Brief

Researchers have introduced AUCp, a new metric designed to improve model selection in abnormality detection tasks, particularly within medical imaging. This metric addresses the challenge of relying on labeled validation data, which is often scarce or time-consuming to acquire for rare diseases. By treating all unannotated test samples as positive and using a traditional AUC calculation, AUCp effectively identifies the optimal model for inference without needing annotated test sets, outperforming conventional metrics in unsupervised and self-supervised learning scenarios. AI

Researchers have developed a new method called G2G to improve the estimation of relative 6-Do-F poses between groups of images. This technique leverages existing intra-group geometry information and pretrained multi-view backbones. G2G introduces lightweight trainable modules that fuse information across different groups, achieving state-of-the-art accuracy on various datasets without retraining the entire foundation model. AI

Researchers have developed BLINQ, a novel model-based algorithm designed to learn Whittle indices for Markov Decision Processes. This new approach constructs an empirical estimate of the MDP and then computes the indices, offering a proven convergence guarantee and a bound on learning time. Numerical experiments indicate BLINQ requires fewer samples than existing Q-learning methods for accurate approximations and has a lower overall computational cost. AI

Researchers have developed a deep learning framework designed to separate true signals from measurement artifacts in multi-sensor data. This method uses a dual-encoder architecture and a counterfactual generation objective to disentangle intrinsic physical properties from sensor-specific distortions. The framework's effectiveness was demonstrated on astrophysical galaxy images from the DESI Legacy Imaging Survey and the Hyper Suprime-Cam Survey, offering a general approach for scientific and multi-modal self-supervised pretraining. AI

Researchers have introduced ERBench, a new benchmark and test suite specifically designed to evaluate algorithms for equation discovery. This framework focuses on assessing how well these algorithms can recover known groundtruth formulas, addressing limitations in existing benchmarks that often use small datasets and lack robustness testing. ERBench aims to provide a more rigorous evaluation of symbolic regression algorithms, which are crucial for automating the discovery of scientific models from data. AI

Researchers have developed WIEN-INR, a novel hierarchical implicit neural representation (INR) framework designed to improve the representation of complex scientific data. This multi-scale architecture distributes modeling across different resolution levels, using an enhancement network to recover fine details. The approach aims to enable smaller networks to capture the full spectrum of information, thereby reducing computational and storage costs while maintaining high fidelity. WIEN-INR has demonstrated its effectiveness on diverse experimental measurements, offering a practical solution for broader adoption of neural representations in scientific workflows. AI

The REACT 2026 challenge focuses on generating multiple appropriate facial reactions in response to speaker behavior, building on previous iterations. This year's challenge introduces personalization by incorporating individual-level personality labels and EEG recordings, moving towards a one-to-many personalized facial reaction generation setting. New baselines and guidelines are provided for both offline and online generic and personalized MAFRG sub-challenges. AI

Researchers have introduced CTS-Bench, a new benchmark suite designed to evaluate the trade-offs between graph coarsening techniques and the accuracy of Graph Neural Networks (GNNs) for Clock Tree Synthesis (CTS) in electronic design automation. The benchmark includes 4,860 converged physical design solutions, allowing for systematic analysis of how graph coarsening impacts prediction accuracy and computational efficiency. While coarsening can significantly reduce memory usage and training time, the study found it often removes critical structural information, leading to poor performance on CTS-specific tasks like clock skew prediction. AI

Researchers have developed a novel multi-branch deep learning framework designed to improve the detection of Parkinson's disease through speech analysis. This approach utilizes three distinct speech representations: Log-Mel spectrograms, MFCCs, and HuBERT embeddings, each processed by specialized neural networks. A key innovation is a context-guided cross-modal attention mechanism that dynamically integrates these diverse features, leading to enhanced accuracy in identifying the disease. AI

Researchers have introduced DAL-PCQA, a new dataset designed to improve point cloud quality assessment by incorporating distortion-level and language-driven reasoning. Unlike previous methods that provide only a single score, DAL-PCQA includes multi-level distortion severity labels, quality categories, and natural language descriptions of artifacts. This dataset aims to enable more interpretable and explainable quality assessment by aligning with how humans perceive and describe point cloud degradations. AI

