Introducing "Neural Module Networks for Reasoning over Text" accepted to ICLR 2020
w/ Kevin Lin (@nlpkevinl), Dan Roth (@dannydanr), Sameer Singh (@sameer_) and Matt Gardner (@nlpmattg).
Paper: arxiv.org/abs/1912.04971
Code: Coming soon! [1/4]
w/ Kevin Lin (@nlpkevinl), Dan Roth (@dannydanr), Sameer Singh (@sameer_) and Matt Gardner (@nlpmattg).
Paper: arxiv.org/abs/1912.04971
Code: Coming soon! [1/4]
@nlpkevinl @dannydanr @sameer_ @nlpmattg We extend NMNs (@jacobandreas) for text by
(a) introducing modules that reason over a paragraph, performing symbolic reasoning (such as arithmetic, sorting, counting) over numbers and dates in a probabilistic and differentiable manner; [2/4]
(a) introducing modules that reason over a paragraph, performing symbolic reasoning (such as arithmetic, sorting, counting) over numbers and dates in a probabilistic and differentiable manner; [2/4]
@nlpkevinl @dannydanr @sameer_ @nlpmattg @jacobandreas (b) proposing an unsupervised auxiliary loss to help extract arguments associated with events.
Additionally, limited amount of heuristically-obtained question program and intermediate module output supervision provides sufficient inductive bias for accurate learning. [3/4]
Additionally, limited amount of heuristically-obtained question program and intermediate module output supervision provides sufficient inductive bias for accurate learning. [3/4]
Our model outperforms state-of-the-art models on a subset of the DROP dataset (@ddua17).
I am particularly interested in this research direction of explicit compositional models for reasoning. Exciting challenges ahead; several of which we lay out in our paper as well. [4/4]
I am particularly interested in this research direction of explicit compositional models for reasoning. Exciting challenges ahead; several of which we lay out in our paper as well. [4/4]
