Learning to Perturb Word Embeddings for Out-of-distribution QA

ACL 2021 Learning to Perturb Word Embeddings for Out-of-distribution QA Seanie Lee, Minki Kang, Juho Lee and Sung Ju Hwang  QA models based on pretrained language models have achieved remarkable performance on various benchmark datasets. However, QA models do not generalize well ...

Learning to Generate Noise for Multi-Attack Robustness

ICML 2021 Learning to Generate Noise for Multi-Attack Robustness Divyam Madaan, Jinwoo Shin and Sung Ju Hwang  Adversarial learning has emerged as one of the successful techniques to circumvent the susceptibility of existing methods against adversarial perturbations. However, the...

Learning How Long to Wait: Adaptively-Constrained Monotonic Multihead Attention for Streaming ASR

IEEE 2021 Learning How Long to Wait: Adaptively-Constrained Monotonic Multihead Attention for Streaming ASR Jaeyun Song; Hajin Shim; Eunho Yang  Monotonic Multihead Attention, which allows multiple heads to learn their own alignments per head, shows great performance on simultane...

Large-Scale Meta-Learning with Continual Trajectory Shifting

ICML 2021 Large-Scale Meta-Learning with Continual Trajectory Shifting JaeWoong Shin, Hae Beom Lee, Boqing Gong and Sung Ju Hwang  Meta-learning of shared initialization parameters has shown to be highly effective in solving few-shot learning tasks. However, extending the framewo...

Hardware-Adaptive Efficient Latency Predictor for NAS via Meta-Learning

NeurIPS 2021 Hardware-Adaptive Efficient Latency Predictor for NAS via Meta-Learning Hayeon Lee, Sewoong Lee, Song Chong and Sung Ju Hwang  For deployment, neural architecture search should be hardware-aware, in order to satisfy the device-specific constraints (e.g., memory usage...

GTA: Graph Truncated Attention for Retrosynthesis

AAAI 2021 GTA: Graph Truncated Attention for Retrosynthesis Seung-Woo Seo, You Young Song, June Yong Yang, Seohui Bae, Hankook Lee, Jinwoo Shin, Sung Ju Hwang, and Eunho Yang  Retrosynthesis is the task of predicting reactant molecules from a given product molecule and is, import...

Hit and Lead Discovery with Explorative RL and Fragment-based Molecule Generation

NeurIPS 2021 Hit and Lead Discovery with Explorative RL and Fragment-based Molecule Generation Soojung Yang, Doyeong Hwang, Seul Lee, Seongok Ryu and Sung Ju Hwang  Recently, utilizing reinforcement learning (RL) to generate molecules with desired properties has been highlighted ...

FedMix: Approximation of Mixup under Mean Augmented Federated Learning

ICLR 2021 FedMix: Approximation of Mixup under Mean Augmented Federated Learning Tehrim Yoon, Sumin Shin, Sung Ju Hwang and Eunho Yang  Federated learning (FL) allows edge devices to collectively learn a model without directly sharing data within each device, thus preserving priv...

Federated Semi-Supervised Learning with Inter-Client Consistency & Disjoint Learning

ICLR 2021 (formerly ICML 2020 Workshop - Best Student Paper Award) Federated Semi-Supervised Learning with Inter-Client Consistency & Disjoint Learning Wonyong Jeong, Jaehong Yoon, Eunho Yang and Sung Ju Hwang  While existing federated learning approaches mostl...

Federated Continual Learning with Weighted Inter-client Transfer

ICML 2021 Federated Continual Learning with Weighted Inter-client Transfer  Jaehong Yoon, Wonyong Jeong, Giwoong Lee, Eunho Yang, Sung Ju Hwang  There has been a surge of interest in continual learning and federated learning, both of which are important in deep neural n...