Combining behavioral research and NLP to combat gender biases
summary
Built a live bias Eliminator app that highlights ~150 gender-coded words and suggests context-aware swaps using a Roberta masked LM with a double-mask prompt and lemma-level filtering. Continuous human edits feed a short-term one-shot similarity hash table plus a shallow mini-batch learner, protecting the base transformer from catastrophic forgetting while improving suggestions. Benchmarks cited: strongly masculine ads cut female applicants by ~13%; deployment follows MLOps practices (Kubernetes).
AI-written summary · about
chapters (9)
- 0:00Start
- 11:54Using the Transformer model to find the best word replacement
- 12:14MASK
- 16:24Keeping the human in the loop Model fine-tuning
- 21:12Hash table of similarity
- 22:53Shallow model (REST)
- 24:14Framework of augmented intelligence: Decision flow
- 26:19Next step for Bias Eliminator Bias in performance reviews
- 29:02CHANGE THE WORLD THROUGH DATA
related talks
20mDebiasing Contextualized Word Embeddings in NLP
01h 30mFairness Tales: How To Measure And Mitigate Unfair Bias in Machine Learning Models
01h 31mHow To Measure And Mitigate Unfair Bias in Machine Learning Models
52mKeynote address
30mBiases in Language Models
35mBias in NLP 101
28mBias in NLP
27mDesigning More Inclusive AI with More Inclusive Workflows