Researchers have developed a new variational method for accurately computing signed distance functions (SDFs) from point clouds. This approach explicitly incorporates the medial axis, which is the jump set of the SDF's gradient, by using a higher-order variational formulation. The method employs a phase field approximation to implicitly describe the medial axis and uses neural networks to approximate both the SDF and the phase field, demonstrating improved accuracy compared to existing methods. AI

Researchers have developed a new framework to improve the user experience of online shopping by personalizing review ranking and summarization. This system integrates user preference modeling, sentiment analysis, and Large Language Models (LLMs) to tailor review content to individual needs. By analyzing historical reviews and user-selected product aspects, the framework ranks and summarizes reviews to reduce information overload and enhance decision-making confidence. Evaluations showed this personalized approach significantly outperformed traditional ranking methods and improved user satisfaction and efficiency. AI

Researchers have developed a new framework to understand neural network representations in modular arithmetic tasks. Their work refines the explanation for why these networks adopt a two-dimensional cyclic geometry, deviating from the predicted neural collapse phenomenon. The study details a layerwise training mechanism where classifier weights form a rank-2 configuration before embeddings align, and explains this cyclic solution's advantage over standard neural collapse under certain conditions. AI

Researchers have developed a novel voice conversion framework that uses K-Nearest Neighbors (KNN) retrieval on WavLM representations to align non-parallel speech data. This method constructs synthetic training pairs from non-parallel source and target audio, enabling supervised learning without requiring explicit alignment or parallel corpora. The framework also incorporates a speaker loss to maintain consistent target-speaker identity, demonstrating high naturalness and speaker similarity across multiple languages, even when trained solely on English data. AI

Researchers have developed an AI-based backcasting approach to generate synthetic IoT sensor data for strawberry yield forecasting. By combining this synthetic data with actual sensor and yield records, they trained models that improved forecasting accuracy. This method addresses data gaps in agricultural settings, enabling more robust data-driven resource management for farmers. AI

Researchers have introduced Conditional Random Ordered Transport Spaces (CROTS), a novel framework for evaluating distributional learning. CROTS equips spaces of random probability measures with an ambient Wasserstein metric and a stochastic order, enabling the assessment of mass movement admissibility. This theory provides a mathematical language to describe issues like evidence overreach and distributional shift in machine learning. AI

Researchers have introduced RAD, a new dataset and benchmark designed to evaluate anomaly detection capabilities in real-world robotic scenarios. Unlike previous benchmarks, RAD features objects captured from numerous robotic viewpoints under uncontrolled lighting, simulating practical deployment challenges. The study found that established 2D feature-based methods surprisingly outperformed newer 3D and vision-language models in image-level anomaly detection, though the gap narrowed for precise defect localization. AI

Researchers have introduced Projection Transfer Learning (ProjectionTL), a novel framework designed to improve learning from multiple, heterogeneous data sources when the target dataset is limited. This method uses a hierarchical Bayesian model to adaptively weigh information from different sources, capturing global alignment. It then refines this transfer at the feature level through a posterior-projection step, selecting features that agree locally with the target signal. ProjectionTL aims to mitigate negative transfer and enhance interpretability, showing improved accuracy and stability in simulations and biomedical applications. AI

Researchers have introduced a new framework called counterfactual transport flows for offline reinforcement learning. This method aims to improve decision-making policies using only logged historical data, without extrapolating beyond the available information. The approach constructs local preference pairs by finding similar trajectories with higher feedback in latent space, which then guides a conservative refinement process. This allows for a controllable trade-off between maintaining the original behavior and applying stronger improvements, as demonstrated on D4RL benchmarks. AI

Researchers have developed a new framework for statistical decision theory that incorporates counterfactual loss, addressing limitations in classical approaches that only consider realized outcomes. This new method allows for the evaluation of decision quality against feasible alternatives at an individual level, which is crucial in fields like pretrial bail decisions. The framework demonstrates that counterfactual risk is identifiable under specific conditions, particularly when the loss function is additive in potential outcomes, and can capture both decision accuracy and difficulty, unlike standard losses that only reflect accuracy. AI

