A Closer Look at In-Distribution vs. Out-of-Distribution Accuracy for Open-Set Test-time Adaptation

Open-set test-time adaptation (TTA) updates models on new data in the presence of input shifts and unknown output classes. While recent methods have made progress on improving in-distribution (InD) accuracy for known classes, their ability to accurately detect out-of-distribution (OOD) unknown classes remains underexplored. We benchmark robust and open-set TTA methods (SAR, OSTTA, UniEnt, and SoTTA) on the standard corruption benchmarks of CIFAR-10-C at the small scale and ImageNet-C at the large scale. For CIFAR-10-C, we use OOD data from SVHN and CIFAR-100 in their respective corrupted forms of SVHN-C and CIFAR-100-C. For ImageNet-C, we use OOD data from ImageNet-O and Textures in their respective corrupted forms of ImageNet-O-C and Textures-C. ImageNet-O is nearer to ImageNet, as unknown but related object classes (like ''garlic bread'' vs. ''hot dog'' for food, or ''highway'' vs. ''dam'' for infrastructure), while Textures is farther from ImageNet, as non-object patterns (like ''cracked'' mud, ''porous'' sponge, ''veined'' leaves). We evaluate the accuracy and confidence of TTA methods for InD vs. OOD recognition on CIFAR-10-C and ImageNet-C. We verify the accuracy of each method's own OOD detection technique on CIFAR-10-C. We also evaluate on ImageNet-C and report both accuracy and standard OOD detection metrics. We further examine more realistic settings, in which the proportions and rates of OOD data can vary. To explore the trade-off between InD recognition and OOD rejection, we propose a new baseline that replaces softmax/multi-class output with sigmoid/multi-label output. Our analysis shows for the first time that current open-set TTA methods struggle to balance InD and OOD accuracy and that they only imperfectly filter OOD data for their own adaptation updates.

Private and Stable Test-Time Adaptation with Differential Privacy

Test-time adaptation (TTA) can reduce error on new and different data by updating the model on these inputs during inference. However, these updates raise the issue of privacy w.r.t. the testing data, because the model parameters now depend on all past inputs. To control this privacy risk, we cast multiple popular TTA methods (Tent, EATA, SAR, DeYO, and COME) into differential privacy (DP) forms that apply per-sample gradient clipping and Gaussian noise for all updates. On ImageNet-C, our DP-TTA methods provide adequate privacy at small cost to accuracy, and in the low-privacy regime the clipping mechanism of DP can even improve the accuracy and stability of adaptation in the continual setting. These improvements to privacy and accuracy come at only modest computational overhead. These first results on private TTA raise awareness of the issue, inform the development of more private test-time updates, and identify per-sample clipping as an effective technique for improving the accuracy and stability of adaptation.

No One Knows the State of the Art in Geospatial Foundation Models

Geospatial foundation models (GFMs) have been proposed as generalizable backbones for disaster response, land-cover mapping, food-security monitoring, and other high-stakes Earth-observation tasks. Yet the published work about these models does not give reviewers or users enough information to tell which model fits a given task. We argue that nobody knows what the current state of the art is in geospatial foundation models. The methods may be useful, but the GFM literature does not standardize evaluations, training and testing protocols, released weights, or pretraining controls well enough for anyone to compare or rank them. In a 152-paper audit, we find 46 cross-paper disagreements of at least 10 points for the same model, benchmark, and protocol; 94/126 papers with extractable pretraining data use a configuration no other paper uses; and 39% of GFM papers release no model weights. This lack of community standards can be solved. We propose six concrete expectations: named-license weight release, shared core evaluations, copied-versus-rerun baseline annotations, variance reporting, one shared evaluation harness, and data-vs-architecture-vs-algorithm controls. These gaps are a coordination failure, not a fault of any individual lab; the authors of this paper, like many others in the GFM community, have contributed to them. Rather than just critiquing the community, we aim to provide concrete steps toward a shared understanding of how to innovate GFMs.

Self-Distillation of Hidden Layers for Self-Supervised Representation Learning

The landscape of self-supervised learning (SSL) is currently dominated by generative approaches (e.g., MAE) that reconstruct raw low-level data, and predictive approaches (e.g., I-JEPA) that predict high-level abstract embeddings. While generative methods provide strong grounding, they are computationally inefficient for high-redundancy modalities like imagery, and their training objective does not prioritize learning high-level, conceptual features. Conversely, predictive methods often suffer from training instability due to their reliance on the non-stationary targets of final-layer self-distillation. We introduce Bootleg, a method that bridges this divide by tasking the model with predicting latent representations from multiple hidden layers of a teacher network. This hierarchical objective forces the model to capture features at varying levels of abstraction simultaneously. We demonstrate that Bootleg significantly outperforms comparable baselines (+10% over I-JEPA) on classification of ImageNet-1K and iNaturalist-21, and semantic segmentation of ADE20K and Cityscapes.

Self-Soupervision: Cooking Model Soups without Labels

Model soups are strange and strangely effective combinations of parameters. They take a model (the stock), fine-tune it into multiple models (the ingredients), and then mix their parameters back into one model (the soup) to improve predictions. While all known soups require supervised learning, and optimize the same loss on labeled data, our recipes for Self-\emph{Soup}ervision generalize soups to self-supervised learning (SSL). Our Self-Souping lets us flavor ingredients on new data sources, e.g. from unlabeled data from a task for transfer or from a shift for robustness. We show that Self-Souping on corrupted test data, then fine-tuning back on uncorrupted train data, boosts robustness by +3.5\% (ImageNet-C) and +7\% (LAION-C). Self-\emph{Soup}ervision also unlocks countless SSL algorithms to cook the diverse ingredients needed for more robust soups. We show for the first time that ingredients can differ in their SSL hyperparameters -- and more surprisingly, in their SSL algorithms. We cook soups of MAE, MoCoV3, and MMCR ingredients that are more accurate than any one single SSL ingredient.

