Results 91 to 100 of about 27,097 (268)

Proximal Backpropagation

CoRR, 2017
We propose proximal backpropagation (ProxProp) as a novel algorithm that takes implicit instead of explicit gradient steps to update the network parameters during neural network training. Our algorithm is motivated by the step size limitation of explicit gradient descent, which poses an impediment for optimization.
Thomas Frerix, Thomas Möllenhoff, Michael Möller 0001, Daniel Cremers   +3 more
openaire   +3 more sources

Holographic Mapping of Orbital Angular Momentum using a Terahertz Diffractive Optical Neural Network

Advanced Intelligent Discovery, EarlyView.
A compact six‐layer diffractive optical neural network enables direct recognition and spatial mapping of terahertz (THz) orbital angular momentum (OAM) beams. Fabricated by 3D printing, the system distinguishes nine OAM modes and their superpositions with high fidelity, good efficiency, and low crosstalk, offering a scalable solution for THz ...
Wei Jia, Miguel Gomez, Steve Blair, Berardi Sensale‐Rodriguez   +3 more
wiley   +1 more source

A Robust Deep Temporal Causal Discovery Platform for Single‐Cell Gene Regulatory Network Reconstruction

Advanced Intelligent Discovery, EarlyView.
scTIGER2.0 is a deep‐learning framework that infers gene regulatory networks from single‐cell RNA sequencing data. By integrating correlation, pseudotime ordering, deep learning and bootstrap‐based significance testing, it reduces false positives and reveals directional gene interactions.
Nishi Gupta, Madison Dautle, Shaoqiang Zhang, Yong Chen   +3 more
wiley   +1 more source

Solving Data Overlapping Problem Using A Class‐Separable Extreme Learning Machine Auto‐Encoder

Advanced Intelligent Systems, Volume 7, Issue 3, March 2025.
The overlapping and imbalanced data in classification present key challenges. Class‐separable extreme learning machine auto‐encoding (CS‐ELM‐AE) is proposed, which is an enhancement of ELM‐AE that better handles overlapping data by clustering points from the same class together. Applying oversampling addresses imbalanced data.
Ekkarat Boonchieng, Wanchaloem Nadda
wiley   +1 more source

Spike-based computation using classical recurrent neural networks

Neuromorphic Computing and Engineering
Spiking neural networks (SNNs) are a type of artificial neural networks in which communication between neurons is only made of events, also called spikes.
Florent De Geeter, Damien Ernst, Guillaume Drion   +2 more
doaj   +1 more source

Predicting Performance of Hall Effect Ion Source Using Machine Learning

Advanced Intelligent Systems, Volume 7, Issue 3, March 2025.
This study introduces HallNN, a machine learning tool for predicting Hall effect ion source performance using a neural network ensemble trained on data generated from numerical simulations. HallNN provides faster and more accurate predictions than numerical methods and traditional scaling laws, making it valuable for designing and optimizing Hall ...
Jaehong Park, Guentae Doh, Dongho Lee, Youngho Kim, Changmin Shin, Su‐Jin Shin, Young‐Chul Ghim, Sanghoo Park, Wonho Choe   +8 more
wiley   +1 more source

Investigation of Analog Memristor Characteristics for Hardware Synaptic Weight in Multilayer Neural Network

Advanced Intelligent Systems, Volume 7, Issue 3, March 2025.
The systematic design of memristor‐based neural network is provided by analog conductance state parameters to accurately emulate the software‐based high‐resolution weight at discrete device level. The requirement of discrete analog conductance of memristor device is measured as ≈50 states with nonlinearity value of ≈0.142 within the deviation range of ...
Jingon Jang, Yoonseok Song, Sungjun Park
wiley   +1 more source

Exploiting Ferroelectric and Spintronic Dynamics for Neural Network Computation

Advanced Intelligent Systems, EarlyView.
Ferroelectric and spintronic devices, relying on the control of polarization and magnetization, offer intrinsically fast, durable, energy‐efficient, and low‐latency building blocks for analog in‐memory computing. The hysteretic dynamics of an order parameter are leveraged to provide nonvolatile, multistate memory and nonlinear switching. Brain‐inspired
Dashiell Harrison, Kevin Nadaud, Josquin Harrison, Guillaume F. Nataf, Laura Bégon‐Lours   +4 more
wiley   +1 more source

A NEAT Approach to Evolving Neural‐Network‐Based Optimization of Chiral Photonic Metasurfaces: Application of a NeuroEvolution‐of‐Augmenting‐Topologies Pipeline

Advanced Intelligent Systems, EarlyView.
Neuro‐evolution can boost machine‐learning optimization of chiral metasurfaces. By integrating the NEAT algorithm into a deep‐learning framework, we enable the efficient design of visible‐spectrum chiroptical responses. NEAT autonomously evolves neural‐network architectures and weights, reducing manual tuning.
Davide Filippozzi, Arash Rahimi‐Iman
wiley   +1 more source

Interpretable Short‐Term Electric Load Forecasting

Advanced Intelligent Systems, EarlyView.
A temporal fusion transformer is implemented to generate day‐ahead forecasts of the hourly electrical load of a departmentbuilding at an Italian university. A forecasting performance improvement of more than 25% compared with established benchmark models and a provision of inherent robust interpretability insights reveal the potential of this model for
Alessandro Nicola, Giorgia Ghione, Vincenzo Randazzo, Vincenzo Franzitta, Marco Badami, Eros Pasero, Giansalvo Cirrincione   +6 more
wiley   +1 more source