Tampa, Florida, United States
Domain Adaptation in Neural Machine Translation
• Proposed an end-to-end algorithm for low-resource domain adaptation in Neural Machine Translation (NMT), Epi-NMT.
Split AI Architechture (SAIA)
• Introduced a Mobile/Web Framework, Split AI Architecture (SAIA) for medical diagnosis on mobile devices (e.g., Android smartphones), which dynamically determines if an image captured by the mobile device should be diagnosed by the mobile side or sent to the server side for more powerful AI solutions.
Cost-Sensitive Active Fusion (CS-AF)
• Proposed a SOTA active fusion algorithm for medical diagnosing, which identifies the capability of each CNN classifier during the training phase and dynamically assigns the weight of each classifier during the testing phase.
Class Imbalanced Image Classification
• Proposed a two-stage training framework for imbalanced image classification, which regularizes the image representation by supervised contrastive learning.
Domain Generalization for Image Recognition
• Proposed an end-to-end algorithm DADG, which leverages GAN (Generative Adversarial Network) and improves the image recognition accuracy when classifying the images in unseen genres and environments (e.g., photo dog and cartoon dog).
