We investigate the presence and spatial characteristics of the jet base emission in M87* at 230 GHz, enabled by the significantly enhanced (u,v) coverage in the 2021 Event Horizon Telescope (EHT) observations. The integration of the 12−m Kitt Peak Telescope (USA) and NOEMA (France) stations into the array introduces two critical intermediate-length baselines to SMT (USA) and IRAM 30−m (Spain), providing sensitivity to emission structures at spatial scales of ∼250 μas and ∼2500 μas (∼ 0.02 pc and ∼ 0.02 pc). Without these new baselines, previous EHT observations of the source in 2017 and 2018 lacked the capability to constrain emission on large scales, where a “missing flux” of order ∼1 Jy is expected to reside. To probe these scales, we analyzed closure phases–robust against station-based gain calibration errors–and model the jet base emission using a simple Gaussian component offset from the compact ring emission at spatial separations > 100 μas. Our analysis revealed a Gaussian feature centered at (ΔRA ≈ 320 μas, ΔDec. ≈ 60 μ as), projected separation of ≈ 5500 AU, with an estimated flux density of only ∼60 mJy, implying that most of the missing flux identified in previous EHT studies had to originate from different, larger scales. Brighter emission at the relevant spatial scales is firmly ruled out, and the data do not favor more complex models. This component aligns with the inferred position of the large-scale jet and is therefore physically consistent with the emission of the jet base. While our findings point to detectable jet base emission at 230 GHz, the limited coverage provided by only two intermediate baselines limits our ability to robustly reconstruct its morphology. Consequently, we treated the recovered Gaussian as an upper limit on the jet base flux density. Future EHT observations with expanded intermediate baseline coverage will be essential to constrain the structure and nature of this component with higher precision.

Sagittarius A* ( the supermassive black hole at the center of the Milky Way, provides a unique laboratory to study accretion dynamics and plasma processes near the event horizon. We investigated the variability and polarization properties of using ALMA observations during the 2018 Event Horizon Telescope campaign. We analyzed high-cadence full-polarization light curves from ALMA at millimeter wavelengths, performed time-series analysis, and investigated the temporal behavior during an X-ray flare observed by on 2018 April 24. The variability characteristics are compared with expectations from standard accretion flow models. Chandra We find low variability in total intensity (σ/μ < 10%), but significantly higher variability in linear and circular polarization (∼ 30% and ∼ 50%, respectively). A time-series analysis reveals red-noise variability, with power spectral densities between -2 and -3 across all Stokes parameters. Polarized intensity shows stable intra-day timescales, while total intensity exhibits more variable timescales, suggesting distinct emission regions, with polarization likely arising from a coherent structure. On April 24, a statistically significant inter-band delay in polarized intensity coincides with a near-simultaneous X-ray and millimeter peak that deviates from the typical delayed flare scenario. This event also features enhanced millimeter variability and coherent polarization loop evolution. The observed simultaneity challenges standard models of transient synchrotron emission with cooling delays, favoring instead a scenario of continuous energy injection in an optically thin region. Our results offer new constraints on the physical mechanisms driving variability in and provide key observational input for refining theoretical models of accretion and plasma behavior in the vicinity of supermassive black holes.

