of 3D geometric priors for cross-view geo-spatial learning.

We present MAGNET (Model Autonomously Growing Network), a decentralized system for autonomous generation, training, and serving of domain-expert language models across commodity hardware. MAGNET integrates four components: (1) autoresearch, an autonomous ML research pipeline that automates dataset generation, hyperparameter exploration, evaluation, and error-driven iteration; (2) BitNet b1.58 ternary training, enabling CPU-native inference via bitnet.cpp without GPU hardware; (3) DiLoCo-based distributed merging for communication-efficient aggregation of domain specialists; and (4) on-chain contribution tracking on the HOOTi EVM chain. We validate autoresearch through three case studies: video safety classification (balanced accuracy 0.9287 to 0.9851), cryptocurrency directional prediction (41% to 54.9% hit rate), and BitNet hyperparameter optimization (10-phase sweep, -16.7% validation loss).

Vision-Language Models (VLMs) have demonstrated impressive capabilities in code generation across various domains. However, their ability to replicate complex, multi-panel visualizations from real-world data remains largely unassessed. To address this gap, we introduce \textbf{\texttt{RealChart2Code}}, a new large-scale benchmark with over 2,800 instances grounded in authentic datasets and featuring tasks with clear analytical intent. Crucially, it is the first benchmark to systematically evaluate chart generation from large-scale raw data and assess iterative code refinement in a multi-turn conversational setting. Our comprehensive evaluation of 14 leading VLMs on \texttt{RealChart2Code} reveals significant performance degradation compared to simpler benchmarks, highlighting their struggles with complex plot structures and authentic data. Our analysis uncovers a substantial performance gap between proprietary and open-weight models and confirms that even state-of-the-art VLMs often fail to accurately replicate intricate, multi-panel charts. These findings provide valuable insights into the current limitations of VLMs and guide future research directions. We release the benchmark and code at \url{https://github.com/Speakn0w/RealChart2Code}.

Training Vision Transformers (ViTs) presents significant challenges, one of which is the emergence of artifacts in attention maps, hindering their interpretability. Darcet et al. (2024) investigated this phenomenon and attributed it to the need of ViTs to store global information beyond the [CLS] token. They proposed a novel solution involving the addition of empty input tokens, named registers, which successfully eliminate artifacts and improve the clarity of attention maps. In this work, we reproduce the findings of Darcet et al. (2024) and evaluate the generalizability of their claims across multiple models, including DINO, DINOv2, OpenCLIP, and DeiT3. While we confirm the validity of several of their key claims, our results reveal that some claims do not extend universally to other models. Additionally, we explore the impact of model size, extending their findings to smaller models. Finally, we untie terminology inconsistencies found in the original paper and explain their impact when generalizing to a wider range of models.

Computational pathology relies on effective representation learning to support cancer research and precision medicine. Although self-supervised learning has driven major progress at the patch and whole-slide image levels, representation learning at the single-cell level remains comparatively underexplored, despite its importance for characterizing cell types and cellular phenotypes. We introduce LEMON (Learning Embeddings from Morphology Of Nuclei), a self-supervised foundation model for scalable single-cell image representation learning. Trained on millions of cell images from diverse tissues and cancer types, LEMON learns robust and versatile morphological representations that support large-scale single-cell analyses in pathology. We evaluate LEMON on five benchmark datasets across a range of prediction tasks and show that it provides strong performance, highlighting its potential as a new paradigm for cell-level computational pathology. Model weights are available at https://huggingface.co/aliceblondel/LEMON.

We present Feature-Align CNN (FA-CNN), a prototype CNN architecture with intrinsic class attribution through end-to-end feature alignment. Our intuition is that the use of unordered operations such as Linear and Conv2D layers cause unnecessary shuffling and mixing of semantic concepts, thereby making raw feature maps difficult to understand. We introduce two new order preserving layers, the dampened skip connection, and the global average pooling classifier head. These layers force the model to maintain an end-to-end feature alignment from the raw input pixels all the way to final class logits. This end-to-end alignment enhances the interpretability of the model by allowing the raw feature maps to intrinsically exhibit class attribution. We prove theoretically that FA-CNN penultimate feature maps are identical to Grad-CAM saliency maps. Moreover, we prove that these feature maps slowly morph layer-by-layer over network depth, showing the evolution of features through network depth toward penultimate class activations. FA-CNN performs well on benchmark image classification datasets. Moreover, we compare the averaged FA-CNN raw feature maps against Grad-CAM and permutation methods in a percent pixels removed interpretability task. We conclude this work with a discussion and future, including limitations and extensions toward hybrid models.

