MG-SpaIR is a training-data-free framework that restores clean images from single observations corrupted by mixed degradations. It uses implicit neural representations with a multi-grade coarse-to-fine hierarchy and explicit sparse regularization to improve fidelity and suppress artifacts. Experiments show it outperforms Deep Image Prior.

Proposes a system that automatically generates realistic assembly sequences and trains real-time inspection models using synthetic data. It requires only CAD models and step descriptions, can be deployed within an hour, and achieves 92.4% accuracy on a real-world assembly case.

Researchers propose DeCoDe, a technique that decomposes few-shot image classification into pairwise comparisons, enabling off-the-shelf multimodal LLMs to become powerful few-shot classifiers without additional training. It outperforms state-of-the-art methods on twelve datasets, with code open-sourced.

This paper introduces SegFS, a dual-stream fast-slow framework for open-vocabulary video instance segmentation (OV-VIS). By using a slow, accurate path on sparse keyframes and a fast, lightweight path on subsequent frames, it achieves up to 14x lower latency than the mobile-oriented MOBIUS model while maintaining competitive accuracy.

This paper presents PixelEyes, a multi-turn visual reasoning agent that decouples reasoning from perception to address the repeated localization failures of multimodal LLMs. It introduces mask-guided visual search and semantic-region breadth-first search, constructs the PixelEyes-6K dataset and Pinpoint-Bench benchmark, and demonstrates significant headroom for existing models.

Urban-scale Visual Place Recognition (VPR) aims to identify the geographic location of a query image by matching it against a geo-tagged database. However, recent methods overlook a serious long-tailed problem in urban datasets, biasing models toward well-photographed areas while failing in sparsely covered regions. This paper systematically characterizes this imbalance and proposes Distribution-Aware Place Recognition (DAPR), a model-agnostic plug-in framework that rebalances gradient contributions across head and tail classes. Within classification-retrieval pipelines, DAPR applies a multi-scale distance search mechanism to compute per-class distributional compactness, providing complementary gains at the retrieval stage. On the SF-XL benchmark, DAPR outperforms the previous baseline by…

This paper proposes a training-free evidence-injection framework that systematically mitigates hallucinations in medical MLLMs by recalibrating visual perception and anchoring textual reasoning with ROI priors from MedSAM and anatomical coordinate mapping. Evaluations show up to ~6% improvement in close-ended accuracy and ~35% reduction in open-ended hallucinations.

Proposes DiSIINet, a unified model based on Denoising Diffusion Implicit Models that jointly performs medical image enhancement and segmentation, with a Symbiotic Information Interaction module enabling dynamic feature-level exchange, achieving significant improvements on multi-modal medical datasets.

A new method called Multi-Scale Layer Attention (MSLA) is proposed to improve Oracle Bone Inscription recognition. It explicitly models multi-scale and cross-layer feature interactions to capture fine-grained details. Experiments show it outperforms existing attention mechanisms while maintaining computational efficiency.

SLIM-RL is a new reinforcement learning method for diffusion large language models (dLLMs). It bounds the commit risk of each rollout step with a tau-budget decoder, reducing aggregate risk without trajectory reconstruction. Using a trace-free random-masking objective with variance reduction, it matches TraceRL's best MATH500 accuracy at 0.46x training samples on SDAR-4B, and improves over TraceRL on math and code benchmarks.

This study applies machine learning to analyze 619 Inka khipus, revealing three structural clusters via unsupervised clustering, achieving 86% F1 score in provenance classification, and identifying cord twist direction as the key discriminator for imperial style. It also uncovers colonial acquisition bias in museum collections and independently validates the moiety structure of Santa Valley khipus.

ALEE is a new evaluation framework that uses Abstract Meaning Representations (AMR) to generate English minimal pairs with controlled semantic shifts, paired with translations into target languages, enabling fine-grained diagnostics for embedding models in any language with English parallel data. A large-scale study across 275+ languages reveals significant performance disparities correlated with language prevalence and subword tokenization.

This paper systematically audits existing KB-VQA benchmarks, uncovering widespread issues such as missing or contradicted answers, ambiguous questions, and trivial visual scenes, which render accuracy a misleading metric. The authors propose an audit-and-repair protocol and a multi-entity augmentation protocol to address these flaws, and demonstrate that corrected evaluations yield significantly different model rankings.

A new study shows that while medical LLM hallucination can be reliably detected (AUROC 0.77-0.86) using a simple probe, the underlying neural signal is distributed and redundant. Crucially, detection does not translate to control: steering hallucination-associated neurons fails to correct outputs, revealing a fundamental gap between readability and controllability.

Researchers introduce a comprehensive hate speech dataset covering six topics in Turkish and Arabic, and develop state-of-the-art BERT-based models for hate category classification, intensity prediction, target identification, and span detection.

