China's massive reforestation effort has transformed more than 77 million hectares of land, but new research shows that planted forests face different drought risks from natural forests. Using satellite observations from 1990 to 2024 and machine learning analysis, researchers found that planted forests are more vulnerable to rapid-onset flash droughts but less prone to prolonged slow droughts. The differences are linked to forest structure: natural forests have greater species diversity, more complex canopies, and deeper root systems that help buffer sudden water shortages, while planted forests are typically younger and less diverse. The findings highlight a trade-off in drought resilience and suggest that future forest management should prioritize mixed-species, structurally diverse forests to better withstand increasing climate extremes.

A new commentary in Forest Ecosystems warns that peer review in forestry journals is under mounting pressure: published articles have quadrupled since 2000 while the reviewer workforce has shrunk. The authors examine how AI tools could help fill the gap, finding they are useful for catching technical errors but cannot replace the deeper judgement of human experts. They call on journals to experiment with new review models, consider compensating reviewers, and develop clear policies for responsible AI use.

Cobalt-free Ni-rich layered cathodes are promising for high-energy lithium-ion batteries because they can reduce material cost and cobalt dependence. However, their practical use is limited by lattice oxygen release and structural degradation under high-voltage operation. Researchers have now developed a synergistic Fe/Zr codoping strategy for LiNi_0.8Mn_0.2O₂ cathodes. Fe helps regulate the local transition-metal–oxygen electronic environment, while Zr acts as a lattice anchor through strong Zr–O bonding. The codoped cathode shows improved cycling stability, rate capability, and lithium-ion transport kinetics.

Researchers have developed a low-cost route to high-performance sodium-ion battery anodes by transforming coal-tar-pitch-derived carbon into a host for ZnO/ZnSe heterostructures. The zinc-aluminum layered double hydroxide template acts both as a structural guide and a pore-forming agent, enabling closely coupled ZnO/ZnSe-carbon interfaces. The resulting anode delivered 637.5 mAh g⁻¹ at 100 mA g⁻¹ and retained 259.7 mAh g⁻¹ after 1000 cycles at 5 A g⁻¹.

As large models advance, there’s growing demand to use knowledge distillation to produce smaller, more portable models (student) that match higher-performing larger models (teacher). However, when a teacher’s capacity far exceeds the student’s, distillation often degrades, which is known as capacity mismatch. This mismatch caps the student model’s performance and has become a bottleneck in large-model distillation. Existing mitigation techniques remain ad hoc, and no study has yet systematically explained its root causes or proposed targeted methods to enable larger teachers to yield better distillation results.

In recent years, domain generalization (DG) has garnered significant attention for its goal of learning models from multiple source domains to mitigate domain shift and enable generalization to unseen target domains. Most existing methods e.g., domain-invariant representation, focus on learning a universal sample-to-label mapping (function) across domains, overlooking semantically intrinsic domain-specific information and failing to maintain invariance across diverse unseen target domains.

Skill-based reinforcement learning has become the mainstream approach to solve sparse-rewards decision making tasks. The skills extracted from the demonstration datasets provide the temporal abstraction. However, in previous skill-based RL methods, the skills are kept fixed during online learning, which brings in sub-optimal asymptotic performance when the dataset contains only sub-optimal behavior modes.

With the growing use of multiple social platforms, aligning user identities across networks, known as Social Network Alignment (SNA), has become crucial for personalized services, fraud detection, and online safety. However, existing SNA methods face significant challenges when dealing with sparse connections, heterogeneous structures, and dynamic changes in real-world social networks.

Artificial intelligence is beginning to transform not only drug discovery but also the design of advanced biomaterials. In a new study, researchers combined machine learning with laboratory validation to develop a data-driven strategy for identifying bioactive nucleoside hydrogels for periodontitis treatment. By integrating predictive modeling with innovative molecular scoring methods, the work aims to streamline biomaterial discovery and accelerate the development of targeted therapies for oral health applications.

The cyber-attack diagnosability (CA-diagnosability) of discrete event systems (DESs) assess the ability to diagnose issues when an attacker interferes with sensor-to-diagnostic communication. In real life, for example, in network systems, a network node (router or user) may be attacked or unable to correctly and promptly transmit messages due to other reasons; in power systems, a current or voltage detector may naturally age or be affected by external factors, thus failing to accurately obtain the current and voltage values of the power system; in automotive systems, damage to a component may cause the sensor readings to differ from the original values, and so on.