📌 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 Breaking ai: Lockheed Martin’s giant runway-free drone flies like

Sikorsky, a Lockheed Martin company, has unveiled its new Nomad family of vertical takeoff and landing (VTOL) aircraft, less than a year after proving the flight efficiency and reliability of its experimental rotor-blown wing drone.

The new twin-proprotor design combines a helicopter’s vertical agility with a fixed-wing aircraft’s speed and range, allowing it to take off, hover, and land vertically while cruising efficiently in forward flight. 

The system marks Sikorsky’s latest effort to advance autonomous hybrid-electric aircraft that can operate in both military and civilian roles without the need for runways.

Giant runway-free drone

“We use the term ‘family’ to point to a key attribute of the design, its ability to be scaled in size from a small Group 3 UAS to the footprint equivalent of a Black Hawk helicopter,” said Rich Benton, Sikorsky’s vice president and general manager. 

“The resulting Nomad family of drones will be adaptable, go-anywhere, runway-independent aircraft capable of land and sea-based missions across defense, national security, forestry, and civilian organizations.”

Sikorsky stated that the Nomad family will help boost operations. It will work alongside crewed aircraft like the Black Hawk to improve awareness, logistics, and strike abilities. This is especially important for operations in the Indo-Pacific, where long distances and spread-out bases present challenges.

The company announced the results of a successful flight test campaign in 2025 for the Nomad 50. This prototype has a wingspan of 10.3 feet and shows the good performance of its rotor-blown wing in terms of aerodynamics and vertical lift.

Sikorsky is now building the Nomad 100, a larger 18-foot wingspan Group 3 variant, with its first flight expected in the coming months.

The Nomad aircraft will use Sikorsky’s MATRIX autonomy technology. This technology includes software and sensors that help the aircraft plan routes, avoid obstacles, and carry out missions independently.

Developed in collaboration with DARPA, MATRIX has been tested on rotary and fixed-wing platforms and demonstrated in aerial firefighting, logistics resupply, and advanced aerial mobility missions.

Takes off like a helicopter, cruises like a jet

“Nomad represents breakthroughs for Sikorsky and the next generation of autonomous, long-endurance drones,” said Dan Shidler, Sikorsky’s director of advanced programs. 

“We are acting on feedback from the Pentagon, adopting a rapid approach and creating a family of drones that can take off and land virtually anywhere and execute the mission, all autonomously and in the hands of Soldiers, Marines, Sailors, and Airmen.”

According to Sikorsky, the Nomad series is designed for missions ranging from reconnaissance and light attack to contested logistics and humanitarian operations. 

Depending on mission requirements, the modular airframe can scale from Group 3 systems, weighing 56 and 1,320 pounds, to Group 4 and 5 classes exceeding 1,320 pounds.

Most variants will use fuel-efficient hybrid-electric propulsion, while larger models will employ conventional drivetrains for extended range and higher payload capacity.

The new family comes amid a broader shift in US military aviation toward autonomous teaming concepts, where drones operate alongside crewed aircraft to expand reach and reduce risk. 

Sikorsky’s Nomad effort aligns with the Pentagon’s push for affordable, rapidly deployable, and runway-independent systems capable of surviving in contested airspace.

Lockheed Martin shares closed slightly higher Thursday after the announcement.

🔗 Sumber: interestingengineering.com