[PDF] CLOSURE: Assessing Systematic Generalization of CLEVR Models | Semantic Scholar

[PDF] CLOSURE: Assessing Systematic Generalization of CLEVR Models | Semantic Scholar

The CLEVR dataset of natural-looking questions about 3D-rendered scenes has recently received much attention from the research community. A number of models have been proposed for this task, many of which achieved very high accuracies of around 97-99%. In this work, we study how systematic the generalization of such models is, that is to which extent they are capable of handling novel combinations of known linguistic constructs. To this end, we define 7 additional question families which test models’ understanding of similarity-based references (such as e.g. “the object that has the same size as ...”) in novel contexts. Our experiments on the thereby constructed CLOSURE benchmark show that state-of-the-art models often do not exhibit systematicity after being trained on CLEVR. Surprisingly, we find that the explicitly compositional Neural Module Network model also generalizes badly on CLOSURE, even when it has access to the ground-truth programs at test time. We improve the NMN’s syste

1 mentions: @nlpmattg
Date: 2020/01/15 17:21

Referring Tweets

@nlpmattg A follow-on to this paper by most of the same folks: t.co/gJFLRtFGXU t.co/t6qlqrhubl

Related Entries

Read more [MIRU2018] Global Average Poolingの特性を用いたAttention Branch Network
2 users, 0 mentions 2018/08/10 09:23
Read more Recommendation System Series Part 2: The 10 Categories of Deep Recommendation Systems That Academic ...
0 users, 12 mentions 2019/12/29 03:51
Read more GitHub - harvardnlp/seq2seq-attn: Sequence-to-sequence model with LSTM encoder/decoders and attentio...
0 users, 0 mentions 2018/04/22 03:40
Read more [1710.10903] Graph Attention Networks
1 users, 1 mentions 2019/10/15 03:48
Read more [1906.01861] GRAM: Scalable Generative Models for Graphs with Graph Attention Mechanism
0 users, 1 mentions 2019/07/16 15:46