Researchers have developed CSFlow, a novel weighting scheme that aligns the iterative denoising process in flow matching models with human contrast sensitivity. This method accounts for the human visual system's varying sensitivity to different spatial frequencies and the tendency of diffusion models to stabilize coarse image content before fine details. By estimating which frequencies are generated at each reverse flow interval and weighting timesteps accordingly, CSFlow has demonstrated improvements in image generation quality, reducing FID scores and enhancing visual realism. AI

Researchers have developed a more precise convergence analysis for Decentralized SGD, a key algorithm in decentralized learning. Unlike previous methods that focused solely on the spectral gap of the network topology, this new analysis considers all eigenvalues of the mixing matrix. Experiments confirmed that this refined approach more accurately describes how different network topologies impact the convergence rate of Decentralized SGD, particularly in heterogeneous settings. AI

This paper provides a comprehensive review of deep multi-task learning (MTL) techniques applied to connected autonomous vehicles (CAVs). It explores how MTL can enable a single model to handle diverse tasks like perception, prediction, planning, and control, which is crucial for efficient and real-time operation in complex driving scenarios. The survey categorizes existing research based on whether tasks are performed solely by the ego vehicle or enhanced through vehicle-to-everything (V2X) communication, and also examines MTL in the context of V2X communications and radio resource management. The authors identify current research gaps and suggest future directions for advancing MTL in CAV systems. AI

Researchers have developed a novel framework for screen content image super-resolution (SCISR) that addresses limitations in existing methods by considering frequency characteristics. Their Frequency Decoupled Framework (FDF) separates images into amplitude and phase streams, utilizing specialized modules to capture periodic patterns and global configuration. This approach has demonstrated state-of-the-art performance across multiple datasets and scales. AI

Researchers have developed MinNav, a novel navigation system for tiny aerial robots that utilizes optical flow and its uncertainty to navigate complex environments. This system can autonomously fly through scenes with static and dynamic obstacles, as well as unknown gaps, without prior knowledge of the surroundings. MinNav achieves a 70% success rate in real-world experiments and offers comparable performance to depth-based methods with significantly less computational power, making it suitable for onboard tiny aerial robots. AI

Researchers have developed a new method to resolve the scale ambiguity in feed-forward 3D reconstruction models by utilizing satellite imagery as a global metric reference. This approach integrates satellite patches with reconstruction backbones, enforcing consistency to infer absolute scale, refine geometry, and estimate camera pose in a metric coordinate frame. The framework demonstrated improvements in metric depth estimation, point-cloud reconstruction, and camera localization across multiple datasets. AI

Researchers have developed a new method called GraCE to estimate ground contact forces from 3D human motion capture data. This approach utilizes the human's center of gravity and body mass distribution to infer contact dynamics, offering a more scalable solution than traditional force plates or pressure mats. GraCE aims to capture complex pressure distributions and has demonstrated superior performance on datasets like GroundLink and MOYO for ground reaction force and contact pressure prediction, respectively. AI

Researchers have introduced DifferSeg, a novel framework for multimodal binary segmentation that addresses challenges in aligning complementary features and balancing high- and low-frequency representations. The framework utilizes a differential perception fusion module to adaptively align multimodal features and enhance their complementarity, while a frequency-guided decoder ensures consistency between detailed structures and semantic information. DifferSeg has demonstrated superior performance across numerous datasets and tasks, outperforming 67 existing methods. AI

Researchers have developed a novel approach for load forecasting in data-scarce environments by leveraging a large language model called Chronos. This LLM framework utilizes its extensive pre-trained knowledge to achieve accurate predictions without requiring extensive fine-tuning on specific datasets. Experiments across five real-world datasets demonstrated that Chronos significantly outperforms nine traditional baseline models in both deterministic and probabilistic forecasting, showing substantial reductions in error metrics. AI

Researchers have introduced a new framework called Stochastic Dimension Implicit Functional Projection (SDIFP) to address challenges in enforcing integral constraints within high-dimensional neural network solvers for partial differential equations. This method replaces traditional grid-based projection techniques with a global affine correction, determined by scalar coefficients derived from a weighted quadrature rule. SDIFP aims to improve scalability and efficiency, particularly for mesh-free methods like physics-informed neural networks (PINNs), by separating quadrature evaluation from automatic differentiation memory costs and enabling pointwise inference efficiency. AI

