A5006968060- Advanced Memory and Neural Computing
- Domain Adaptation and Few-Shot Learning
- Ferroelectric and Negative Capacitance Devices
- Multimodal Machine Learning Applications
- Neuroscience and Neural Engineering
- Machine Learning and ELM
- Semiconductor materials and devices
- Neural dynamics and brain function
- Transition Metal Oxide Nanomaterials
- Neural Networks and Reservoir Computing
- Conducting polymers and applications
- Photoreceptor and optogenetics research
Sungkyunkwan University
2021-2024
We consider class incremental learning (CIL) problem, in which a agent continuously learns new classes from incrementally arriving training data batches and aims to predict well on all the learned so far. The main challenge of problem is catastrophic forgetting, for exemplar-memory based CIL methods, it generally known that forgetting commonly caused by classification score bias injected due imbalance between old (in exemplar-memory). While several methods have been proposed correct such...
The field of biomimetic electronics that mimic synaptic functions has expanded significantly to overcome the limitations von Neumann bottleneck. However, scaling down technology led an increasingly intricate manufacturing process. To address issue, this work presents a one-shot integrable electropolymerization (OSIEP) method with remote controllability for deposition elements on chip by exploiting bipolar electrochemistry. Condensing synthesis, deposition, and patterning into single...
Abstract Reversible metal‐filamentary mechanism has been widely investigated to design an analog resistive switching memory (RSM) for neuromorphic hardware‐implementation. However, uncontrollable filament‐formation, inducing its reliability issues, a fundamental challenge. Here, RSM with 3D ion transport channels that can provide unprecedentedly high and robustness is demonstrated. This architecture realized by laser‐assisted photo‐thermochemical process, compatible the back‐end‐of‐line...
Brain-inspired neuromorphic computing systems, based on a crossbar array of two-terminal multilevel resistive random-access memory (RRAM), have attracted attention as promising technologies for processing large amounts unstructured data. However, the low reliability and inferior conductance tunability RRAM, caused by uncontrollable metal filament formation in uneven switching medium, result lower accuracy compared to software neural network (SW-NN). In this work, we present highly reliable...
Reservoir computing (RC) system is based upon the reservoir layer, which non-linearly transforms input signals into high-dimensional states, facilitating simple training in readout layer-a linear neural network. These layers require different types of devices-the former demonstrated as diffusive memristors and latter prepared drift memristors. The integration these components can increase structural complexity RC system. Here, a reconfigurable resistive switching memory (RSM) capable...
Convolutional weight mapping plays a stapling role in facilitating convolution operations on Processing-in-memory (PIM) architecture which is, at its essence, matrix-vector multiplication (MVM) accelerator. Despite importance, convolutional methods are under-studied and existing fail to exploit the sparse redundant characteristics of heavily quantized weights, leading low array utilization ineffectual computations. To address these issues, this paper proposes novel weight-aware activation...
We consider class incremental learning (CIL) problem, in which a agent continuously learns new classes from incrementally arriving training data batches and aims to predict well on all the learned so far. The main challenge of problem is catastrophic forgetting, for exemplar-memory based CIL methods, it generally known that forgetting commonly caused by classification score bias injected due imbalance between old (in exemplar-memory). While several methods have been proposed correct such...
Resistive Switching Memory In article number 2213064, Hui Joon Park, Jong Hwan Ko, Daeho Lee, and co-workers demonstrate analog resistive switching memory (RSM) with unprecedentedly high reliability robustness by laser-assisted photo-thermochemical process. This is compatible back-end-of-line process flexible format. With its superior characteristics, practical adaptive learning rule designed applied to ultrasonic tissue-classification task computing accuracy. RSM also has reconfigurability...
The ReRAM-based neuromorphic computing system (NCS) has been widely used as an energy-efficient platform for deep neural network (DNN) acceleration. However, ReRAM commonly suffers from stuck-at-fault (SAF), resulting in permanent device failure. SAF tolerance is essential task to ensure the reliability of by minimizing DNN inference accuracy degradation. Since hardware-based solutions incur additional overhead and power consumption, it necessary seek a solution that can be executed offline...