Beijing Academy of Science and Technology

Perovskite RTSDs remain limited by reproducibility. We report a universal reverse-bias aging protocol that drives halogen migration and self-heals bromine Frenkel defects, lowering dark current and boosting charge collection efficiency. Validated on hundreds of APbBr3 single crystals from multiple sources, it delivers reproducible, near state-of-the-art (2%) energy resolution and a transferable recipe.

GABAergic medial septal neurons play a key role in regulating hippocampus-dependent spatial memory, but the underlying mechanisms are not yet fully understood. In male mice, we establish an object-place recognition impairment model using an acute sleep deprivation protocol. Here we show that parvalbumin-positive neurons in the medial septum regulate object-location memory specificity during encoding by modulating hippocampal place cell population activity. Using in vivo electrophysiology combined with optogenetics, we characterize the role of medial septal parvalbumin-positive neurons in object-place recognition during memory encoding. Preference for objects relocated to novel positions parallels directional shifts in place fields of remapping neuronal populations, as well as decreased co-activity among place cells in dorsal CA1 during the encoding phase. Activating medial septal parvalbumin-positive neurons during memory encoding rescues object-place recognition impairments. These findings suggest that parvalbumin-positive neurons in the medial septum play a causal role in object-place recognition memory.

Organotin compounds (OTCs) are toxic pollutants threatening ecosystems and human health, among which dioctyltin (DOCT), widely used in skin-contact textiles, can induce immune dysfunction and metabolic disorders. Although DOCT levels in textiles are strictly regulated by international standards, traditional GC-MS suffers from cumbersome derivatization, unsatisfactory repeatability, and lengthy analysis, highlighting the urgent demand for a rapid and sensitive detection approach. Herein, we developed a fast SERS-based strategy for DOCT determination using size-optimized Au@Ag core-shell nanoparticles as the substrate, which offers simple pretreatment, high efficiency, good uniformity, and excellent reproducibility. The SERS spectra and functional group vibration modes of DOCT were elucidated by density functional theory (DFT) calculations combined with experimental validation, and the peak at 301 cm-1 was identified as the characteristic peak for quantitative analysis. After extractant optimization, the method achieved a low LOD of 0.1 μg/L in real textile samples, with recoveries ranging from 86% to 108% and good linearity from 0.1 to 1000 μg/L (R2 = 0.9804). This approach provides a reliable, high-sensitivity alternative for rapid monitoring of DOCT residues in textiles.

In-context learning (ICL) has emerged as an effective solution for few-shot learning with large language models (LLMs). However, how LLMs leverage demonstrations to specify a task and learn a corresponding computational function through ICL is underexplored. Drawing from the way humans learn from content-label mappings in demonstrations, we categorize the tokens in an ICL prompt into content, stopword, and template tokens. Our goal is to identify the types of tokens whose representations directly influence LLM's performance, a property we refer to as being performance-critical. By ablating representations from the attention of the test example, we find that the representations of informative content tokens have less influence on performance compared to template and stopword tokens, which contrasts with the human attention to informative words. We give evidence that the representations of performance-critical tokens aggregate information from the content tokens. Moreover, we demonstrate experimentally that lexical meaning, repetition, and structural cues are the main distinguishing characteristics of these tokens. Our work sheds light on how LLMs learn to perform tasks from demonstrations and deepens our understanding of the roles different types of tokens play in LLMs.

Understanding 3D scene-level affordances from natural language instructions is essential for enabling embodied agents to interact meaningfully in complex environments. However, this task remains challenging due to the need for semantic reasoning and spatial grounding. Existing methods mainly focus on object-level affordances or merely lift 2D predictions to 3D, neglecting rich geometric structure information in point clouds and incurring high computational costs. To address these limitations, we introduce Task-Aware 3D Scene-level Affordance segmentation (TASA), a novel geometry-optimized framework that jointly leverages 2D semantic cues and 3D geometric reasoning in a coarse-to-fine manner. To improve the affordance detection efficiency, TASA features a task-aware 2D affordance detection module to identify manipulable points from language and visual inputs, guiding the selection of task-relevant views. To fully exploit 3D geometric information, a 3D affordance refinement module is proposed to integrate 2D semantic priors with local 3D geometry, resulting in accurate and spatially coherent 3D affordance masks. Experiments on SceneFun3D demonstrate that TASA significantly outperforms the baselines in both accuracy and efficiency in scene-level affordance segmentation.

Based on the atmospheric pressure carbon deoxidation process in steelmaking, this study calculated thermodynamic calculations under multi‐variable conditions and combined permutation feature importance (PFI) analysis to evaluate the feature importance of machine learning models. It revealed the dependence of key parameters for carbon deoxidation in molten steel on slag system composition, providing a basis for selecting appropriate slag systems tailored to the composition of target steel grades. This study also established a paradigm for high‐throughput slag system optimization research, aiding future selection of suitable slag systems for different deoxidation methods. Five machine learning algorithms were used to establish relevant models, among which Gradient Boosting‐2 demonstrated the best predictive performance on the test set, with R 2 and root mean square error values of 0.992 and 0.005, respectively. PFI analysis based on this model indicated that features such as CaO/SiO 2 , SiO 2 , and MgO exhibit significant influence under different scenarios. This study provides new insights for customized slag system design and customized steel grade production in novel clean deoxidation processes.