Search

Your search keyword '"Lee, Youngmoon"' showing total 47 results
Start Over Author "Lee, Youngmoon"
47 results on '"Lee, Youngmoon"'

1. AT-SNN: Adaptive Tokens for Vision Transformer on Spiking Neural Network

Author

Kang, Donghwa, Lee, Youngmoon, Lee, Eun-Kyu, Kang, Brent, Lee, Jinkyu, and Baek, Hyeongboo

Subjects
Computer Science - Artificial Intelligence, Computer Science - Computer Vision and Pattern Recognition
Abstract

In the training and inference of spiking neural networks (SNNs), direct training and lightweight computation methods have been orthogonally developed, aimed at reducing power consumption. However, only a limited number of approaches have applied these two mechanisms simultaneously and failed to fully leverage the advantages of SNN-based vision transformers (ViTs) since they were originally designed for convolutional neural networks (CNNs). In this paper, we propose AT-SNN designed to dynamically adjust the number of tokens processed during inference in SNN-based ViTs with direct training, wherein power consumption is proportional to the number of tokens. We first demonstrate the applicability of adaptive computation time (ACT), previously limited to RNNs and ViTs, to SNN-based ViTs, enhancing it to discard less informative spatial tokens selectively. Also, we propose a new token-merge mechanism that relies on the similarity of tokens, which further reduces the number of tokens while enhancing accuracy. We implement AT-SNN to Spikformer and show the effectiveness of AT-SNN in achieving high energy efficiency and accuracy compared to state-of-the-art approaches on the image classification tasks, CIFAR10, CIFAR-100, and TinyImageNet. For example, our approach uses up to 42.4% fewer tokens than the existing best-performing method on CIFAR-100, while conserving higher accuracy., Comment: 8 pages

Published
2024

9. Hydra: Resilient and Highly Available Remote Memory

Author

Lee, Youngmoon, Maruf, Hasan Al, Chowdhury, Mosharaf, Cidon, Asaf, and Shin, Kang G.

Subjects
Computer Science - Distributed, Parallel, and Cluster Computing, Computer Science - Networking and Internet Architecture
Abstract

We present Hydra, a low-latency, low-overhead, and highly available resilience mechanism for remote memory. Hydra can access erasure-coded remote memory within a single-digit microsecond read/write latency, significantly improving the performance-efficiency trade-off over the state-of-the-art -- it performs similar to in-memory replication with 1.6X lower memory overhead. We also propose CodingSets, a novel coding group placement algorithm for erasure-coded data, that provides load balancing while reducing the probability of data loss under correlated failures by an order of magnitude. With Hydra, even when only 50% of memory is local, unmodified memory-intensive applications achieve performance close to that of the fully in-memory case in the presence of remote failures and outperform the state-of-the-art solutions by up to 4.35X.

Published
2019

47. Thermal and QoS-Aware Embedded Systems

Author

Lee, Youngmoon

Subjects
Thermal Management, Battery Management, Embedded Systems, Mobile Systems
Abstract

While embedded systems such as smartphones and smart cars become essential parts of our lives, they face urgent thermal challenges. Extreme thermal conditions (i.e., both high and low temperatures) degrade system reliability, even risking safety; devices in the cold environments unexpectedly go offline, whereas extremely high device temperatures can cause device failures or battery explosions. These thermal limits become close to the norm because of ever-increasing chip power densities and application complexities. Embedded systems in the wild, however, lack adaptive and effective solutions to overcome such thermal challenges. An adaptive thermal management solution must cope with various runtime thermal scenarios under a changing ambient temperature. An effective solution requires the understanding of the dynamic thermal behaviors of underlying hardware and application workloads to ensure thermal and application quality-of-service (QoS) requirements. This thesis proposes a suite of adaptive and effective thermal management solutions to address different aspects of real-world thermal challenges faced by modern embedded systems. First, we present BPM, a battery-aware power management framework for mobile devices to address the unexpected device shutoffs in cold environments. We develop BPM as a background service that characterizes and controls real-time battery behaviors to maintain operable conditions even in cold environments. We then propose eTEC, building on the thermoelectric cooling solution, which adaptively controls cooling and computational power to avoid mobile devices overheating. For the real-time embedded systems such as cars, we present RT-TRM, a thermal-aware resource management framework that monitors changing ambient temperatures and allocates system resources to individual tasks. Next, we target in-vehicle vision systems running on CPUs–GPU system-on-chips and develop CPU–GPU co-scheduling to tackle thermal imbalance across CPUs caused by GPU heat. We evaluate all of these solutions using representative mobile/automotive platforms and workloads, demonstrating their effectiveness in meeting thermal and QoS requirements.

Published
2019

Catalog

Books, media, physical & digital resources
See catalog results