We present a clustering-based explainability technique for digital pathology models based on convolutional neural networks. Unlike commonly used methods based on saliency maps, such as occlusion, GradCAM, or relevance propagation, which highlight regions that contribute the most to the prediction for a single slide, our method shows the global behaviour of the model under consideration, while also providing more fine-grained information. The result clusters can be visualised not only to understand the model, but also to increase confidence in its operation, leading to faster adoption in clinical practice. We also evaluate the performance of our technique on an existing model for detecting prostate cancer, demonstrating its usefulness.

We present a novel decentralized algorithm for coverage control in unknown spatial environments modeled by Gaussian Processes (GPs). To trade-off between exploration and exploitation, each agent autonomously determines its trajectory by minimizing a local cost function. Inspired by the GP-UCB (Upper Confidence Bound for GPs) acquisition function, the proposed cost combines the expected locational cost with a variance-based exploration term, guiding agents toward regions that are both high in predicted density and model uncertainty. Compared to previous work, our algorithm operates in a fully decentralized fashion, relying only on local observations and communication with neighboring agents. In particular, agents periodically update their inducing points using a greedy selection strategy, enabling scalable online GP updates. We demonstrate the effectiveness of our algorithm in simulation.

Autonomous graphical user interface (GUI) agents rely on accurate GUI grounding, which maps language instructions to on-screen coordinates, to execute user commands. However, current models, whether trained via supervised fine-tuning (SFT) or reinforcement learning (RL), often provide confidence signals that are poorly aligned with actual grounding correctness, leading to overconfident and unreliable predictions. To address this, we propose HyperClick, a novel framework that enhances trustworthy GUI grounding through self-critiqued reinforcement learning (SCRL). HyperClick combines a correctness reward and a confidence alignment reward, training the policy model to output both a click prediction and an explicit confidence estimate. This approach jointly optimizes grounding accuracy and confidence reliability through confidence-based self-assessment. Extensive experiments on challenging benchmarks show that HyperClick maintains strong grounding performance while providing better-aligned confidence estimates. By exposing uncertainty alongside GUI actions, HyperClick supports confidence-based abstention in GUI automation. Code will be released here.

Selecting an appropriate evaluation metric for classifiers is crucial for model comparison, parameter optimization, and deployment decisions, yet there is no consensus on a broadly accepted evaluation paradigm explicitly aligned with Total Classification Cost (TCC) minimization. At the same time, class imbalance is often treated as a problem to be corrected \emph{per se}, potentially causing misalignments with TCC minimization. To address these limitations, (\emph{i}) we define Weighted Accuracy (WA), an evaluation metric for binary classifiers with a straightforward interpretation as a weighted version of accuracy and (\emph{ii}) we propose a general reweighting framework for handling class imbalance in cost-sensitive scenarios, providing an alternative to resampling techniques. This framework applies to any evaluation metric or loss function that can be expressed as a linear combination of example-dependent quantities; it enables meaningful comparison of evaluation results obtained on different datasets and accounts for discrepancies between the \emph{development} dataset, used for training, validation, and testing, and the \emph{target} dataset, where the model will be deployed. Within this framework, we derive the conditions under which standard rebalancing techniques remain coherent with TCC minimization, and when they may instead become misleading. We prove that, under example-independent Unit Classification Costs, maximizing WA is equivalent to minimizing TCC. Finally, we analyze the robustness of WA in realistic example-dependent cost scenarios by studying its correlation with TCC across a broad range of class imbalance and cost regimes. The results show that WA maintains robust alignment with TCC across almost all examined scenarios.

This book presents the core principles that have guided the development of diffusion models, tracing their origins and showing how diverse formulations arise from shared mathematical ideas. Diffusion modeling starts by defining a forward process that gradually corrupts data into noise, linking the data distribution to a simple prior through a continuum of intermediate distributions. The goal is to learn a reverse process that transforms noise back into data while recovering the same intermediates. We describe three complementary views. The variational view, inspired by variational autoencoders, sees diffusion as learning to remove noise step by step. The score-based view, rooted in energy-based modeling, learns the gradient of the evolving data distribution, indicating how to nudge samples toward more likely regions. The flow-based view, related to normalizing flows, treats generation as following a smooth path that moves samples from noise to data under a learned velocity field. These perspectives share a common backbone: a time-dependent velocity field whose flow transports a simple prior to the data. Sampling then amounts to solving a differential equation that evolves noise into data along a continuous trajectory. On this foundation, the book discusses guidance for controllable generation, efficient numerical solvers, and diffusion-motivated flow-map models that learn direct mappings between arbitrary times. It provides a conceptual and mathematically grounded understanding of diffusion models for readers with basic deep-learning knowledge.

