MAROKO133 Update ai: Japanese supercomputer challenges 45-year-old theory about how sun-li

📌 MAROKO133 Update ai: Japanese supercomputer challenges 45-year-old theory about

For nearly half a century, astronomers have believed that stars like our sun eventually change the way they rotate. The theory suggested that when such stars grow old and slow down, their rotation pattern flips—causing their poles to spin faster than their equators. 

However, a new study from scientists at Nagoya University in Japan now suggests that this long-standing picture may be wrong. By running the most detailed simulations of stellar interiors ever performed, the researchers found that sun-like stars may keep the same rotation pattern for their entire lives. 

“The simulation can reproduce the sun’s observed rotation pattern almost perfectly. When we apply it to slower-rotating stars, it also matches astronomical observations and shows no anti-solar rotation,” Yoshiki Hatta, study co-author and a professor at NU, said.

Instead of flipping to the predicted anti-solar rotation, the equator continues to rotate faster than the poles even when the star becomes very slow. These findings indicate that magnetic fields inside stars play a much larger role in shaping their behavior than earlier models suggested.

Why scientists expect stars to flip their rotation

Unlike Earth, which spins as a rigid body, stars are made of extremely hot, moving gas. This means different parts of a star can rotate at different speeds—a phenomenon called differential rotation. 

In our sun, for example, the equator completes one rotation in roughly 25 days, while the polar regions take about 35 days. Scientists had long assumed that this pattern would eventually change as stars age. This is mainly because over billions of years, stars gradually lose rotational speed. 

Earlier theoretical studies suggested that slower rotation would alter the movement of gas deep inside the star. Those internal flows were expected to reorganize in a way that would make the poles spin faster than the equator—a state known as anti-solar differential rotation.

However, there was a problem. Astronomers have never clearly observed such stars. The predicted rotation pattern appeared in computer models, but real observations failed to confirm it.

To investigate the discrepancy, researchers turned to powerful numerical simulations. The team built an extremely detailed model of the interior of solar-type stars using magnetohydrodynamic simulations, which simultaneously calculate the motion of hot plasma and the behavior of magnetic fields.

High-resolution simulations reveal the hidden role of magnetism

The calculations were carried out on Fugaku, one of the most powerful supercomputers in the world. The simulation was extraordinarily detailed. Each modeled star was divided into about 5.4 billion grid points, allowing scientists to track tiny turbulent motions and magnetic structures inside the stellar interior.

This level of detail turned out to be essential. Earlier simulations used far fewer grid points, which caused magnetic fields to weaken artificially during the calculations. Due to this limitation, earlier studies underestimated how important magnetism might be in shaping stellar rotation.

When the new high-resolution simulation was run, the magnetic fields remained strong and stable. The results revealed that magnetic forces together with turbulent gas motions keep the equator rotating faster than the poles, even when the star rotates very slowly. 

“We found that these two processes, turbulence and magnetism, keep the equator spinning faster than the poles throughout the star’s life, not just when the star is young. So even though stars do slow down, the switch doesn’t happen because magnetic fields, which previous simulations missed, prevent it,” Hideyuki Hotta, one of the lead researchers and a professor at Nagoya, said.

The model also reproduced the sun’s observed rotation pattern with remarkable accuracy. When researchers applied the same simulation to stars rotating more slowly than the sun, the rotation pattern still did not flip. Instead, it remained solar-like. 

This provides a possible explanation for why astronomers have struggled to find evidence of anti-solar rotation in real stars. The simulations also uncovered another trend. As a star ages, its magnetic field steadily weakens. 

Earlier theories suggested the magnetic field might become strong again when the rotation pattern reversed, but the new results show no such revival. “Our results show that the magnetic field monotonically decreases over the stellar lifetime,” the study authors note.

Rethinking stellar evolution and magnetic activity

If confirmed, these findings could significantly change how astronomers understand the life cycles of stars. Stellar rotation influences many processes, including magnetic activity and the emission of energetic particles. 

A better picture of these processes could also improve predictions about how stellar environments affect the planets orbiting them—especially whether those planets remain suitable for life over billions of years.

At the same time, the new results are based on simulations rather than direct measurements. Observing the internal rotation of distant stars remains extremely challenging. Future research will likely test these predictions using improved astronomical observations.