Test-Time Training (TTT) models context dependencies by adapting part of the model's weights (often referred to as fast weights) at inference time. This adapted fast weight, similar to recurrent states in RNNs, stores temporary memories of past tokens in the current sequence. Existing TTT methods have struggled to demonstrate effectiveness in handling long-sequence data, due to their computational inefficiency on modern GPUs. The TTT layers in many of these approaches operate with extremely low FLOPs utilization (often below 5%) because they deliberately apply small online mini-batch sizes (e.g., updating fast weights every 16 or 64 tokens). Moreover, a small mini-batch implies fine-grained block-wise causal dependencies in the data, making them unsuitable for data beyond 1D ordered sequences, like sets or N-dimensional grids such as images or videos. In contrast, we pursue the opposite direction by proposing an extremely large chunk update, ranging from 2K to 1M tokens across tasks of varying modalities, which we refer to as Large Chunk Test-Time Training (LaCT). This approach improves hardware utilization by orders of magnitude, and more importantly, facilitates scaling of nonlinear state size (up to 40% of model parameter size), hence substantially improving state capacity, all without requiring cumbersome and error-prone custom kernel implementations. It also allows easy integration of sophisticated optimizers like Muon for online memory updates. We validate our approach across diverse data modalities and tasks, including novel view synthesis from image sets, language models, and auto-regressive video diffusion models. Our approach can scale up to 14-billion-parameter auto-regressive video diffusion models handling sequences of up to 56K tokens. In our longest sequence experiment, we perform novel view synthesis with more than one million context length. Our results highlight the computational and performance benefits of large-chunk test-time training, paving the way for more efficient and scalable long-context sequence modeling. We hope that this work will inspire and accelerate new research in the field of long-context modeling and test-time training.

Machine learning frameworks for physical problems must capture and enforce physical constraints that preserve the structure of dynamical systems. Many existing approaches achieve this by integrating physical operators into neural networks. While these methods offer theoretical guarantees, they face two key limitations: (i) they primarily model local relations between adjacent time steps, overlooking longer-range or higher-level physical interactions, and (ii) they focus on forward simulation while neglecting broader physical reasoning tasks. We propose the Denoising Hamiltonian Network (DHN), a novel framework that generalizes Hamiltonian mechanics operators into more flexible neural operators. DHN captures non-local temporal relationships and mitigates numerical integration errors through a denoising mechanism. DHN also supports multi-system modeling with a global conditioning mechanism. We demonstrate its effectiveness and flexibility across three diverse physical reasoning tasks with distinct inputs and outputs.

Event Horizon Telescope (EHT) images of the supermassive black hole M87* depict an asymmetric ring of emission. General relativistic magnetohydrodynamic (GRMHD) models of M87* and its accretion disk predict that the amplitude and location of the ring's peak brightness asymmetry should fluctuate due to turbulence in the source plasma. We compare the observed distribution of brightness asymmetry amplitudes to the simulated distribution in GRMHD models, across varying black hole spin a(*). We show that, for strongly magnetized (MAD) models, three epochs of EHT data marginally disfavor divided by a(*)divided by less than or similar to 0.2. This is consistent with the Blandford-Znajek model for M87's jet, which predicts that M87* should have nonzero spin. We show quantitatively how future observations could improve spin constraints and discuss how improved spin constraints could distinguish between differing jet-launching mechanisms and black hole growth scenarios.

Data collected from the physical world is always a combination of multiple sources: an underlying signal from the physical process of interest and a signal from measurement-dependent artifacts from the sensor or instrument. This secondary signal acts as a confounding factor, limiting our ability to extract information about the physics underlying the phenomena we observe. Furthermore, it complicates the combination of observations in heterogeneous or multi-instrument settings. We propose a deep learning framework that leverages overlapping observations, a dual-encoder architecture, and a counterfactual generation objective to disentangle these factors of variation. The resulting representations explicitly separate intrinsic signals from sensor-specific distortions and noise, and can be used for counterfactual view generation, parameter inference unconfounded by measurement distortions, and instrument-independent similarity search. We demonstrate the effectiveness of our approach on astrophysical galaxy images from the DESI Legacy Imaging Survey (Legacy) and the Hyper Suprime-Cam (HSC) Survey as a representative multi-instrument setting. This framework provides a general recipe for scientific and multi-modal self-supervised pretraining: construct training pairs from overlapping observations of the same physical system, treat sensor- or modality-specific effects as augmentations, and learn invariant representations through counterfactual generation.