Researchers have developed a new framework for anomaly detection in multispectral point clouds captured by drones under dense foliage. This method addresses the challenge of varying illumination conditions, which can obscure targets. It achieves this by first estimating solar angles to distinguish shadows from dark objects and then employing an illumination-consistent sparse representation to separate spectral reflectance from lighting effects. The framework demonstrates improved performance in complex forest environments for tasks like identifying hidden military targets or archaeological sites. AI

Researchers have introduced Energy-Regularized Spatial Masking (ERSM), a new framework designed to improve the robustness and interpretability of vision models. ERSM treats feature selection as a differentiable energy minimization problem, assigning each visual token an energy value based on its importance and spatial coherence. This approach allows models to autonomously find an optimal balance of information density, leading to emergent sparsity and enhanced performance in robustness tests without explicit supervision. AI

Researchers have uncovered a new relationship between the noise introduced by Stochastic Gradient Descent (SGD) and the curvature of the loss landscape in deep learning models. Their findings indicate that this noise is not directly proportional to the Hessian of the loss, as previously assumed under specific conditions. Instead, the study reveals a more general connection where the SGD noise covariance is related to the expected value of per-sample Hessians, suggesting these two factors approximately commute rather than coincide. AI

A new research paper explores the theoretical limits of function approximation, demonstrating that compositional methods, such as neural networks, can significantly outperform superpositional methods. The study constructs specific examples where the approximation error gap between these two approaches can be arbitrarily large. This work has implications for understanding the fundamental capabilities of different model architectures in machine learning. AI

Researchers have introduced a new representation called Geometric Primary Structure (GPS) for understanding articulated parts in robotic manipulation. This method aims to balance scalability and quality by abstracting the geometric structure of object parts. An efficient VR-based annotation system was used to collect a dataset of 41,000 frames for 234 objects, enabling the training of a generalizable GPS model that achieved a 73% success rate in object manipulation tasks. AI

Researchers have developed a new metric called the directional linear separability measure (LSM) to analyze the geometric properties of neural network representations. This measure quantifies how well a target class can be separated from other classes using affine halfspaces, providing a class-wise and asymmetric assessment. LSM is designed to distinguish between changes due to linear reparameterization and those caused by information loss or nonlinear transformations, offering a tool to diagnose class-wise intrusion in deep learning architectures. AI

A new survey paper details the emerging field of Test-Time Scaling (TTS) for Multimodal Foundation Models (MFMs). The paper categorizes existing TTS methods into sampling-based, feedback-based, and search-based approaches. It also outlines common applications, benchmarks, and future research directions for enhancing MFM performance in generation and reasoning tasks. AI

Researchers have developed a new framework called GraphNC to improve semi-supervised graph anomaly detection. This method calibrates normality by leveraging both labeled and unlabeled data, using a teacher model to guide the process. GraphNC incorporates anomaly score distribution alignment and perturbation-based normality regularization to enhance the accuracy and separability of anomaly scores and node representations. AI

Researchers have developed GNSS-FM, a novel self-supervised foundation model designed for analyzing daily Global Navigation Satellite System (GNSS) displacement time series. This model utilizes a dual-stream input combining displacement and velocity data, pre-trained with a masked latent prediction objective. After pre-training on data from over 17,000 GNSS stations, GNSS-FM demonstrated strong performance when fine-tuned for displacement forecasting and seismic step localization, outperforming existing task-specific baselines. AI

Researchers have introduced the bulk-boundary decomposition, a novel framework for analyzing the training dynamics of deep neural networks. This approach separates the network's Lagrangian into a data-independent bulk term and a data-dependent boundary term. The bulk term characterizes the inherent dynamics influenced by network architecture and activation functions, while the boundary term reflects the stochastic interactions arising from training samples at the input and output layers. This decomposition reveals the local and homogeneous structure within deep networks, leading to the derivation of an energy continuity equation. AI

