Oscar Chang

Oscar Chang is a 2020 research intern in the Applied Sciences Group (ASG) at Microsoft.  His general research interests revolve around deep learning and its applications.  Oscar has done extensive work on meta-learning algorithms applied to deep neural networks.  He developed principled weight initialization formulas for hypernetworks, which are meta neural networks that generate other neural networks in an end-to-end differentiable fashion.  He has also proposed gradient-based meta-learning algorithms that mitigates inner loop over-fitting and speeds up the convergence of the outer loop.  His current area of research interest is centered around deep learning models targeting speech and audio specific applications.

• Single-Channel Speech Enhancement Using Learnable Loss Mixup

  Proceedings, Interspeech 2021 August, 2021 Pages 2696-2700

  By Oscar Chang, Dung Tran, Kazuhito Koishida