🤖 Computer Science

A model-free method to learn multiple skills in parallel on modular robots

Legged robots need unique control schemes for unstructured environments. Simulation-based controllers don't transfer well to real-world scenarios, so rapid real-world learning methods are essential. This paper introduces a method using Central Pattern Generators to quickly learn basic locomotion skills by finding good initial states. Tests on six robot types show robots can learn locomotion in under 15 minutes, demonstrated in a targeted experiment.

Trajectory-aligned Space-time Tokens for Few-shot Action Recognition

We propose a method for few-shot action recognition that disentangles motion and appearance representations using trajectory-aligned tokens (TATs). Leveraging recent advancements in tracking and self-supervised learning, our approach reduces data needs while retaining key information. A Masked Space-time Transformer processes these representations, achieving state-of-the-art results across multiple datasets.

⚛️ Physics and Chemistry

Optimization algorithm successfully computes the ground state of interacting quantum matter

Researchers at the University of Augsburg have developed a new stochastic-reconfiguration optimization algorithm to train deep neural quantum networks with up to 106 parameters. This algorithm, described in Nature Physics, effectively computes the ground state of a quantum spin liquid using neural quantum states (NQSs). This advancement addresses previous limitations in training algorithms for quantum many-body problems.

A recipe for zero-emissions fuel: Soda cans, seawater, and caffeine

MIT engineers have discovered that exposing pure aluminum from soda cans to seawater produces hydrogen gas, which can be used for clean energy. This reaction is sped up by adding caffeine. Their study, published in Cell Reports Physical Science, demonstrates that pretreating aluminum pellets with a rare-metal alloy allows them to react with seawater to generate hydrogen, while the alloy can be recycled for sustainable hydrogen production.

Element 120 may now be in reach and the hunt for it could begin next year

Researchers at Lawrence Berkeley National Laboratory have successfully used titanium-50 to create a superheavy element, aiming to search for element 120 next year. New elements are made by fusing lighter and heavier elements in particle accelerators. These unstable, short-lived isotopes decay quickly and require specialized detection equipment.

🧬 Biology & Medicine

Two Diets Linked to Improved Cognition, Slowed Brain Aging

New research shows that both intermittent fasting (IF) and a standard healthy living (HL) diet result in weight loss, reduced insulin resistance, and slowed brain aging in older overweight adults with insulin resistance. However, neither diet affects Alzheimer's disease biomarkers. The study, led by Dimitrios Kapogiannis, MD, suggests further research on these diets for brain health optimization. Findings were published on June 19 in Cell Metabolism.

Machine learning models predict disability progression in multiple sclerosis

Machine learning models can predict multiple sclerosis (MS) disability progression, according to a study using data from 15,240 adults treated at 146 centers. The models achieved an average ROC-AUC of 0.71, indicating that disability progression history is more predictive than treatment or relapse history.

🔭 Space & Astronomy

SpaceX finds cause of Falcon 9 failure, eyes return to flight as soon as July 27

SpaceX identified and fixed the issue that caused its Falcon 9 rocket to fail on July 11, when a liquid oxygen leak in the upper stage prevented an orbit-raising burn, causing the Starlink satellites to fall back to Earth. The leak was due to a crack in a pressure sensor line, caused by engine vibrations and a loose clamp.

Thanks For Reading!

Please move this email to your Primary Inbox (if it is not already there) to make sure you don’t miss any of the exciting news and updates!
Reply directly to this email with any questions, comments or suggestions. 🚀