Accurate long-term localization using onboard sensors is crucial for robots operating in Global Navigation Satellite System (GNSS)-denied environments. While complementary sensors mitigate individual degradations, carrying all the available sensor types on a single robot significantly increases the size, weight, and power demands. Distributing sensors across multiple robots enhances the deployability but introduces challenges in fusing asynchronous, multi-modal data from independently moving platforms. We propose a novel adaptive multi-modal multi-robot cooperative localization approach using a factor-graph formulation to fuse asynchronous Visual-Inertial Odometry (VIO), LiDAR-Inertial Odometry (LIO), and 3D inter-robot detections from distinct robots in a loosely-coupled fashion. The approach adapts to changing conditions, leveraging reliable data to assist robots affected by sensory degradations. A novel interpolation-based factor enables fusion of the unsynchronized measurements. LIO degradations are evaluated based on the approximate scan-matching Hessian. A novel approach of weighting odometry data proportionally to the Wasserstein distance between the consecutive VIO outputs is proposed. A theoretical analysis is provided, investigating the cooperative localization problem under various conditions, mainly in the presence of sensory degradations. The proposed method has been extensively evaluated on real-world data gathered with heterogeneous teams of an Unmanned Ground Vehicle (UGV) and Unmanned Aerial Vehicles (UAVs), showing that the approach provides significant improvements in localization accuracy in the presence of various sensory degradations.

The vast majority of cardiovascular diseases may be preventable if early signs and risk factors are detected. Cardiovascular monitoring with body-worn sensor devices like sensor patches allows for the detection of such signs while preserving the freedom and comfort of patients. However, the analysis of the sensor data must be robust, reliable, efficient, and highly accurate. Deep learning methods can automate data interpretation, reducing the workload of clinicians. In this work, we analyze the feasibility of applying deep learning models to the classification of synchronized electrocardiogram (ECG) and phonocardiogram (PCG) recordings on resource-constrained medical edge devices. We propose a convolutional neural network with early fusion of data to solve a binary classification problem. The model is trained and validated on the synchronized ECG and PCG recordings from the Physionet Challenge 2016 dataset. Our approach reduces memory footprint and compute cost by approximately three orders of magnitude compared with the state-of-the-art while maintaining competitive accuracy. We further demonstrate the applicability of the proposed model on medical edge devices by measuring its energy consumption on a microcontroller equipped with a neural processing unit (NPU) and benchmarking the energy of Bluetooth Low Energy (BLE) communication on a representative BLE evaluation kit across a range of payload sizes. The comparison confirms that on-device inference can be more energy efficient than continuous data streaming.

Large language models may encode sensitive information or outdated knowledge that needs to be removed, to ensure responsible and compliant model responses. Unlearning has emerged as an efficient alternative to full retraining, aiming to remove specific knowledge while preserving overall model utility. Existing evaluations of unlearning methods focus on (1) the extent of forgetting of the target knowledge (forget set) and (2) maintaining performance on the retain set (i.e., utility). However, these evaluations overlook an important usability aspect: users may still want the model to leverage the removed information if it is re-introduced in the prompt. In a systematic evaluation of six state-of-the-art unlearning methods, we find that they consistently impair such contextual utility. To address this, we augment unlearning objectives with a plug-in term that preserves the model's ability to use forgotten knowledge when it is present in context. Extensive experiments demonstrate that our approach restores contextual utility to near original levels while still maintaining effective forgetting and retain-set utility.