Existing hand-object interactions (HOI) methods are largely limited to rigid objects, while 4D reconstruction methods of articulated objects generally require pre-scanning the object or even multi-view videos. It remains an unexplored but significant challenge to reconstruct 4D human-articulated-object interactions from a single monocular RGB video. Fortunately, recent advancements in foundation models present a new opportunity to address this highly ill-posed problem. To this end, we introduce ArtHOI, an optimization-based framework that integrates and refines priors from multiple foundation models. Our key contribution is a suite of novel methodologies designed to resolve the inherent inaccuracies and physical unreality of these priors. In particular, we introduce an Adaptive Sampling Refinement (ASR) method to optimize object's metric scale and pose for grounding its normalized mesh in world space. Furthermore, we propose a Multimodal Large Language Model (MLLM) guided hand-object alignment method, utilizing contact reasoning information as constraints of hand-object mesh composition optimization. To facilitate a comprehensive evaluation, we also contribute two new datasets, ArtHOI-RGBD and ArtHOI-Wild. Extensive experiments validate the robustness and effectiveness of our ArtHOI across diverse objects and interactions. Project: https://arthoi-reconstruction.github.io.

Groundwater represents a key element of the water cycle, yet it exhibits intricate and context-dependent relationships that make its modeling a challenging task. Theory-based models have been the cornerstone of scientific understanding. However, their computational demands, simplifying assumptions, and calibration requirements limit their use. In recent years, data-driven models have emerged as powerful alternatives. In particular, deep learning has proven to be a leading approach for its design flexibility and ability to learn complex relationships. We proposed an attention-based pure deep learning model, named STAINet, to predict weekly groundwater levels at an arbitrary and variable number of locations, leveraging both spatially sparse groundwater measurements and spatially dense weather information. Then, to enhance the model's trustworthiness and generalization ability, we considered different physics-guided strategies to inject the groundwater flow equation into the model. Firstly, in the STAINet-IB, by introducing an inductive bias, we also estimated the governing equation components. Then, by adopting a learning bias strategy, we proposed the STAINet-ILB, trained with additional loss terms adding supervision on the estimated equation components. Lastly, we developed the STAINet-ILRB, leveraging the groundwater body recharge zone information estimated by domain experts. The STAINet-ILB performed the best, achieving overwhelming test performances in a rollout setting (median MAPE 0.16%, KGE 0.58). Furthermore, it predicted sensible equation components, providing insights into the model's physical soundness. Physics-guided approaches represent a promising opportunity to enhance both the generalization ability and the trustworthiness, thereby paving the way to a new generation of disruptive hybrid deep learning Earth system models.

Endoscopic video analysis is essential for early gastrointestinal screening but remains hindered by limited high-quality annotations. While self-supervised video pre-training shows promise, existing methods developed for natural videos prioritize dense spatio-temporal modeling and exhibit motion bias, overlooking the static, structured semantics critical to clinical decision-making. To address this challenge, we propose Focus-to-Perceive Representation Learning (FPRL), a cognition-inspired hierarchical framework that emulates clinical examination. FPRL first focuses on intra-frame lesion-centric regions to learn static semantics, and then perceives their evolution across frames to model contextual semantics. To achieve this, FPRL employs a hierarchical semantic modeling mechanism that explicitly distinguishes and collaboratively learns both types of semantics. Specifically, it begins by capturing static semantics via teacher-prior adaptive masking (TPAM) combined with multi-view sparse sampling. This approach mitigates redundant temporal dependencies and enables the model to concentrate on lesion-related local semantics. Following this, contextual semantics are derived through cross-view masked feature completion (CVMFC) and attention-guided temporal prediction (AGTP). These processes establish cross-view correspondences and effectively model structured inter-frame evolution, thereby reinforcing temporal semantic continuity while preserving global contextual integrity. Extensive experiments on 11 endoscopic video datasets show that FPRL achieves superior performance across diverse downstream tasks, demonstrating its effectiveness in endoscopic video representation learning. The code is available at https://github.com/MLMIP/FPRL.