MAE: A 3nm 0.168mm2 576MAC Mini Autoencoder with Line-Based Depth-First Scheduling for Generative AI in Vision on Edge Devices
Shih-Wei Hsieh, Chia-Hung Yuan, Ming-Hung Lin, Ping-Yuan Tsai, You-Yu Nian, Chia-Yuan Cheng, Hung-Wei Chih, Po-Han Chiang, Ming-Hsuan Chiang, Yuan-Jung Kuo, Yu-Wei We, Yi-Syuan Chen, Po-Heng Chen, Sandy Huang, Ming-En Shih, Chia-Ping Chen, Abrams Chen, ShenKai Chang, Chih-Ming Wang, Po-Yu Yeh, Jett Liu, Yung-Chang Chang, Chung-Yi Chen, Chi-Cheng Ju, CH Wang, Keven Jou
ISSCC 2025 (Highlight)
Paper

Proposed 3nm Al accelerator, which delivers 0.63 TOPS within an ultra-compact 0.168mm2 area, optimized for generative Al applications.

Meta Adversarial Perturbations
Chia-Hung Yuan, Pin-Yu Chen, Chia-Mu Yu
AAAI 2022 Workshop
Paper

Proposed a meta adversarial perturbation (MAP), a better initialization that causes data to be misclassified with high probability after being updated through only a one-step gradient ascent update.

Neural Tangent Generalization Attacks
Chia-Hung Yuan, Shan-Hung Wu
ICML 2021
Paper / Video / Slide / Code / Competition

Proposed generalization attack, where an attacker aims to modify training data in order to spoil training process such that a trained network lacks generalizability.

Adversarial Robustness via Runtime Masking and Cleansing
Yi-Hsuan Wu, Chia-Hung Yuan, Shan-Hung Wu
ICML 2020
Paper / Video / Slide / Code

Devised runtime masking and cleansing (RMC), a new defense method, to improve adversarial robustness.

TensorFlow2 Classification Model Zoo
95.76% on CIFAR-10 with TensorFlow2. A TF2 implementation of the classification models, including VGG, ResNet, DenseNet, SENet, MobileNet, etc.

Neural Networks from Scratch
A tutorial about how to build neural networks on our own, without the help of the deep learning frameworks. In this way, we can better understand deep learning and how all of the elements work.

Awesome Real-world Adversarial Examples
A curated list of awesome real-world adversarial examples resources. This repository only lists the mechanism which can be realized in the real-world, in other words, the physical attack or defense.

Scene Recognition with Bag of Words
An example of a typical bag of words classification pipeline. It begins with simplest method, tiny images and k-NN(k nearest neighbors) classification, and then move forward to bags of quantized local features and linear classifiers learned by SVC(support vector classifier).