OlmoEarth: Stable Latent Image Modeling for Multimodal Earth Observation

Earth observation data presents a unique challenge: it is spatial like images, sequential like video or text, and highly multimodal. We present OlmoEarth: a multimodal, spatio-temporal foundation model that employs a novel self-supervised learning formulation, masking strategy, and loss all designed for the Earth observation domain. OlmoEarth achieves state-of-the-art performance compared to 12 other foundation models across a variety of research benchmarks and real-world tasks from external partners. When evaluating embeddings OlmoEarth achieves the best performance on 15 out of 24 tasks, and with full fine-tuning it is the best on 19 of 29 tasks. We deploy OlmoEarth as the backbone of an end-to-end platform for data collection, labeling, training, and inference of Earth observation models. The OlmoEarth Platform puts frontier foundation models and powerful data management tools into the hands of non-profits and NGOs working to solve the world's biggest problems. OlmoEarth source code, training data, and pre-trained weights are available at $\href{https://github.com/allenai/olmoearth_pretrain}{\text{https://github.com/allenai/olmoearth_pretrain}}$.

AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label data

Unprecedented volumes of Earth observation data are continually collected around the world, but high-quality labels remain scarce given the effort required to make physical measurements and observations. This has led to considerable investment in bespoke modeling efforts translating sparse labels into maps. Here we introduce AlphaEarth Foundations, an embedding field model yielding a highly general, geospatial representation that assimilates spatial, temporal, and measurement contexts across multiple sources, enabling accurate and efficient production of maps and monitoring systems from local to global scales. The embeddings generated by AlphaEarth Foundations are the only to consistently outperform all previous featurization approaches tested on a diverse set of mapping evaluations without re-training. We will release a dataset of global, annual, analysis-ready embedding field layers from 2017 through 2024.

Asymmetric Duos: Sidekicks Improve Uncertainty

The go-to strategy to apply deep networks in settings where uncertainty informs decisions--ensembling multiple training runs with random initializations--is ill-suited for the extremely large-scale models and practical fine-tuning workflows of today. We introduce a new cost-effective strategy for improving the uncertainty quantification and downstream decisions of a large model (e.g. a fine-tuned ViT-B): coupling it with a less accurate but much smaller "sidekick" (e.g. a fine-tuned ResNet-34) with a fraction of the computational cost. We propose aggregating the predictions of this \emph{Asymmetric Duo} by simple learned weighted averaging. Surprisingly, despite their inherent asymmetry, the sidekick model almost never harms the performance of the larger model. In fact, across five image classification benchmarks and a variety of model architectures and training schemes (including soups), Asymmetric Duos significantly improve accuracy, uncertainty quantification, and selective classification metrics with only ${\sim}10-20\%$ more computation.

LookWhere? Efficient Visual Recognition by Learning Where to Look and What to See from Self-Supervision

Vision transformers are ever larger, more accurate, and more expensive to compute. The expense is even more extreme at high resolution as the number of tokens grows quadratically with the image size. We turn to adaptive computation to cope with this cost by learning to predict where to compute. Our LookWhere method divides the computation between a low-resolution selector and a high-resolution extractor without ever processing the full high-resolution input. We jointly pretrain the selector and extractor without task supervision by distillation from a self-supervised teacher, in effect, learning where and what to compute simultaneously. Unlike prior token reduction methods, which pay to save by pruning already-computed tokens, and prior token selection methods, which require complex and expensive per-task optimization, LookWhere economically and accurately selects and extracts transferrable representations of images. We show that LookWhere excels at sparse recognition on high-resolution inputs (Traffic Signs), maintaining accuracy while reducing FLOPs by up to 34x and time by 6x. It also excels at standard recognition tasks that are global (ImageNet classification) or local (ADE20K segmentation), improving accuracy while reducing time by 1.36x.

ReservoirTTA: Prolonged Test-time Adaptation for Evolving and Recurring Domains

This paper introduces ReservoirTTA, a novel plug-in framework designed for prolonged test-time adaptation (TTA) in scenarios where the test domain continuously shifts over time, including cases where domains recur or evolve gradually. At its core, ReservoirTTA maintains a reservoir of domain-specialized models -- an adaptive test-time model ensemble -- that both detects new domains via online clustering over style features of incoming samples and routes each sample to the appropriate specialized model, and thereby enables domain-specific adaptation. This multi-model strategy overcomes key limitations of single model adaptation, such as catastrophic forgetting, inter-domain interference, and error accumulation, ensuring robust and stable performance on sustained non-stationary test distributions. Our theoretical analysis reveals key components that bound parameter variance and prevent model collapse, while our plug-in TTA module mitigates catastrophic forgetting of previously encountered domains. Extensive experiments on the classification corruption benchmarks, including ImageNet-C and CIFAR-10/100-C, as well as the Cityscapes$\rightarrow$ACDC semantic segmentation task, covering recurring and continuously evolving domain shifts, demonstrate that ReservoirTTA significantly improves adaptation accuracy and maintains stable performance across prolonged, recurring shifts, outperforming state-of-the-art methods.