NTR Webinar: Towards Trustworthy Natural Language Processing

NTR organizes and hosts scientific webinars on neural networks and invites speakers from all over the world to present their recent work. 

On November 16 Katja Filippova, Google Research Berlin, Berlin, Germany, presented a technical Zoom webinar on Towards Trustworthy Natural Language Processing. 

About the webinar: 

In this talk I presented work done at Google aimed at making natural language processing more trustworthy. I first established what trustworthiness and trust mean and when we can say that a system is trustworthy. We then discussed what criteria a trustworthy NLP system should meet, before we looked at recent work done at Google towards this end.

Materials: 

• Jacovi et al. Formalizing Trust in Artificial Intelligence: Prerequisites, Causes and Goals of Human Trust in AI
• Brundage et al. Toward Trustworthy AI: Mechanisms for Supporting Verifiable Claims

• Robustness:
  • Gorman & Bedrick. We Need to Talk About Standard Splits
  • Søgaard, Ebert, Bastings, Filippova. We Need to Talk About Random Splits
• Fairness:
  • Webster et al. Measuring and Reducing Gendered Correlations in Pre-trained Models
• Explainability:
  • The Language Explainability Tool: https://pair-code.github.io/lit
  • «Will you find these shortcuts?» A Protocol for Evaluating the Faithfulness of Input Salience Methods for Text Classification

Extracting Training Data from Large Language Models

Moderator and contact: NTR CEO Nick Mikhailovsky: nickm@ntrlab.com.