Sinica Podcast Andrew Ng on artificial intelligence and startup culture from Beijing to Silicon Valley
Sep 8, 2016
Andrew Ng, leading AI researcher and entrepreneur who co-founded Coursera and led Baidu Research. He discusses China vs US AI progress, differences in engineering and startup cultures, end-to-end speech recognition for Mandarin, self-driving car deployment strategies, AR on smartphones, and how data and big companies shape AI’s future.
AI Snips
Chapters
Books
Transcript
Episode notes
Silicon Valley And Beijing Offer Complementary Strengths
- Silicon Valley and Beijing are the two premier tech ecosystems, each missing strengths the other has developed.
- China leads in mobile/product innovation and rapid iteration while the US retains advantages in GPU hardware and some foundational resources.
End-To-End Deep Learning Outperforms Phoneme Models
- Deep learning learns direct mappings from raw audio to text, removing human-created intermediate representations like phonemes.
- Andrew Ng's team found end-to-end neural models transcribe Mandarin phrases more accurately than typical humans on short out-of-context clips.
AI Generalizes Across Languages But Needs Massive Data
- The same deep-learning architecture generalizes across languages with minimal changes; Mandarin and English require few adaptations.
- The bottleneck is data hunger: models need enormous labeled audio corpora that take years to collect and train on.
