Please feel free to click on the following icons.
Cross-lingual adaptive response generation
ACL Outstanding Paper Award (Oral Presentation)
[Meta Reality Labs]
Adopt mSeq2seqs for cross-lingual transfer in low-resource language generation.
Multi-intent Zero-shot Detection
Enabled zero-shot intent detection with linear approximation of a user utterance using label BERT embeddings.
Contrastive Context Clustering for intent induction
An unsupervised approach to induce dialog intents with a new contrastive learning-based clustering method.
Context and Knowledge augmented SLU network. Published in Interspeech '21, ICASSP '22, EACL '23.
Enabled context and commonsense knowledge awareness for multi-intent and slot filling detection.
Learning to rank in ASR second pass rescoring
Enabled zero-shot intent detection with linear approximation of a user utterance using label BERT embeddings.
Heterogeneous data augmentation in skill routing
[Amazon Alexa NLU] Adopt generative language models for heterogeneous data augmentation to improve skill routing model robustness in tail domains.
Retinal image report generation
Generate medical descriptions for retinal images with keyword reinforced. This is a collaborative project with published papers. See more in the following link in github pages.
Read MoreInteractive Chatroom with AI chatbot
Play Now!!
A fancy chatroom for you to chat with a deep minded chatbot both in english and chinese.
Three chatbots: StackBot, MovieBot, ChickBot!!
Comic image generation
Simulated the style drawing of Naruto figures to construct new naruto characters completely by artificial intelligence with deep convolutional generative adversarial networks (GAN).
Read MoreClassify and diagnose disease possiblity of a malaria cell dataset with modified resnet50 structure implemented with pytorch, achieving testing accuracy around 85%.
Read MoreTrained machine to generate Chinese lyrics based on dataset of composed songs from four popular singers scraped from internet in Taiwan by pytorch with self-defined input words.
Read MoreDescribe a general image with 1-2 sentences by using datasets from Google's Conceptual Captions Competition. Extract and train with ~50000 images to get decent results.
Read MorePlease feel free to click on the following icons to explore more my personal life
Extracurricular and volunteering activities
Here is my personal life experiences! Click below to find more!