The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Deep Learning for NLP: From the Trenches with Charlene Chambliss - #433
Dec 3, 2020
Charlene Chambliss, a Machine Learning Engineer at Primer AI with expertise in NLP, discusses her unique transition from psychology to data science. She shares insights on working with BERT models, detailing projects like her multilingual BERT initiative and a COVID-19 classifier. The conversation dives into challenges in data labeling, the use of innovative techniques for topic drift, and debugging NLP models. Charlene also offers advice for those looking to shift into tech from non-technical backgrounds, emphasizing the importance of mentorship.
AI Snips
Chapters
Transcript
Episode notes
Multilingual NER Project
- Charlene Chambliss built multilingual NER models for IQT Labs' machine translation models.
- These models highlighted entities in Russian and English translations for quality assessment.
ML for ML Assessment
- The NER models were used to assess the quality of machine translation models.
- This involved highlighting entities like names to check for mistranslations.
Fast Tokenizers
- Use Hugging Face Transformers' new Fast Tokenizers.
- They simplify aligning text spans with tokens, eliminating boilerplate code.