📌 TOPINDIATOURS Hot ai: Adobe Research Unlocking Long-Term Memory in Video World M

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: Google Rapidly Deploying Huge CO2 Battery Facilities That

While we’ve made major strides in generating renewable energy, storing that green power to use when the Sun isn’t shining or the wind isn’t blowing remains a major engineering challenge.

Researchers have developed many creative concepts — storing it in cranes that hoist humongous concrete blocks up and down, inside hot giant rocks, or spinning turbines by pumping water out of deep, decommissioned mines — none have yet proved practical enough for wide deployment.

Now, as IEEE Spectrum reports, a Milan-based company called Energy Dome has come up with an intriguing approach that stores energy in enormous domes that are filled with compressed carbon dioxide gas.

The idea behind the “CO2 battery” is simple. By compressing the gas using excess green power, it can later be depressurized to spin large turbines. A fully charged facility can store a formidable 200 megawatt-hours of electricity — enough to power around 6,000 homes for a full day.

To charge, the battery uses a thermal-energy storage system to cool the CO2 down to ambient pressure, and a condenser turns it into a liquid over a span of ten hours. To discharge it, the CO2 is evaporated and heated to power the turbine.

The goal is to bridge the gap between when renewable energy is available and when it’s actually needed through a “long-term duration energy storage” (LDES) solution. For instance, solar energy generation may hit its peak during the day, but peak household demand lags hours behind when people are home in the evening.

The idea has even caught the attention of Google, which announced a partnership with Energy Dome earlier this year. Now, IEEE Spectrum reports that the tech giant “plans to rapidly deploy the facilities in all of its key data-center locations in Europe, the United States, and the Asia-Pacific region.”

Energy Dome is currently working on a pilot CO2 battery built on five hectares of flat land in Sardinia, Italy. If successful, it wants to expand rapidly, popping up similar facilities across the world, starting with a separate plant in Karnataka, India. Authorities are also working on laying the groundwork for another in Wisconsin.

Google’s senior lead for energy strategy, Ainhoa Anda, told IEEE Spectrum that one big benefit of the approach is that it’s one-size-fits-all.

“We’ve been scanning the globe seeking different solutions,” he said, adding that “standardization is really important, and this is one of the aspects that we really like” about Energy Dome.

“They can really plug and play this,” Anda added.

The tech giant is looking to start in places where the electricity grid is already reliable and has a surplus of renewable energy that needs to be stored. Nearby data centers can then tap into the CO2 battery.

Unlike other renewable energy storage solutions, CO2 batteries don’t need special minerals, supply chains for complex parts, or constant upkeep.

And it’s not just Google looking to harness the benefits of LDES. China is also working on constructing CO2 batteries, according to IEEE.

Nonetheless, questions surrounding the concept’s long-term economic viability remain. For one thing, a CO2 battery’s footprint is considerably larger than a lithium-ion battery storage facility. There’s also the shortcoming that plagues all bubbles: the threat of a puncture, which could release thousands of tons of CO2 into the atmosphere.

But proponents argue it’s worth the risks.

“It’s negligible compared to the emissions of a coal plant,” Energy Dome CEO Claudio Spadacini told IEEE Spectrum.

More on renewable energy storage: The Amount of Electricity Generated From Solar Is Suddenly Unbelievable

The post Google Rapidly Deploying Huge CO2 Battery Facilities That Store 200 Megawatt Hours of Power appeared first on Futurism.

🔗 Sumber: futurism.com