Random Utility Models (RUMs) are a classical framework for modeling user preferences and play a key role in reward modeling for Reinforcement Learning from Human Feedback (RLHF). However, a crucial shortcoming of many of these techniques is the Independence of Irrelevant Alternatives (IIA) assumption, which collapses \emph{all} human preferences to a universal underlying utility function, yielding a coarse approximation of the range of human preferences. On the other hand, statistical and computational guarantees for models avoiding this assumption are scarce. In this paper, we investigate the statistical and computational challenges of learning a \emph{correlated} probit model, a fundamental RUM that avoids the IIA assumption. First, we establish that the classical data collection paradigm of pairwise preference data is \emph{fundamentally insufficient} to learn correlational information, explaining the lack of statistical and computational guarantees in this setting. Next, we demonstrate that \emph{best-of-three} preference data provably overcomes these shortcomings, and devise a statistically and computationally efficient estimator with near-optimal performance. These results highlight the benefits of higher-order preference data in learning correlated utilities, allowing for more fine-grained modeling of human preferences. Finally, we validate these theoretical guarantees on several real-world datasets, demonstrating improved personalization of human preferences.

Accurate brain tumor segmentation from MRI is vital for diagnosis and treatment planning. Although Monte Carlo (MC) Dropout is widely used to estimate model uncertainty, the effectiveness of variance-based uncertainty - computed as pixel-wise variance across stochastic forward passes - in identifying segmentation errors, particularly near tumor boundaries, remains insufficiently studied. This study empirically examines the relationship between variance-based MC Dropout uncertainty and segmentation error in 2D brain tumor MRI segmentation using a U-Net trained under four augmentation settings: none, horizontal flip, rotation, and scaling. Uncertainty was estimated as the pixel-wise variance across 50 stochastic forward passes and correlated with pixel-wise errors using Pearson and Spearman coefficients. Results show weak global correlations (r ~ 0.30-0.38) and negligible boundary correlations (|r| < 0.05). Although differences across augmentations were statistically significant (p < 0.001), they lacked practical relevance. These findings suggest that variance-based MC Dropout uncertainty provides limited cues for global and boundary error localization, and that the choice of uncertainty representation critically affects the utility of MC Dropout in medical image segmentation. Alternative representations such as predictive entropy or mutual information may better capture segmentation errors, particularly at boundaries.

With the rise of reasoning language models and test-time scaling methods as a paradigm for improving model performance, substantial computation is often required to generate multiple candidate sequences from the same prompt. This enables exploration of different reasoning paths toward the correct solution, however, allocates the same compute budget for each prompt. Grounded on the assumption that different prompts carry different degrees of complexity, and thus different computation needs, we propose EAGer, a training-free generation method that leverages model uncertainty through token-wise entropy distribution to reduce redundant computation and concurrently improve overall performance. EAGer allows branching to multiple reasoning paths only in the presence of high-entropy tokens, and reallocates the saved compute budget to instances where exploration of alternative paths is most needed. We validate EAGer across multiple open-source models on complex reasoning benchmarks, with gains specifically demonstrated on AIME 2025. When target labels are accessible -- as in RLVR training pipelines -- EAGer achieves up to +37% in Pass@k and 59% fewer tokens; in test-time settings it still yields +12% in Pass@k and 64% fewer tokens compared to Full Parallel Sampling.

How should we analyze memory in deep RL? We introduce tools for analyzing policies under partial observability and revealing how agents use memory to make decisions. To utilize these tools, we present POPGym Arcade, a collection of Atari-inspired, hardware-accelerated environments sharing a single observation and action space. Each environment provides fully and partially observable variants, enabling counterfactual studies on observability. We find that controlled studies are necessary for fair comparisons and identify a pathology where value functions smear credit over irrelevant history. Using this pathology, we demonstrate how out-of-distribution scenarios can contaminate memory, perturbing the policy far into the future. Our code is available at https://github.com/bolt-research/popgym-arcade.

