Baidu Silicon Valley AI Lab
Around 2017, I was keen to expand my practical knowledge of machine learning, and this led me to the realization that Andrew Ng's team was looking for someone with my skills. As much as spending time away from education for any period was hard to imagine for me, working near and with Andrew Ng in a Chinese company was hard to pass. Ultimately, he left Baidu days before my start date, but I got to work on some of his ideas under a culture he had put into place at Baidu's Silicon Valley AI Lab. It felt like being in the Xerox PARC, at the peak of AI innovations.
I got to work on many, many small and big projects while at Baidu, to the point that when I was leaving, we had to make a bulleted list of about 20 things for handoff. Some of these included:
-
Stabilizing and re-architecting SwiftScribe, one of the first generally available based speech-to-text deep-learning AI-based transcription editing apps.
-
Rebuilding PaddlePaddle.org, the website for the English and Chinese documentation for the deep learning framework (like TensorFlow and PyTorch), PaddlePaddle, and an ecosystems for tens of scientists and developers to teach ML to Chinese developers.
-
Working on the PaddlePaddle framework, to add support for distributed concurrency in batch training. Spent a lot of time designing and rethinking neural network design and monitoring (among other things, designed the first version VisualDL; the TensorBoard equivalent). Also, supporting interoperabiility by building the ONNX library to convert PaddlePaddle models to run through TensorFlow/PyTorch, and vice-versa. Served on The Linux Foundation's Deep learning committee and NVIDIA's deep learning framework council during this period.
-
Helping in the international marketing efforts of Xiaodu / Baidu's DuerOS.
-
Several smaller projects involving productionizing and testing of automated speech recognition and speech synthesis models.