In a one-hour panel, I had the opportunity to discuss topics related to inclusivity in scientific conferences with Sabine Weber, Naomi Saphra, Diana Galván, and Salomey Osei. The discussions included many topics such as tiered conference pricing, payment options, and how to promote leadership from underrepresented communities in the scientific D&I scene.

The even page is available here and recordings are available on youtube.

[Slides] [Paper] [ML Collective]

A talk at ML Collective Deep Learning Classics and Trends reading group. The talk is about Generation with Distributional Control” (GDC) a novel framework for controlled NLG that enjoys great flexibility by being able to define “pointwise’” and “distributional” constraints over target language models.

[slides] [paper]

A talk with Muhammad Khalifa, We survey the historical development of controlled NLG. We then introduce “Generation with Distributional Control” (GDC) a novel framework for controlled NLG that formalizes the problem as constraint satisfaction over a probability distribution. This framework enjoys great flexibility by being able to define “pointwise’” and “distributional” constraints over target language models. We will then wrap up by showing a summary of experiments over various control conditions demonstrating this framework’s effectiveness for Controlled NLG and its potential as a remedy to the problem of Bias in Language Models.

UCL - Bloomberg - London Feb. 2019

ML models fail due to Domain shift. However, Domain-shift detection metrics are model and task-dependent
In this Talk I present:

• Task dependent modifications to domain-shift detection metrics

• A method to predict the performance drop of ML models

  • Cheap & Fast without target domain annotations

  • Accurate ± 2.15% Sentiment analysis & ± 0.89 % POS tagging

[Paper] [Slides] [Blogpost]

[slides] [blogpost]

EMNLP2019, Hongkong.

[slides]

EMNLP2019, Hongkong.

[Slides]