A study evaluates frontier LLMs on Arabic cultural and sociolinguistic knowledge using human expert grading. The cross-evaluation framework tests models on Egyptian and Iraqi Arabic, finding that GPT-5.4 is the most reliable judge and that implicit cultural reasoning remains a major challenge.

A research paper presented at ACL 2026 BigPicture Workshop introduces techniques to harness language model latent spaces for better control and trust, including steering vectors and model calibrators.

Large language models face a binary failure in creative writing, oscillating between safe surface editing and uncontrolled plot expansion. Researchers propose Loom, a framework based on the narratological distinction between story and discourse, using a three-layer pipeline with intent-centered semiotic chain-of-thought to achieve precise control over narrative intent and rendering density. Evaluation shows Loom resolves this tension, achieving highest quality scores with gains in factual integrity and descriptive intensity.

A new paper challenges Beckmann & Butlin's ontological framework for LLM individuation, presenting four empirical wedges from persona-topology experiments and proposing regime-indexed individuation.

Federated learning is bandwidth-bound on two orthogonal axes: model size, which limits how often parameter-averaging methods can afford to merge, and class count, which makes per-probe soft-label distillation prohibitive at large vocabularies. TallyTrain collapses the class-count axis to ⌈log2C⌉ bits per probe by transmitting only each peer's argmax class index. Under non-IID training, hard-label majority voting filters noise where soft-label averaging amplifies it. TallyTrain matches or beats soft-label distillation at up to three orders of magnitude less communication. A bandwidth-bridge variant combining hard-label consensus with sparse parameter merges Pareto-dominates FedAvg, FedProx, and FedDF.

Thermodynamic computing devices based on the Ising model show great promise for low-power AI inference and edge computing, but scalable training methods remain limited. This paper presents a scalable backpropagation-based algorithm for training deep convolutional networks on Ising machines, achieving 94.9% on CIFAR-10 and 76.0% on CIFAR-100. It also develops a theory for the tradeoff between inference cost and accuracy, and discusses implications for hardware development.

This paper proposes a reinforcement learning method with a novel verifiable, label-free reward to train calibrated probabilistic forecasters. Applied to NFL win probability prediction, a 7B model trained solely with this reward matches betting market calibration without human labels or supervised fine-tuning, outperforming zero-shot frontier models.

Parameter-efficient fine-tuning (PEFT) reparameterizes weight updates in a fixed basis: low-rank adapters operate in the spatial domain, while spectral methods operate in a fixed Fourier domain. This paper introduces Fractional-Fourier Mixture of Experts, where each expert has a learnable fractional-Fourier order that interpolates between spatial and Fourier domains. Routing tokens through experts of different orders allows low-rank updates to be placed in their most compact domain, and the experts are naturally decorrelated, reducing interference and improving multi-task composition. The method adds negligible cost and outperforms strong baselines on LLaMA-3.1-8B and Qwen2.5-7B across various benchmarks.

Evolutionary neural architecture design for multivariate time-series forecasting remains underexplored. This paper introduces EVOTS, an evolutionary search framework that uses a modular genome representation and repair mechanism to discover task-adaptive Transformer-like models, achieving competitive results on ETT benchmarks.

A new paper proves that three popular methods for training language models to reason—GRPO, Dr. GRPO, and DAPO—are all adjusting the same number: the standard deviation of answer correctness across multiple samples. The group-standard-deviation identity shows that disagreement directly determines the size of the training update, validated on the Big-Math dataset and in controlled experiments.

FoGS is a novel synthetic data method for survival analysis that selects samples from a pool of multiple generators instead of generating directly, addressing data scarcity and privacy restrictions in clinical settings. On 16 public datasets, it significantly improves model performance over single-generator approaches while maintaining privacy.

This study disentangles the effects of classifier tuning vs. joint optimization in semi-supervised learning (SSL) pipelines for security classification, finding that simply tuning the classifier with Bayesian Optimization recovers 86% of the performance gain, and a simplified recipe matches the full pipeline.

Generative models used as surrogates for physical simulation often fail to enforce physics constraints. Constrained sampling can enforce constraints at inference but is computationally expensive. This paper introduces SNAP-FM, which leverages sparse GPU nonlinear optimization to accelerate constraint projection. Using block-sparse Jacobian and KKT systems with ExaModels.jl and MadNLP.jl, the method achieves faster nonlinear constraint projection on PDE benchmarks while maintaining accuracy.

This paper introduces the Manifestation Unit Protocol, a structured representation scheme for mechanistic interpretability that organizes component-level analysis outputs into queryable, reusable fields, validated across multiple models.

Seed2.0 is a model series that takes a meaningful step toward solving complex, real-world tasks. It identifies user needs, builds a reliable evaluation system, and targets long-tail knowledge and complex instruction following. It also delivers world-leading reasoning, visual understanding, and search capabilities.