MAROKO133 Hot ai: Adobe Research Unlocking Long-Term Memory in Video World Models with Sta

📌 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

📌 MAROKO133 Hot ai: US startup’s fusion energy device hits record plasma pressure

A US startup has just reached a new milestone in the race for fusion energy after hitting plasma pressures comparable to those found deep beneath Earth’s crust with its Fusion Z-pinch Experiment 3 (FuZE-3) device.

Researchers at Seattle’s fusion power developer Zap Energy registered plasmas with electron pressures reaching 830 megapascals (MPa), or approximately 1.6 gigapascals (GPa) total.

The results, presented this week at the American Physical Society’s Division of Plasma Physics meeting in Long Beach, California, are the highest-pressure performance ever recorded in a sheared-flow-stabilized Z pinch.

According to the company, the outcomes represent an important marker on the path to scientific energy gain, or Q>1.

“There are some big changes in FuZE-3 compared to Zap’s previous systems, and it’s great to see it perform this well so quickly out of the gate,” Colin Adams, Zap Energy’s head of experimental physics, noted.

Record-breaking pressures

Unlocking fusion energy requires extremely hot and dense plasma. The higher the pressure, the more fusion reactions occur and the more energy is produced.

Even though some fusion devices push for maximum pressure and others focus on longer confinement times, Zap Energy’s sheared-flow-stabilized Z pinches aim to balance both compression and confinement.

FuZE-3 is the US company’s newest and most advanced fusion platform. It is also the firm’s first device to incorporate a third electrode to separate the forces that drive plasma acceleration and compression.

The researchers’ latest measurements showed electron pressures reaching 830 megapascals. When both electrons and ions are accounted for, the total plasma pressure approached 1.6 gigapascals (GPa).

This is about 10,000 times atmospheric pressure at sea level, or 10 times the pressure at the bottom of the Mariana Trench. These extreme conditions were sustained for about one microsecond and measured using optical Thomson scattering, which is the gold standard for plasma diagnostics.

The recent FuZE-3 campaigns featured multiple repeated shots achieving electron densities between 3 and 5 x 1024 m-3 per cubic meter, and electron temperatures exceeding one kilo-electronvolts (keV). This is equivalent to more than 21 million degrees Fahrenheit.

Although the new results are still preliminary, they represent another step toward scientific energy gain, also known as Q > 1. This is the point at which a fusion system generates more energy than it consumes.

“We’re really just getting started with FuZE-3,” Ben Levitt, Zap Energy’s R&D vice president, stated. The company plans to continue scientific campaigns on FuZE-3 through the coming months while preparing the next-generation FuZE platform, slated to begin operation this winter.

“It was built and commissioned just recently, we’re generating lots of high-quality shots with high repeatability, and we have plenty of headroom to continue making rapid progress in fusion performance,” Levitt concluded in a press release.

🔗 Sumber: interestingengineering.com