Latent diffusion models (LDMs) enable high-fidelity synthesis by operating in learned latent spaces. However, training state-of-the-art LDMs requires complex staging: a tokenizer must be trained first, before the diffusion model can be trained in the frozen latent space. We propose UNITE - an autoencoder architecture for unified tokenization and latent diffusion. UNITE consists of a Generative Encoder that serves as both image tokenizer and latent generator via weight sharing. Our key insight is that tokenization and generation can be viewed as the same latent inference problem under different conditioning regimes: tokenization infers latents from fully observed images, whereas generation infers them from noise together with text or class conditioning. Motivated by this, we introduce a single-stage training procedure that jointly optimizes both tasks via two forward passes through the same Generative Encoder. The shared parameters enable gradients to jointly shape the latent space, encouraging a "common latent language". Across image and molecule modalities, UNITE achieves near state of the art performance without adversarial losses or pretrained encoders (e.g., DINO), reaching FID 2.12 and 1.73 for Base and Large models on ImageNet 256 x 256. We further analyze the Generative Encoder through the lenses of representation alignment and compression. These results show that single stage joint training of tokenization & generation from scratch is feasible.

Human visual perception offers valuable insights for understanding computational principles of motion-based scene interpretation. Humans robustly detect and segment moving entities that constitute independently moveable chunks of matter, whether observing sparse moving dots, textured surfaces, or naturalistic scenes. In contrast, existing computer vision systems lack a unified approach that works across these diverse settings. Inspired by principles of human perception, we propose a generative model that hierarchically groups low-level motion and appearance features into particles (small Gaussians representing local matter), and groups particles into clusters capturing coherently and independently moveable physical entities. We develop a hardware-accelerated inference algorithm based on parallelized block Gibbs sampling to recover stable particle motion and groupings. Our model operates on different kinds of inputs (random dots, stylized textures, or naturalistic RGB video), enabling it to work across settings where biological vision succeeds but existing computer vision approaches do not. We validate this unified framework across three domains: on 2D random dot kinematograms, our approach captures human object perception including graded uncertainty across ambiguous conditions; on a Gestalt-inspired dataset of camouflaged rotating objects, our approach recovers correct 3D structure from motion and thereby accurate 2D object segmentation; and on naturalistic RGB videos, our model tracks the moving 3D matter that makes up deforming objects, enabling robust object-level scene understanding. This work thus establishes a general framework for motion-based perception grounded in principles of human vision.

Creating images from noise is image generation; reconstructing fine details from coarse inputs is super-resolution. Despite their practical differences, both can be understood as reversing information loss across scales. We introduce SKILD, a Scale-invariant K-Space Image Learning Diffusion model that unifies generation and continuous super-resolution within a single unconditional framework. Both natural images and critical physical systems exhibit scale invariance, and we leverage it to design a forward process that attenuates image content from fine to coarse scales while injecting spectrum-matched Gaussian noise, making scale an explicit coordinate of the diffusion dynamics. The same trained reverse process performs generation and continuous super-resolution by varying only the starting timestep: no task-specific architecture, no conditioning branch, no classifier-free guidance, no retraining per scale factor. Empirically, SKILD reaches FID 2.65 and Inception Score 9.63 on unconditional CIFAR-10, performs 2×–8× super-resolution on ImageNet from a single unconditional checkpoint while outperforming conditional models across perceptual metrics, and reconstructs critical Ising models whose connected four-point correlations closely track the ground truth.

In very-long baseline interferometric arrays, nearly co-located stations probe the largest scales and typically cannot resolve the observed source. In the absence of a large-scale structure, closure phases constructed with these stations are zero and, since they are independent of station-based errors, they can be used to probe data issues. Here, we show how these trivial closure phases become nonzero with a brightness distribution on smaller scales than their short baseline would suggest. When applied to sources that are made up of a bright compact and large-scale diffuse component, the trivial closure phases directly measure the centroid relative to the compact source and higher-order image moments. We present a technique to measure these image moments with minimal model assumptions and validate it on synthetic Event Horizon Telescope (EHT) data. We then apply this technique to 2017 and 2018 EHT observations of M87* and find a weak preference for extended emission in the direction of the large-scale jet. We also apply it to 2021 EHT data and measure the source centroid about 1 mas northwest of the compact ring, which is consistent with the jet observed at lower frequencies.