Large language models are increasingly being assembled into medical multi-agent systems that emulate multidisciplinary consultation through specialist roles, peer review and consensus formation. In clinical decision support, however, apparent consensus is not enough. Clinicians also need to know whether agents checked the evidence, addressed disagreement and kept uncertainty visible. Current evaluations largely score final accuracy, leaving the safety of the collaborative process untested. Here we introduce MedAgentAudit, a clinically grounded workflow audit framework for diagnosing and quantifying collaborative failure modes in medical multi-agent systems. From 3,600 execution logs, we derive an expert-validated taxonomy of ten recurrent failures spanning task comprehension, collaborative discussion, and synthesis and decision-making. We then deploy an expert-validated automated auditor as non-interventional probes across 14,400 cases, covering six multi-agent architectures, six medical text and vision datasets, and four large language model settings per modality. Across systems, collaboration yields uneven accuracy gains and frequent process failures. Unsupported observations affect 16.63% of cases and propagate downstream. In discussion, agents repeat initial views in 98.42% of cases rather than re-examining evidence, and fail to activate specialist reasoning in 42.73%. During synthesis, final answers often substitute authority or majority count for evidence checking, showing authority bias in 28.76% (rising from 35.30% to 68.75% across rounds), self-contradiction in 18.53%, contradiction neglect in 5.48% and minority suppression in 5.11%. MedAgentAudit reframes medical AI evaluation from output scoring to process-level safety and accountability, providing a practical foundation for transparent, auditable and clinician-supervised agentic systems in medicine.

Prototype-based neural networks offer interpretable predictions by comparing inputs to learned, representative signal patterns anchored in training data. While such models have shown promise in the classification of physiological data, it remains unclear whether their prototypes capture an underlying structure that aligns with broader clinical phenotypes. We use a prototype-based deep learning model trained for multi-label ECG classification using the PTB-XL dataset. Then without modification we performed inference on the MIMIC-IV clinical database. We assess whether individual prototypes, trained solely for classification, are associated with hospital discharge diagnoses in the form of phecodes in this external population. Individual prototypes demonstrate significantly stronger and more specific associations with clinical outcomes compared to the classifier's class predictions, NLP-extracted concepts, or broader prototype classes across all phecode categories. Prototype classes with mixed significance patterns exhibit significantly greater intra-class distances (p $<$ 0.0001), indicating the model learned to differentiate clinically meaningful variations within diagnostic categories. The prototypes achieve strong predictive performance across diverse conditions, with AUCs ranging from 0.89 for atrial fibrillation to 0.91 for heart failure, while also showing substantial signal for non-cardiac conditions such as sepsis and renal disease. These findings suggest that prototype-based models can support interpretable digital phenotyping from physiologic time-series data, providing transferable intermediate phenotypes that capture clinically meaningful physiologic signatures beyond their original training objectives.

Existing controllable video generation methods are typically designed for rigid, task-specific settings, such as first-frame image-to-video, inpainting, or interpolation, treating spatio-temporal control as a set of isolated problems. We formalize a unified task, arbitrary spatio-temporal video completion, where a model generates a coherent video from user-specified patches placed at any spatial location and timestamp. However, realizing such a unified framework within modern latent video diffusion models is non-trivial: causal video VAEs compress multiple frames into a single latent slot, making frame-level conditioning fundamentally ill-posed, and directly feeding sparsely populated, zero-padded video inputs into the VAE leads to severe out-of-distribution artifacts. To address these challenges, we propose VideoCanvas, a simple yet effective framework that adapts the In-Context Conditioning paradigm to arbitrary spatio-temporal completion without modifying or retraining the VAE. Our key idea is a hybrid conditioning strategy that decouples spatial and temporal control: spatially, we encode zero-padded full-frame canvases in image mode to keep VAE inputs in-distribution, and temporally we use Temporal RoPE Interpolation to assign each condition a continuous fractional index in the latent sequence for precise frame-level alignment. To evaluate this capability, we develop VideoCanvasBench, the first benchmark for arbitrary spatio-temporal video completion, covering both intra-scene fidelity and inter-scene creativity. Extensive experiments demonstrate that VideoCanvas achieves state-of-the-art performance across a diverse range of video generation tasks under a single, unified framework.

Thompson Sampling is one of the most widely used and studied bandit algorithms, known for its simple structure, low regret performance, and solid theoretical guarantees. Yet, in stark contrast to most other families of bandit algorithms, the exact mechanism through which posterior sampling (as introduced by Thompson) is able to "properly" balance exploration and exploitation, remains a mystery. In this paper, we show that the core insight to address this question stems from recasting Thompson Sampling as an online optimization algorithm. To distill this, we introduce a suitable time invariant notion of regret that leads to a stationarized bandit problem, and a stationary Bellman-optimal policy. We then show that Thompson Sampling admits an online optimization form that mimics the structure of the aforementioned Bellman-optimal policy, where "greediness" is regularized by a measure of residual uncertainty. This new lens of online optimization allows both a better understanding of Thompson Sampling dynamics, as well as a principled manner for policy improvement that mimics the Bellman-optimal benchmark.

Large language models (LLMs) are increasingly deployed in user-facing applications, raising concerns that they may reflect and amplify social biases. We investigate social identity biases in Chinese LLMs using Mandarin-specific prompts across ten representative models. Our evaluation compares ingroup ("We") and outgroup ("They") framings across 240 social groups salient in the Chinese context, using a two-tiered measurement framework that assesses both sentiment and toxicity. The prompt design explicitly accounts for linguistic properties of Mandarin, including the distinction between the default gender-neutral plural pronoun and its explicitly feminine counterpart, enabling a controlled comparison of social identity framing effects. Across models, we observe systematic ingroup-outgroup asymmetries, although their expression differs across measurement dimensions. In particular, instruction tuning often reduces sentiment asymmetries, while toxicity gaps remain more persistent. Moreover, the feminine-marked plural pronoun is associated with higher toxicity than the default gender-neutral plural in several models. Our study introduces a language-aware evaluation framework for Chinese LLMs and shows that (i) social identity biases previously documented in English also manifest in Chinese and that (ii) Mandarin-specific linguistic structure can reveal bias patterns that are not directly observable in English-only settings.

Autoregressive models have emerged as a powerful paradigm for visual content creation, but often overlook the intrinsic structural properties of visual data. Our prior work, IAR, initiated a direction to address this by reorganizing the visual codebook based on embedding similarity, thereby improving generation robustness. However, it is constrained by the rigidity of pre-trained codebooks and the inaccuracies of hard, uniform clustering. To overcome these limitations, we propose IAR2, an advanced autoregressive framework that enables a hierarchical semantic-detail synthesis process. At the core of IAR2 is a novel Semantic-Detail Associated Dual Codebook, which decouples image representations into a semantic codebook for global semantic information and a detail codebook for fine-grained refinements. It expands the quantization capacity from a linear to a polynomial scale, significantly enhancing expressiveness. To accommodate this dual representation, we propose a Semantic-Detail Autoregressive Prediction scheme coupled with a Local-Context Enhanced Autoregressive Head, which performs hierarchical prediction-first the semantic token, then the detail token-while leveraging a local context window to enhance spatial coherence. Furthermore, for conditional generation, we introduce a Progressive Attention-Guided Adaptive CFG mechanism that dynamically modulates the guidance scale for each token based on its relevance to the condition and its temporal position in the generation sequence, improving conditional alignment without sacrificing realism. Extensive experiments demonstrate that IAR2 sets a new state-of-the-art for autoregressive image generation, achieving a FID of 1.50 on ImageNet. Our model not only surpasses previous methods in performance but also demonstrates superior computational efficiency, highlighting the effectiveness of our structured, coarse-to-fine generation strategy.

Retrieval-augmented generation (RAG) has become a common practice in multimodal large language models (MLLM) to enhance factual grounding and reduce hallucination. Yet, its reliance on retrieval exposes MLLMs to knowledge poisoning attacks, in which adversaries deliberately inject malicious multimodal content into external knowledge bases to steer models toward generating incorrect or even harmful responses. We present MM-PoisonRAG, a framework to systematically study the vulnerability of multimodal RAG under knowledge poisoning. Specifically, we design two novel attack strategies: Localized Poisoning Attack (LPA), which implants targeted, query-specific multimodal misinformation to manipulate outputs toward attacker-controlled responses, and Globalized Poisoning Attack (GPA), which uses a single, untargeted adversarial injection to broadly corrupt reasoning and collapse generation quality across all queries. Extensive experiments on diverse tasks, multimodal RAG components, and attacker access levels reveal severe vulnerabilities: LPA achieves up to 56% attack success rate even under restricted access, and transfers effectively across four different retrievers without re-optimizing the adversaries. GPA completely disrupts model generation to 0% accuracy with just one poisoned content. Moreover, both LPA and GPA bypass existing defenses, underscoring the fragility of multimodal RAG and establishing MM-PoisonRAG as a foundation for future research on securing RAG frameworks against multimodal knowledge poisoning.

As Large Language Models (LLMs) develop stronger multilingual capabilities, their sensitivity to culturally diverse entities becomes increasingly important. Prior work by Naous et al. (2024) has shown that LLMs often favor Western-associated entities in Arabic. Due to the lack of entity-centric multilingual benchmarks, it remains unclear if such biases also manifest in various non-Western languages. In this paper, we introduce Camellia, a benchmark for evaluating entity-centric cultural biases in nine Asian languages, spanning six Asian cultures. Camellia includes 19,530 manually annotated entities associated with the covered Asian or Western cultures, as well as 2,173 masked contexts for these entities derived from social media posts. Using Camellia, we evaluate cultural biases in four recent multilingual LLMs across three tasks: cultural context adaptation, sentiment association, and entity extractive QA. Our analyses show that LLMs struggle with cultural adaptation across these languages, with performance differing across models developed in different regions. We further observe that different LLM families can hold distinct biases, reflected in the ways they link cultures to particular sentiments. Lastly, we find that LLMs can struggle with context understanding in some Asian languages, creating performance gaps between cultures in entity extraction.

Speculative decoding accelerates LLM inference by verifying candidate tokens from a draft model against a larger target model. Recent judge decoding boosts this process by relaxing verification criteria by accepting draft tokens that may exhibit minor discrepancies from target model output, but existing methods are restricted by their reliance on human annotations or tasks with verifiable ground truths, limiting generalizability across diverse NLP tasks. We propose SelfJudge, which trains judge verifiers via self-supervision of the target model. Our method measures semantic preservation by assessing whether token-substituted responses preserve the meaning of original responses, enabling automatic verifier training across diverse NLP tasks. Our experiments show SelfJudge achieves superior inference-accuracy trade-offs than judge decoding baselines, offering a broadly applicable solution for faster LLM inference.

Mass spectrometry-based metabolomics generates complex, high-dimensional data that holds vast potential for biological discovery but remains difficult to integrate and interpret. Knowledge graphs (KGs) unify this heterogeneous information by representing spectra, annotations, taxa, chemical classes, and biological activities as a single interoperable network; however, their practical use is limited by the steep learning curve of corresponding specialized representation and query languages. Here we introduce MetaboT, an open-source multi-agent Large Language Model (LLM) framework that translates natural-language questions into executable SPARQL queries over metabolomics knowledge graphs. MetaboT mitigates the hallucination and schema-compliance limitations of single-model approaches through a modular architecture in which specialised agents handle scope validation, entity resolution against authoritative resources, schema-aware query generation, iterative refinement, and result interpretation. We validated MetaboT on the Experimental Natural Products Knowledge Graph (ENPKG), using an expert-authored benchmark of natural-language questions paired with reference SPARQL queries, and demonstrate its ability to answer complex questions about plant--metabolite relationships and biological activities. MetaboT lowers the technical barrier for metabolomics researchers and enables semantic data mining without specialised programming expertise.

Recent Vision-Language-Action (VLA) models built on pre-trained Vision-Language Models (VLMs) require extensive post-training, resulting in high computational overhead that limits scalability and deployment.We propose CogVLA, a Cognition-Aligned Vision-Language-Action framework that leverages instruction-driven routing and sparsification to improve both efficiency and performance. CogVLA draws inspiration from human multimodal coordination and introduces a 3-stage progressive architecture. 1) Encoder-FiLM based Aggregation Routing (EFA-Routing) injects instruction information into the vision encoder to selectively aggregate and compress dual-stream visual tokens, forming a instruction-aware latent representation. 2) Building upon this compact visual encoding, LLM-FiLM based Pruning Routing (LFP-Routing) introduces action intent into the language model by pruning instruction-irrelevant visually grounded tokens, thereby achieving token-level sparsity. 3) To ensure that compressed perception inputs can still support accurate and coherent action generation, we introduce V-L-A Coupled Attention (CAtten), which combines causal vision-language attention with bidirectional action parallel decoding. Extensive experiments on the LIBERO benchmark and real-world robotic tasks demonstrate that CogVLA achieves state-of-the-art performance with success rates of 97.4% and 70.0%, respectively, while reducing training costs by 2.5-fold and decreasing inference latency by 2.8-fold compared to OpenVLA. CogVLA is open-sourced and publicly available at https://github.com/JiuTian-VL/CogVLA.

The Robbins-Siegmund theorem establishes the convergence of stochastic processes that are almost supermartingales and is one of the most commonly used approaches for analyzing stochastic iterative algorithms in stochastic approximation and reinforcement learning (RL). However, its original form has a significant limitation as it requires the zero-order term to be summable. In many important RL applications, this summable condition, however, cannot be met. This limitation motivates us to extend the Robbins-Siegmund theorem for almost supermartingales where the zero-order term is not summable, but only square-summable. In particular, we introduce a novel and mild assumption on the increments of the stochastic processes. This together with the square-summable condition enables an almost sure convergence to a bounded set. Additionally, we further provide almost sure convergence rates, high probability concentration bounds, and $L^p$ convergence rates. We then apply the new results to stochastic approximation and RL. Notably, we obtain the first almost sure convergence rate, the first high probability concentration bound, and the first $L^p$ convergence rate for $Q$-learning with linear function approximation.

Deep learning (DL) has emerged as a powerful tool for Synthetic Aperture Radar (SAR) ship classification. This survey comprehensively analyzes the diverse DL techniques employed in this domain. We identify critical trends and challenges, highlighting the importance of integrating handcrafted features, utilizing public datasets, data augmentation, fine-tuning, explainability techniques, and fostering interdisciplinary collaborations to improve DL model performance. This survey establishes a first-of-its-kind taxonomy for categorizing relevant research based on DL models, handcrafted feature use, SAR attribute utilization, and the impact of fine-tuning. We discuss the methodologies used in SAR ship classification tasks and the impact of different techniques. Finally, the survey explores potential avenues for future research, including addressing data scarcity, exploring novel DL architectures, incorporating interpretability techniques, and establishing standardized performance metrics. By addressing these challenges and leveraging advancements in DL, researchers can contribute to developing more accurate and efficient ship classification systems, ultimately enhancing maritime surveillance and related applications.

Large language models (LLMs) are typically trained by reinforcement learning (RL) with verifiable rewards (RLVR) and supervised fine-tuning (SFT) on reasoning traces to improve their reasoning abilities. However, how these methods shape reasoning capabilities remains largely elusive. Going beyond an accuracy-based investigation of how these two components sculpt the reasoning process, this paper introduces a novel analysis framework that quantifies reasoning paths and captures their qualitative changes under each training process (with models of 1.5B, 7B, and 14B parameters on mathematical domains). Specifically, we investigate the reasoning process at two levels of granularity: the trajectory-level, which examines complete reasoning outputs, and the step-level, which analyzes reasoning graphs whose nodes correspond to individual reasoning steps. Notably, clustering of unique reasoning trajectories shows complementary effects: RL compresses incorrect trajectories, whereas SFT expands correct ones. Step-level analysis reveals that RL steepens (about 2.5 times), while SFT flattens (reduced to about one-third), the decay rates of node visitation frequency, degree, and betweenness centrality distributions in the reasoning graph. This indicates that RL concentrates reasoning functionality into a small subset of steps, while SFT homogenizes it across many steps. Furthermore, by evaluating the reasoning graph topologies from multiple perspectives, we delineate the shared and distinct characteristics of RL and SFT. Our work presents a novel reasoning path perspective that explains why the current best practice of two-stage training, with SFT followed by RL, is successful, and offers practical implications for data construction and more efficient learning approaches.

