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

📌 MAROKO133 Hot ai: Adobe Research Unlocking Long-Term Memory in Video World Model

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 Breaking ai: Voltage pulses allow hair-thin carbon fibers to move like

In laboratories around the world, scientists have long dreamed of building machines so small that they could grip, bend, and move objects thinner than a strand of hair. However, controlling something so tiny precisely and reversibly has been a big challenge. 

Now, a team of researchers from the Polish Academy of Sciences has shown that even a bare carbon fiber, no thicker than a human hair, can bend and straighten on command, without any direct wiring. 

Their proof-of-concept study reveals a new way to turn ordinary carbon fibers into miniature actuators. This achievement could reshape micromechanics and soft robotics by offering a simpler route to motion at the microscopic scale.

“We anticipate that these results will enrich the tool case for research in the field of soft robotics and micromechanics,” the study authors note.

The long-standing challenge of smart fibers

For years, researchers have tried to make smart fibers—materials that change shape when exposed to electricity, light, heat, or changes in acidity. Smart polymers already exist and can respond to such stimuli, altering their color or shape and then returning to their original state. 

However, when it comes to microfibers and nanofibers, things become harder. Many systems require special coatings, structural modifications, or complex fabrication steps to make the fibers responsive. This adds cost, complexity, and limits real-world use.

The core problem has been precise and reversible control. Scientists could sometimes make a fiber move, but not in a controlled and repeatable way. That is where the study authors took a different approach. 

Instead of heavily modifying the fibers, they used uncoated and unaltered carbon fibers and focused on how electricity interacts with them.

Carbon fibers are already valued in engineering. They are lighter than steel or aluminum but extremely strong. They also conduct electricity, which makes them ideal for electrochemical experiments. 

The researchers placed a single microdiameter carbon fiber inside a special electrochemical setup called a bipolar cell. This type of system has been used since the 1970s in biosensors, reactors, and batteries. Here’s how their setup worked in simple terms:

Electricity made a carbon fiber bend

The fiber was placed between two compartments filled with a liquid containing ions—charged particles such as lithium (Li⁺) and perchlorate (ClO₄⁻). The solution also contained a redox pair: benzoquinone and hydroquinone, which help drive oxidation and reduction reactions. 

When an external voltage was applied across the cell, something remarkable happened. The team compared two kinds of fibers: smooth ones and naturally rough ones. The rough fibers had tiny grooves and uneven pores on their surfaces. In these rough fibers, the distribution of pores was not symmetrical. That natural asymmetry turned out to be crucial.

When voltage was applied, ions from the liquid entered the fiber surface unevenly. On one side of the fiber, oxidation occurred; on the other, reduction. As ion insertion was stronger on one side than the other, the fiber experienced uneven tension. 

This caused it to bend. When the voltage was reversed or removed, the ions left the fiber surface. The tension disappeared, and the fiber straightened again. In short, ions moving in and out of the carbon fiber caused it to bend and unbend. 

The motion was fully reversible and depended on the applied voltage and the fiber’s length. 

Importantly, the fiber was not directly connected to a wire. The closed bipolar cell allowed simultaneous oxidation at one end and reduction at the other, enabling wireless actuation.

“We successfully used the closed bipolar cell to wirelessly actuate a freestanding carbon fiber electrochemically,” Wojciech Nogala, one of the study authors, said.

The researchers also showed that voltage pulses could be applied in cycles. By carefully controlling the pulse duration and voltage level, the fiber could move up and down repeatedly—like microscopic tweezers. This demonstrates that the system is not just a one-time effect but a controllable mechanical response.

A promising proof of concept

This study is still at the proof-of-concept stage, but its implications are wide. If simple, prefabricated asymmetric carbon fibers can act as tiny actuators, engineers may not need complicated coatings or redesigns to build micro-scale devices. 

Such fibers could be used in synthetic muscles for microrobots, in microelectromechanical systems, or in devices that need to move or grip objects at extremely small scales. The motion strength depends on voltage and fiber length, which means the system can be tuned. 

Going forward, the team plans to explore actuators based on prefabricated asymmetric carbon fibers and to optimize performance. 

If successful, this simple mechanism, driven by ions flowing in and out, could help power the next generation of soft robotic systems and microscopic tools, bringing us closer to machines that operate at the scale of cells and tiny structures.

The study is published in the journal Nature Communications.

🔗 Sumber: interestingengineering.com