Mathematical problem solving remains a challenging test of reasoning for large language and multimodal models, yet existing benchmarks are limited in size, language coverage, and task diversity. We introduce *MathNet*, a large-scale, high-quality, multilingual, and multimodal dataset of Olympiad-level problems. MathNet spans 40 countries, 10 languages, and two decades of competitions, comprising 17,512 **expert-authored problems with solutions** across diverse domains. *MathNet* supports three tasks: (i) *mathematical comprehension*, (ii) *mathematical retrieval*, an underexplored but essential capability and (iii) *Math RAG*, which evaluates how retrieval-augmented generation improves problem solving. For retrieval, we construct 39K pairs of mathematically equivalent problems to enable equivalence-based evaluation, in addition to 70 expert-curated pairs from real competitions. Experimental results show that even state-of-the-art reasoning models (76.8% for GPT-5 and 46.8% for Claude 4.5 Opus) are challenged, while embedding models struggle to retrieve equivalent problems. Finally, we show that LLM performance in RAG-based math problem solving is highly sensitive to retrieval quality; for example, DeepSeek-V3.2-Speciale achieves gains of up to 12%, obtaining the highest scores on the benchmark. *MathNet* provides the largest high-quality Olympiad dataset and the first retrieval benchmark for problem equivalence. We publicly release both the dataset and benchmark at http://mathnet.netlify.app/.

Context. The 2017 observing campaign of the Event Horizon Telescope (EHT) delivered the first very long baseline interferometry (VLBI) images at the observing frequency of 230 GHz, leading to a number of unique studies on black holes and relativistic jets from active galactic nuclei (AGN). In total, eighteen sources were observed, including the main science targets, Sgr A* and M 87, and various calibrators. Sixteen sources were AGN. Aims. We investigated the morphology of the sixteen AGN in the EHT 2017 data set, focusing on the properties of the VLBI cores: size, flux density, and brightness temperature. We studied their dependence on the observing frequency in order to compare it with the Blandford-Konigl (BK) jet model. In particular, we aimed to study the signatures of jet acceleration and magnetic energy conversion. Methods. We modeled the source structure of seven AGN in the EHT 2017 data set using linearly polarized circular Gaussian components (1749+096, 1055+018, BL Lac, J0132-1654, J0006-0623, CTA 102, and 3C 454.3) and collected results for the other nine AGN from dedicated EHT publications, complemented by lower frequency data in the 2-86 GHz range. Combining these data into a multifrequency EHT+ data set, we studied the dependences of the VLBI core component flux density, size, and brightness temperature on the frequency measured in the AGN host frame (and hence on the distance from the central black hole), characterizing them with power law fits. We compared the observations with the BK jet model and estimated the magnetic field strength dependence on the distance from the central black hole. Results. Our observations spanning event horizon to parsec scales indicate a deviation from the standard BK model, particularly in the decrease of the brightness temperature with the observing frequency. Only some of the discrepancies may be alleviated by tweaking the model parameters or the jet collimation profile. Either bulk acceleration of the jet material, energy transfer from the magnetic field to the particles, or both are required to explain the observations. For our sample, we estimate a general radial dependence of the Doppler factor delta proportional to r(<= 0.5). This interpretation is consistent with a magnetically accelerated sub-parsec jet. We also estimate a steep decrease of the magnetic field strength with radius B proportional to r(-3), hinting at jet acceleration or efficient magnetic energy dissipation.

