Results 131 to 140 of about 5,008 (215)

Transformers are Bayesian Networks

CoRR
Transformers are the dominant architecture in AI, yet why they work remains poorly understood. This paper offers a precise answer: a transformer is a Bayesian network. We establish this in five ways. First, we prove that every sigmoid transformer with any weights implements weighted loopy belief propagation on its implicit factor graph.
openaire   +3 more sources

Probing the Energy Landscape of α‐Synuclein Amyloid Fibril Formation by Systematic K‐to‐Q Mutagenesis

Advanced Science, EarlyView.
We use lysine‐to‐glutamine mutations to study the effect of electrostatics on the kinetics and thermodynamics of alpha‐synuclein amyloid fibril formation. We find that mutational effects map on their structural context within fibrils and identify residues that modulate the energy landscape of alpha‐synuclein self‐assembly. Our work outlines a scalable,
Antonin Kunka, Azad Farzadfard, Jacob Aunstrup Larsen, Rasmus Krogh Norrild, Federica Saraceno, Celia Fricke, Hossein Mohammad‐Beigi, Ahmed Sadek, Jonas Folke, Susana Aznar, Alexander K. Buell   +10 more
wiley   +1 more source

Smart Bioinks for 4D Bioprinting: Requirements, Design, and Applications

Advanced Science, EarlyView.
Smart bioinks empower 4D‐bioprinted constructs to dynamically adapt and remodel in response to stimuli, effectively biomimicking native tissues. Artificial Intelligence (AI) and Machine Learning (ML) play a guiding role in their rational design by optimizing relevant properties.
Shangsi Chen, Jiahui Lai, Qiongjiao Zeng, Liangbin Zhou, Boguang Yang, Bin Zhang, Min Wang, Jiajing Zhou, Kieran Lau, Khoon S. Lim, Zhong Alan Li, Rocky S. Tuan   +11 more
wiley   +1 more source

Retrospective Grid Cells in the Medial Entorhinal Cortex Encode Past Paths During Spatial Navigation

Advanced Science, EarlyView.
Retrospective grid cells are identified in the medial entorhinal cortex, with spatial firing patterns that become most grid‐like when activity is projected onto recently traversed locations. Their coding loses spatial precision in darkness and is disrupted during assisted movement, indicating that entorhinal representations incorporate recent path ...
Ruojin Liu, Xin Yuan, Zilong He, Hong Jin, Jiayi Tian, Cong Chen, Yuanjing Liu, Xiang Zhang, Shidan Wen, Emilio Kropff, Chenglin Miao   +10 more
wiley   +1 more source

Functional MXene Inks for Printed Electronics: Chemistry, Rheology, and Devices

Advanced Science, EarlyView.
Functional MXene inks are emerging as building blocks for MXetronics, connecting MXene synthesis, surface functionalization, ink formulation, and advanced printing with device‐level applications. This review maps how composition, stability, rheology, and processability govern printed structures and enable high‐performance sensors, energy‐storage ...
Shahzaib Ali, Faisal Shahzad, Hamza Hussain, Muhammad Abbas, Dae‐Hyun Cho, Doyoung Byun, Hassan A. Arafat   +6 more
wiley   +1 more source

LLMs as Representation Generators for Gradient‐Free Scientific Optimization: A Case Study in Battery Fast Charging

Advanced Science, EarlyView.
Large language models serve as representation generators for gradient‐free scientific optimization. Two complementary approaches produce executable battery‐charging policies either directly or through LLM‐designed neural representations coupled to Bayesian optimization.
Ge Lei, Ferran Brosa Planella, Sterling G. Baird, Samuel J. Cooper   +3 more
wiley   +1 more source

An Expertise Transfer Framework For Autonomous Surgical Assistance

Advanced Science, EarlyView.
This study introduces an expertise transfer framework for procedure‐spanning autonomous surgical assistance. By emulating expert logic through hierarchical perception, attention modeling, and knowledge graph‐based decision‐making, the system provides near‐expert surgical view assistance.
Yuan Gao, Guangdi Chu, Wei Jiao, Yuchuan Zhong, Hang Yuan, Chengjun Li, Jinhua Li, Shuxin Wang, Bo Guan, Jianchang Zhao, Lizhi Pan, Haitao Niu, Jianmin Li   +12 more
wiley   +1 more source

Q‐LEAP: Millisecond Hyperdimensional Optimization for Full‐Spectrum Optical Metamaterials

Advanced Science, EarlyView.
Q‐LEAP integrates physics‐informed residual machine learning with factorization‐machine‐encoded quantum annealing to design full‐spectrum optical metamaterials. It explores a 2108 design space and, in a single 2.56 ms annealing step, reaches 85.83% of the theoretical FoM limit, enabling selective 5‐8 µm emission with 3–5 and 8–14 µm suppression and ∼40×
Zikang Guo, Yunwen Wu, Yuhui Guo, Shenghong Ju   +3 more
wiley   +1 more source

Omnidirectional Interaction Networks Modulating Phenotypic Plasticity in the Gut Microbiota Towards Ulcerative Colitis

Advanced Science, EarlyView.
This study employs a spouse‐pair cohort design and an idopNetwork model that integrates evolutionary game theory, predator‐prey dynamics, and developmental modularity theory to reconstruct maximally informative microbial interaction networks. This approach reveals how interspecies interactions drive the phenotypic plasticity of the gut microbiota in ...
Huizhen Li, Jiayi Li, Ang Dong, Xiaoyan Guo, Qiuyue Tu, Zelin Feng, Haojia Lin, Yanru Zhang, Yuxuan Hao, Huiying Gong, Yihan Meng, Yu Wang, Shing‐Tung Yau, Xiaocang Cao, Rongling Wu, Xinjuan Liu   +15 more
wiley   +1 more source

Hardware Acceleration of Stochastic Neural Networks Enabled by Bit‐Cell Level Co‐Design of Magnetic Tunnel Junctions

Advanced Science, EarlyView.
An MTJ‐based probabilistic processing‐in‐memory architecture unites nonvolatile storage, local entropy generation, and stochastic computation within each memory cell. Stored multi‐bit weights are converted directly into programmable stochastic bitstreams, enabling neural‐network inference, adaptive Ising optimization, and uncertainty‐aware Bayesian ...
Qiuyuan Wang, Dooyong Koh, Brooke McGoldrick, Tudor Mocioi, Yabin Fan, Marc Baldo, Luqiao Liu   +6 more
wiley   +1 more source