Robustness to Label Noise Depends on the Shape of the Noise Distribution in Feature Space

no code implementations • 2 Jun 2022 • Diane Oyen, Michal Kucer, Nick Hengartner, Har Simrat Singh

However, for the special case of class-dependent label noise (independent of features given the class label), the tipping point can be as low as 50%.

Analysis on Image Set Visual Question Answering

no code implementations • 31 Mar 2021 • Abhinav Khattar, Aviral Joshi, Har Simrat Singh, Pulkit Goel, Rohit Prakash Barnwal

We tackle the challenge of Visual Question Answering in multi-image setting for the ISVQA dataset.

Question Answering Visual Question Answering

Zero-Shot Language Transfer vs Iterative Back Translation for Unsupervised Machine Translation

1 code implementation • 31 Mar 2021 • Aviral Joshi, Chengzhi Huang, Har Simrat Singh

This work focuses on comparing different solutions for machine translation on low resource language pairs, namely, with zero-shot transfer learning and unsupervised machine translation.

Transfer Learning Translation +1

Lexically-constrained Text Generation through Commonsense Knowledge Extraction and Injection

no code implementations • 19 Dec 2020 • Yikang Li, Pulkit Goel, Varsha Kuppur Rajendra, Har Simrat Singh, Jonathan Francis, Kaixin Ma, Eric Nyberg, Alessandro Oltramari

Conditional text generation has been a challenging task that is yet to see human-level performance from state-of-the-art models.

Common Sense Reasoning Conditional Text Generation +3