📌 TOPINDIATOURS Eksklusif ai: Adobe Research Unlocking Long-Term Memory in Video W

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

📌 TOPINDIATOURS Hot ai: New liquid-based cooling approach promises zero-emission r

Researchers from the Chinese Academy of Sciences have overcome a long-standing bottleneck in refrigeration by developing a new cooling method that could cut carbon emissions to zero.

Led by Prof. Li Bing at the Institute of Metal Research, the team introduced a technique based on the dissolution barocaloric effect, offering a clean alternative to traditional vapor-compression refrigeration.

Cooling systems are essential to modern life, from food storage to data centers, but they come at a steep environmental cost.

Conventional refrigeration relies heavily on electricity and contributes significantly to global carbon emissions.

Solid-state cooling has long been viewed as a greener option, but its real-world use has been limited. The main challenge has been poor heat transfer, which prevents solid coolants from working efficiently at scale.

Cooling without solid limits

The researchers found a way around this barrier by combining solid cooling effects with liquid flow. While studying the salt ammonium thiocyanate, they observed that when the salt dissolves in water, it releases large amounts of heat.

Applying pressure reverses the process, causing the salt to precipitate again. This reversible cycle allows continuous cooling when pressure is applied and released, making it suitable for refrigeration systems.

Unlike conventional solid-state cooling, where heat struggles to move across material boundaries, this method integrates the refrigerant and heat-transfer medium into a single flowing liquid.

The approach solves what researchers describe as the “impossible triangle” of caloric materials by delivering low emissions, high cooling power, and efficient heat transfer at the same time.

Lab experiments showed striking performance. At room temperature, the system produced a temperature drop of nearly 30 kelvins in just 20 seconds. At higher temperatures, the cooling span reached up to 54 kelvins.

These figures far exceed the performance of existing solid-state barocaloric materials. Simulations of a prototype cooling cycle also suggested a cooling capacity of 67 joules per gram, with efficiency approaching 77 percent.

Built for real machines

Using in-situ spectroscopic techniques, the team confirmed that the cooling process is stable, reversible, and responds instantly to pressure changes.

These characteristics are critical for real-world refrigeration systems that must operate reliably over long periods.

The technology departs from traditional cooling principles that rely on gas compression or solid phase changes.

By turning the coolant into a pumpable fluid, the system can move directly through heat exchangers, simplifying design and boosting performance.

This opens the door to zero-emission refrigeration for industrial facilities and households alike.

The strong performance at high temperatures also makes the technology particularly attractive for cooling next-generation artificial intelligence computing centers, where heat loads are extreme and energy efficiency is critical.

The researchers say the method could help reshape how cooling systems are designed, reducing both energy use and carbon emissions across multiple industries.

The study was published in Nature.

🔗 Sumber: interestingengineering.com