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 Update ai: Lead-cooled nuclear reactors edge closer to reality as Fren

French firm Newcleo has submitted the design of its lead-cooled small modular reactor (SMR) to Euratom, the European regulatory body overseeing nuclear safeguards, for review.

This is part of a new regulation that came into effect this year, requiring operators to integrate Euratom safeguards into the design phase. The review process is expected to take two years. 

As countries look to reduce their carbon emissions, nuclear energy is poised to make a comeback due to its non-carbon-emitting nature. While nuclear projects of the past were marred by their gigantic scale, nuclear fission energy of the future relies on small modular reactors (SMRs) that are built and deployed at scale rather than standalone projects. 

SMRs are relatively small nuclear reactors that produce no more than a third of the output of conventional reactors. These reactors are manufactured at a central facility and can be shipped anywhere in the world due to their relatively small size. 

During manufacturing, economies of scale are applied, thereby reducing costs, while their small size also reduces deployment costs. Multiple reactor units can be teamed up to build a power plant with higher output.

Newcleo’s SMR follows the same approach but uses lead as its coolant. 

Why lead-cooled SMR?

There are multiple approaches to building SMRs, some of which Interesting Engineering has extensively covered. In a lead-cooled SMR, molten lead or a lead-bismuth eutectic is used as a coolant. The material is widely available, inexpensive, and not chemically reactive.

A lead-cooled SMR works at normal atmospheric pressure. So, the reactor design does not need to include safety measures for pressure-loss accidents. It also has a high boiling point, so no safety injection systems are required.

At the same time, this increases the reactor’s thermal efficiency to 50 percent, thereby boosting the overall power generation efficiency. 

The reactor uses convection for heat removal. This is a passive process that will occur naturally, even during a meltdown, without the need for pumps or manual intervention.

Newcleo has been working on the reactor design for over two decades and has now settled on a 200 MWe reactor design that it plans to take to market. 

Newcleo’s path ahead

As part of Euratom’s new mandatory requirement, operators of new or modified nuclear facilities must follow the 3S framework of safety, security, and safeguards. In this process, the manufacturer incorporates material control requirements at the design stage, which the competent authorities will rigorously review. 

The process is expected to take two years, during which the company will run licensing procedures with authorities such as Autorité de Sûreté Nucléaire et de Radioprotection (ASNR). By 2027, the company expects to have submitted its licence application to the French Ministry of Environment. 

“Early collaboration with European and French regulators not only builds our licensing experience but also provides invaluable knowledge that we can leverage in other jurisdictions as we progress with our licensing initiatives worldwide,” said Stéphane Calpena, Newcleo’s Group Executive Director for Licensing Affairs, in a press release.

Newcleo expects its non-nuclear precursor prototype reactor to be ready in Italy in 2026, with an investment decision on the commercial power plant in 2029.

Alongside this, the company is also building a mixed uranium/plutonium oxide (MOX) plant to produce fuel for its reactors. 

If all goes to plan, Newcleo’s first commercial power plant will be ready in France by 2032. 

🔗 Sumber: interestingengineering.com