Current video diffusion models achieve impressive generation quality but struggle in interactive applications due to bidirectional attention dependencies. The generation of a single frame requires the model to process the entire sequence, including the future. We address this limitation by adapting a pretrained bidirectional diffusion transformer to an autoregressive transformer that generates frames on-the-fly. To further reduce latency, we extend distribution matching distillation (DMD) to videos, distilling 50-step diffusion model into a 4-step generator. To enable stable and high-quality distillation, we introduce a student initialization scheme based on teacher’s ODE trajectories, as well as an asymmetric distillation strategy that supervises a causal student model with a bidirectional teacher. This approach effectively mitigates error accumulation in autoregressive generation, allowing long-duration video synthesis despite training on short clips. Our model achieves a total score of 84.27 on the VBench-Long benchmark, surpassing all previous video generation models. It enables fast streaming generation of high-quality videos at 9.4 FPS on a single GPU thanks to KV caching. Our approach also enables streaming video-to-video translation, image-to-video, and dynamic prompting in a zero-shot manner. We release our code and pretrained models.

As the field of representation learning grows, there has been a proliferation of different loss functions to solve different classes of problems. We introduce a single information-theoretic equation that generalizes a large collection of modern loss functions in machine learning. In particular, we introduce a framework that shows that several broad classes of machine learning methods are precisely minimizing an integrated KL divergence between two conditional distributions: the supervisory and learned representations. This viewpoint exposes a hidden information geometry underlying clustering, spectral methods, dimensionality reduction, contrastive learning, and supervised learning. This framework enables the development of new loss functions by combining successful techniques from across the literature. We not only present a wide array of proofs, connecting over 23 different approaches, but we also leverage these theoretical results to create state-of-the-art unsupervised image classifiers that achieve a +8% improvement over the prior state-of-the-art on unsupervised classification on ImageNet-1K. We also demonstrate that I-Con can be used to derive principled debiasing methods which improve contrastive representation learners.

Despite the unprecedented progress in the field of 3D generation, current systems still often fail to produce high-quality 3D assets that are visually appealing and geometrically and semantically consistent across multiple viewpoints. To effectively assess the quality of the generated 3D data, there is a need for a reliable 3D evaluation tool. Unfortunately, existing 3D evaluation metrics often overlook the geometric quality of generated assets or merely rely on black-box multimodal large language models for coarse assessment. In this paper, we introduce Eval3D, a fine-grained, interpretable evaluation tool that can faithfully evaluate the quality of generated 3D assets based on various distinct yet complementary criteria. Our key observation is that many desired properties of 3D generation, such as semantic and geometric consistency, can be effectively captured by measuring the consistency among various foundation models and tools. We thus leverage a diverse set of models and tools as probes to evaluate the inconsistency of generated 3D assets across different aspects. Compared to prior work, Eval3D provides pixel-wise measurement, enables accurate 3D spatial feedback, and aligns more closely with human judgments. We comprehensively evaluate existing 3D generation models using Eval3D and highlight the limitations and challenges of current models. Project page: http://eval3d.github.io.

The Information Contrastive (I-Con) framework revealed that over 23 representation learning methods implicitly minimize KL divergence between data and learned distributions that encode similarities between data points. However, a KL-based loss may be misaligned with the true objective, and properties of KL divergence such as asymmetry and unboundedness may create optimization challenges. We present Beyond I-Con, a framework that enables systematic discovery of novel loss functions by exploring alternative statistical divergences. Key findings: (1) on unsupervised clustering of DINO-ViT embeddings, we achieve state-of-the-art results by modifying the PMI algorithm to use total variation (TV) distance; (2) supervised contrastive learning with Euclidean distance as the feature space metric is improved by replacing the standard loss function with Jenson-Shannon divergence (JSD); (3) on dimensionality reduction, we achieve superior qualitative results and better performance on downstream tasks than SNE by replacing KL with a bounded f-divergence. Our results highlight the importance of considering divergence choices in representation learning optimization.