Researchers have introduced a new framework for understanding generative models, focusing on the concept of "self-consistent generative paths." This framework defines a path as self-consistent if it represents a random fixed point of admissible local variational transport corrections. The theory yields a metric called the random fixed-point path residual (R-FPR) to quantify the gap between a generated path and its correction, offering a principle for diagnosing and improving various generative models. AI

Researchers have developed a new method for generating 3D shapes by diffusing over superquadric parameters instead of dense geometric representations. This approach significantly reduces the dimensionality of the diffusion state, requiring only 7KB of parameters per shape. The diffusion-over-superquadrics method enables faster generation, improved scalability, and supports advanced capabilities like part-level editing and constraint-based design, while achieving competitive performance on standard benchmarks. AI

Researchers have developed a new decentralized online Riemannian optimization algorithm capable of operating beyond the limitations of Hadamard manifolds, extending its applicability to spaces with positive curvature. The algorithm incorporates a curvature-aware consensus step that facilitates linear convergence even in these more complex geometric settings. This advancement leads to a $O(\sqrt{T})$ regret bound for the decentralized online Riemannian gradient descent method, with similar bounds achieved in a two-point bandit feedback scenario using efficient gradient estimators. AI

Researchers have developed MB-Loc, a new framework for multi-planar bird's-eye-view localization in outdoor LiDAR scenes. This method addresses computational inefficiency and viewpoint sensitivity in existing scene coordinate regression techniques. MB-Loc projects LiDAR scans into a 2.5D representation, enabling faster processing with standard 2D CNNs while retaining crucial 3D geometric information. The framework also incorporates a KL-regularized latent bottleneck for spatial uncertainty modeling and 3D spatial augmentations for rotation robustness, outperforming current state-of-the-art methods on the NCLT dataset at real-time inference speeds. AI

Researchers have developed IR-SIM, a new lightweight simulator designed to streamline robotics research, particularly for tasks involving large language models. This simulator allows for the creation and modification of navigation scenarios using simple YAML configuration files and text prompts, making it easier to prototype and develop algorithms. IR-SIM also facilitates automated benchmarking and data generation for robot learning, with capabilities to bridge to higher-fidelity simulators and real-world deployments. AI

Researchers have developed a new framework called RGB-S that explicitly aligns tactile sensor data with visual information for robotic manipulation. This method projects tactile sensor locations directly onto RGB images, creating saliency maps that account for spatial uncertainty. By integrating these 2D anchors, the system injects physical contact priors into visual models, improving their ability to handle unreliable or occluded visual inputs. Experiments demonstrated a significant improvement in success rates for dexterous manipulation tasks under severe visual occlusion. AI

Researchers have developed a new method for training machine learning models to downscale climate data, focusing on how to select training years effectively. Their study, using the CESM2 Large Ensemble, found that training models on years distributed across the entire climate trajectory, rather than contiguous historical periods, significantly improves their ability to reproduce climate variability. This approach, even with limited data, outperforms models trained solely on historical data and suggests that broad sampling of climate states is more beneficial than temporal continuity for allocating scarce high-resolution simulation resources. AI

A new paper explores the relationship between traditional differential equation models and modern data-driven approaches like neural operators. It argues that many modeling strategies share a common structure, differing primarily in their assumed input-output mappings. The research suggests that only certain models are capable of true mechanism discovery and subsequent generalization, offering insights into their appropriate applications. AI

Researchers have developed MPC-Flow, a novel framework for solving inverse problems using flow-based generative models. This method employs model predictive control to guide the model's dynamics, making conditional generation more practical. MPC-Flow offers a spectrum of guidance algorithms, some of which bypass the need for backpropagation through the generative model's trajectory. The framework has demonstrated strong performance and scalability on image restoration tasks, including in-painting, deblurring, and super-resolution, even with large-scale models like FLUX.2 on consumer hardware. AI

Researchers have developed DALE-CT, a new family of 2D foundation models for processing computed tomography (CT) data. Built from scratch using a self-supervised learning approach called LeJEPA, DALE-CT incorporates a novel 3D depth-aware pre-training strategy with both automated and human-annotated supervision. This model achieved a Macro AUROC of 0.833 on the CT-RATE dataset for multi-abnormality detection, nearing the performance of state-of-the-art 3D vision-language models with less data and no textual supervision. AI