Natural Language Processing 2020 Discussion
summary
Panel assesses whether 2020’s transformer wave changes practice and gives concrete takeaways: model families and deployments named (BERT, RoBERTa, TinyBERT, Marian), inference tooling and optimizers discussed (Triton, ONNX Runtime, Turbo Transformers), interpretability libraries used (LIME, SHAP, Captum), surprising empirical notes like Roberta on 300 samples and a simple regex yielding 20% engagement lift, and system limits stated plainly (neuromorphic chips experimental, GPT-3 impractical to serve at scale).
AI-written summary · about
chapters (15)
- 0:00Introduction
- 4:18NLP Imagenet
- 7:50Interpretability
- 13:48Interpretation
- 18:01Challenges
- 24:14Transformerbased
- 25:42Neuromorphic architectures
- 29:18Neuromorphic hardware
- 31:37Running big models in production
- 36:14Future of NLP
- 38:03Question
- 46:05Bias
- 48:03One more question
- 49:09Personal experience
- 53:07Career effects
related talks
01h 01mKeynote: Transformers: The Rise of ChatGPT
12mRevolutionizing Cloud-Native Applications with NLP
41mRepresentations from natural language data: successes and challenges
40mThe Future of Language Models
43mDiscovering the BERT family of models
49mBen Hall & Barbara Fusinska
59mKeynote: The past, present, and future of AI for application developers
01h 03mIt's All About AI