We investigate the origin of the elliptical ring structure observed in the images of the supermassive black hole M87*, aiming to disentangle contributions from gravitational, astrophysical, and imaging effects. Leveraging the enhanced capabilities of the Event Horizon Telescope (EHT)'s 2018 array, including improved (u,v)-coverage from the Greenland Telescope, we measured the ring's ellipticity using five independent imaging methods, obtaining a consistent average value of tau = 0.08(-0.02)(+0.03) with a position angle of xi = 50.1(-7.6)(+6.2) degrees. To interpret this measurement, we compared it to general relativistic magnetohydrodynamic (GRMHD) simulations spanning a wide range of physical parameters including the thermal or nonthermal electron distribution function, spins, and ion-to-electron temperature ratios in both low- and high-density regions. We find no statistically significant correlation between spin and ellipticity in GRMHD images. Instead, we identify a correlation between ellipticity and the fraction of non-ring emission, particularly in nonthermal models and models with higher jet emission. These results indicate that the ellipticity measured from the M87* emission structure is consistent with that expected from simulations of turbulent accretion flows around black holes, where it is dominated by astrophysical effects rather than gravitational ones. Future high-resolution imaging, including space very long baseline interferometry and long-term monitoring, will be essential to isolate gravitational signatures from astrophysical effects.

We introduce RandAR, a decoder-only visual autoregressive (AR) model capable of generating images in arbitrary token orders. Unlike previous decoder-only AR models that rely on a predefined generation order, RandAR removes this inductive bias, unlocking new capabilities in decoder-only generation. Our essential design enables random order by inserting a "position instruction token" before each image token to be predicted, representing the spatial location of the next image token. Trained on randomly permuted token sequences - a more challenging task than fixed-order generation, RandAR achieves comparable performance to its conventional raster-order counterpart. More importantly, decoder-only transformers trained from random orders acquire new capabilities. For the efficiency bottleneck of AR models, RandAR adopts parallel decoding with KV-Cache at inference time, enjoying 2.5x acceleration without sacrificing generation quality. Additionally, RandAR supports inpainting, outpainting and resolution extrapolation in a zero-shot manner. We hope RandAR inspires new directions for decoder-only visual generation models and broadens their applications across diverse scenarios. Our project page is at https://rand-ar.github.io/.

Current vision systems typically assign fixed-length representations to images, regardless of the information content. This contrasts with human intelligence —and even large language models—which allocate varying representational capacities based on entropy, context and familiarity. Inspired by this, we propose an approach to learn variable-length token representations for 2D images. Our encoder-decoder architecture recursively processes 2D image tokens, distilling them into 1D latent tokens over multiple iterations of recurrent rollouts. Each iteration refines the 2D tokens, updates the existing 1D latent tokens, and adaptively increases representational capacity by adding new tokens. This enables compression of images into a variable number of tokens, ranging from 32 to 256. We validate our tokenizer using reconstruction loss and FID metrics, demonstrating that token count aligns with image entropy, familiarity and downstream task requirements. Recurrent token processing with increasing representational capacity in each iteration shows signs of token specialization, revealing potential for object / part discovery.

We present WonderWorld, a novel framework for interactive 3D scene generation that enables users to interactively specify scene contents and layout and see the created scenes in low latency. The major challenge lies in achieving fast generation of 3D scenes. Existing scene generation approaches fall short of speed as they often require (1) progressively generating many views and depth maps, and (2) time-consuming optimization of the scene geometry representations. Our approach does not need to generate multiple views, and it leverages a geometry-based initialization that significantly reduces optimization time. Another challenge is generating coherent geometry that allows all scenes to be connected. We introduce the guided depth diffusion that allows partial conditioning of depth estimation. WonderWorld creates connected and diverse 3D scenes, each generated in less than 10 seconds on a single A6000 GPU, enabling real-time user interaction and exploration. We release full code, software, and interactive demos in https://kovenyu.com/WonderWorld/.