The study is published in the journal Nature Astronomy.

🔗 Sumber: interestingengineering.com

📌 MAROKO133 Update ai: Adobe Research Unlocking Long-Term Memory in Video World Mo

Video world models, which predict future frames conditioned on actions, hold immense promise for artificial intelligence, enabling agents to plan and reason in dynamic environments. Recent advancements, particularly with video diffusion models, have shown impressive capabilities in generating realistic future sequences. However, a significant bottleneck remains: maintaining long-term memory. Current models struggle to remember events and states from far in the past due to the high computational cost associated with processing extended sequences using traditional attention layers. This limits their ability to perform complex tasks requiring sustained understanding of a scene.

A new paper, “Long-Context State-Space Video World Models” by researchers from Stanford University, Princeton University, and Adobe Research, proposes an innovative solution to this challenge. They introduce a novel architecture that leverages State-Space Models (SSMs) to extend temporal memory without sacrificing computational efficiency.

The core problem lies in the quadratic computational complexity of attention mechanisms with respect to sequence length. As the video context grows, the resources required for attention layers explode, making long-term memory impractical for real-world applications. This means that after a certain number of frames, the model effectively “forgets” earlier events, hindering its performance on tasks that demand long-range coherence or reasoning over extended periods.

The authors’ key insight is to leverage the inherent strengths of State-Space Models (SSMs) for causal sequence modeling. Unlike previous attempts that retrofitted SSMs for non-causal vision tasks, this work fully exploits their advantages in processing sequences efficiently.

The proposed Long-Context State-Space Video World Model (LSSVWM) incorporates several crucial design choices:

1. Block-wise SSM Scanning Scheme: This is central to their design. Instead of processing the entire video sequence with a single SSM scan, they employ a block-wise scheme. This strategically trades off some spatial consistency (within a block) for significantly extended temporal memory. By breaking down the long sequence into manageable blocks, they can maintain a compressed “state” that carries information across blocks, effectively extending the model’s memory horizon.
2. Dense Local Attention: To compensate for the potential loss of spatial coherence introduced by the block-wise SSM scanning, the model incorporates dense local attention. This ensures that consecutive frames within and across blocks maintain strong relationships, preserving the fine-grained details and consistency necessary for realistic video generation. This dual approach of global (SSM) and local (attention) processing allows them to achieve both long-term memory and local fidelity.

The paper also introduces two key training strategies to further improve long-context performance:

• Diffusion Forcing: This technique encourages the model to generate frames conditioned on a prefix of the input, effectively forcing it to learn to maintain consistency over longer durations. By sometimes not sampling a prefix and keeping all tokens noised, the training becomes equivalent to diffusion forcing, which is highlighted as a special case of long-context training where the prefix length is zero. This pushes the model to generate coherent sequences even from minimal initial context.
• Frame Local Attention: For faster training and sampling, the authors implemented a “frame local attention” mechanism. This utilizes FlexAttention to achieve significant speedups compared to a fully causal mask. By grouping frames into chunks (e.g., chunks of 5 with a frame window size of 10), frames within a chunk maintain bidirectionality while also attending to frames in the previous chunk. This allows for an effective receptive field while optimizing computational load.

The researchers evaluated their LSSVWM on challenging datasets, including Memory Maze and Minecraft, which are specifically designed to test long-term memory capabilities through spatial retrieval and reasoning tasks.

The experiments demonstrate that their approach substantially surpasses baselines in preserving long-range memory. Qualitative results, as shown in supplementary figures (e.g., S1, S2, S3), illustrate that LSSVWM can generate more coherent and accurate sequences over extended periods compared to models relying solely on causal attention or even Mamba2 without frame local attention. For instance, on reasoning tasks for the maze dataset, their model maintains better consistency and accuracy over long horizons. Similarly, for retrieval tasks, LSSVWM shows improved ability to recall and utilize information from distant past frames. Crucially, these improvements are achieved while maintaining practical inference speeds, making the models suitable for interactive applications.

The Paper Long-Context State-Space Video World Models is on arXiv

The post Adobe Research Unlocking Long-Term Memory in Video World Models with State-Space Models first appeared on Synced.

🔗 Sumber: syncedreview.com