Self-supervised learning holds the promise of eliminating the need for manual data annotation, enabling models to scale effortlessly to massive datasets and larger architectures. By not being tailored to specific tasks or domains, this training paradigm has the potential to learn visual representations from diverse sources, ranging from natural to aerial images -- using a single algorithm. This technical report introduces DINOv3, a major milestone toward realizing this vision by leveraging simple yet effective strategies. First, we leverage the benefit of scaling both dataset and model size by careful data preparation, design, and optimization. Second, we introduce a new method called Gram anchoring, which effectively addresses the known yet unsolved issue of dense feature maps degrading during long training schedules. Finally, we apply post-hoc strategies that further enhance our models' flexibility with respect to resolution, model size, and alignment with text. As a result, we present a versatile vision foundation model that outperforms the specialized state of the art across a broad range of settings, without fine-tuning. DINOv3 produces high-quality dense features that achieve outstanding performance on various vision tasks, significantly surpassing previous self- and weakly-supervised foundation models. We also share the DINOv3 suite of vision models, designed to advance the state of the art on a wide spectrum of tasks and data by providing scalable solutions for diverse resource constraints and deployment scenarios.

In digital pathology, whole-slide images (WSIs) are often difficult to handle due to their gigapixel scale, so most approaches train patch encoders via self-supervised learning (SSL) and then aggregate the patch-level embeddings via multiple instance learning (MIL) or slide encoders for downstream tasks. However, patch-level SSL may overlook complex domain-specific features that are essential for biomarker prediction, such as mutation status and molecular characteristics, as SSL methods rely only on basic augmentations selected for natural image domains on small patch-level area. Moreover, SSL methods remain less data efficient than fully supervised approaches, requiring extensive computational resources and datasets to achieve competitive performance. To address these limitations, we present EXAONE Path 2.0, a pathology foundation model that learns patch-level representations under direct slide-level supervision. Using only 37k WSIs for training, EXAONE Path 2.0 achieves state-of-the-art average performance across 10 biomarker prediction tasks, demonstrating remarkable data efficiency.

However, the "sparse-to-dense" paradigm complicates the pipeline of SSOD, and

We further evaluate SGBS+EAS on nine real-world based instance sets from [15]. Each instance set consists of 20 instances that have similar characteristics (i.e., they have been sampled from the same underlying distribution). The instance sets differ significantly in terms of several structural properties, for example, the number of customers n and their position (e.g., clustered vs. random positions). A more detailed description of instance sets can be found in [15]. One major advantage of neural combinatorial optimization approaches over traditional handcrafted optimization methods is their ability to quickly learn customized heuristics for new problem settings.

A wide range of models have been proposed for Graph Generative Models, necessitating effective methods to evaluate their quality. So far, most techniques use either traditional metrics based on subgraph counting, or the representations of randomly initialized Graph Neural Networks (GNNs). We propose using representations from contrastively trained GNNs, rather than random GNNs, and show this gives more reliable evaluation metrics. Neither traditional approaches nor GNN-based approaches dominate the other, however: we give examples of graphs that each approach is unable to distinguish. We demonstrate that Graph Substructure Networks (GSNs), which in a way combine both approaches, are better at distinguishing the distances between graph datasets.

In semi-supervised learning (SSL) for enhancing the performance of graph neural networks (GNNs) with unlabeled data, introducing mutually independent decision factors for cross-validation is regarded as an effective strategy to alleviate pseudo-label confirmation bias and training collapse. However, obtaining such factors is challenging in practice: additional and valid information sources are inherently scarce, and even when such sources are available, their independence from the original source cannot be guaranteed. To address this challenge, In this paper we propose a Differentiated Factor Consistency Semi-supervised Framework (DiFac), which derives differentiated factors from a single information source and enforces their consistency. During pre-training, the model learns to extract these factors; in training, it iteratively removes samples with conflicting factors and ranks pseudo-labels based on the shortest stave principle, selecting the top candidate samples to reduce overconfidence commonly observed in confidence-based or ensemble-based methods. Our framework can also incorporate additional information sources. In this work, we leverage the large multimodal language model to introduce latent textual knowledge as auxiliary decision factors, and we design a accountability scoring mechanism to mitigate additional erroneous judgments introduced by these auxiliary factors. Experiments on multiple benchmark datasets demonstrate that DiFac consistently improves robustness and generalization in low-label regimes, outperforming other baseline methods.