Industrial monitoring systems, especially when deployed in Industry 4.0 environments, are experiencing a shift in paradigm from traditional rule-based architectures to data-driven approaches leveraging machine learning and artificial intelligence. This study presents a comparison between these two methodologies, analyzing their respective strengths, limitations, and application scenarios, and proposes a basic framework to evaluate their key properties. Rule-based systems offer high interpretability, deterministic behavior, and ease of implementation in stable environments, making them ideal for regulated industries and safety-critical applications. However, they face challenges with scalability, adaptability, and performance in complex or evolving contexts. Conversely, data-driven systems excel in detecting hidden anomalies, enabling predictive maintenance and dynamic adaptation to new conditions. Despite their high accuracy, these models face challenges related to data availability, explainability, and integration complexity. The paper suggests hybrid solutions as a possible promising direction, combining the transparency of rule-based logic with the analytical power of machine learning. Our hypothesis is that the future of industrial monitoring lies in intelligent, synergic systems that leverage both expert knowledge and data-driven insights. This dual approach enhances resilience, operational efficiency, and trust, paving the way for smarter and more flexible industrial environments.

Robotic-assisted minimally invasive surgery (RAMIS) requires accurate enforcement of the remote center of motion (RCM) constraint to ensure safe tool motion through a trocar. Existing virtual RCM controllers are commonly formulated either at the kinematic level or as task-space objectives, which makes torque-level enforcement under trocar motion and physical interaction difficult to formulate consistently. This paper models the RCM as a rheonomic holonomic constraint and incorporates it into a projection-based inverse-dynamics controller with explicit constrained/free-motion torque decomposition. The resulting formulation unifies kinematic RCM enforcement and task-space tracking at the torque level, while preserving a constraint-consistent structure for residual regulation and null-space compliance. The proposed controller is validated in simulation and on a RAMIS training platform against representative projection-based and constrained-dynamics baselines. Across spiral tracking, varying insertion depth, moving trocar conditions, and human interaction, the method achieves lower RCM residuals and smoother torque profiles while maintaining accurate tool-tip tracking. These results support the use of constraint-consistent torque control for reliable virtual RCM enforcement in surgical robotics. The project page is available at https://rcmpc-cube.github.io

Continuum robots are advancing bronchoscopy procedures by accessing complex lung airways and enabling targeted interventions. However, their development is limited by the lack of realistic training and test environments: Real data is difficult to collect due to ethical constraints and patient safety concerns, and developing autonomy algorithms requires realistic imaging and physical feedback. We present ROOM (Realistic Optical Observation in Medicine), a comprehensive simulation framework designed for generating photorealistic bronchoscopy training data. By leveraging patient CT scans, our pipeline renders multi-modal sensor data including RGB images with realistic noise and light specularities, metric depth maps, surface normals, optical flow and point clouds at medically relevant scales. We validate the data generated by ROOM in two canonical tasks for medical robotics: multi-view pose estimation and monocular depth estimation, demonstrating diverse challenges that state-of-the-art methods must overcome to transfer to these medical settings. Furthermore, we show that the data produced by ROOM can be used to fine-tune existing depth estimation models to overcome these challenges, also enabling other downstream applications such as navigation. We expect that ROOM will enable large-scale data generation across diverse patient anatomies and procedural scenarios that are challenging to capture in clinical settings. Code and data: https://github.com/iamsalvatore/room.

Early-exit neural networks (EENNs) accelerate inference by allowing intermediate classifiers to stop computation once predictions are confident enough. Most methods rely on confidence thresholds for exiting, and consequently, improving classifier calibration is widely assumed to improve performance. In this work, we challenge this assumption and show that calibration alone is not sufficient for EENNs to exploit adaptive computation. To address this insufficiency, we introduce Early-Exit Failure Prediction (EEFP), which accounts for both prediction correctness and the cost of further computation. We also propose a lightweight, EEFP-motivated procedure to improve the intermediate classifiers, which can directly replace calibration in EENNs. Extensive experiments demonstrate that our approach achieves superior cost-accuracy trade-offs compared to calibration, and EEFP more reliably reflects overall EENN performance. Our code is available at https://github.com/gmum/rethinking-calibration-for-eenns.