Timezone: »

Compositional Generalization by Learning Analytical Expressions
Qian Liu · Shengnan An · Jian-Guang Lou · Bei Chen · Zeqi Lin · Yan Gao · Bin Zhou · Nanning Zheng · Dongmei Zhang

Wed Dec 09 09:00 AM -- 11:00 AM (PST) @ Poster Session 3 #985

Compositional generalization is a basic and essential intellective capability of human beings, which allows us to recombine known parts readily. However, existing neural network based models have been proven to be extremely deficient in such a capability. Inspired by work in cognition which argues compositionality can be captured by variable slots with symbolic functions, we present a refreshing view that connects a memory-augmented neural model with analytical expressions, to achieve compositional generalization. Our model consists of two cooperative neural modules, Composer and Solver, fitting well with the cognitive argument while being able to be trained in an end-to-end manner via a hierarchical reinforcement learning algorithm. Experiments on the well-known benchmark SCAN demonstrate that our model seizes a great ability of compositional generalization, solving all challenges addressed by previous works with 100% accuracies.

Author Information

Qian Liu (Beihang University)

My name is Qian Liu, currently a fourth-year Ph.D. in Beihang University. Meawhile, I am also very lucky to participate the joint program with Microsoft Reserach Asia. In the past few years, my reserach interest focuses on semantic parsing and dialogue. I am also familiar with reinforcement learning such as Policy Gradient, Self-play, Hierarchical Reinforcement Learning, which have been successfully employed on our reserach projects.

Shengnan An (Xi’an Jiaotong University)
Jian-Guang Lou (Microsoft)
Bei Chen (Microsoft Research Asia)
Zeqi Lin (Microsoft Research)
Yan Gao (Microsoft Research Asia, Beijing, China)
Bin Zhou (State Key Laboratory of Virtual Reality Technology and Systems, School of Computer Science and Engineering,Beihang University)
Nanning Zheng (Xi'an Jiaotong University)
Dongmei Zhang (Microsoft Research)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors