Analysing Mathematical Reasoning Abilities of Neural Models

Abstract

Mathematical reasoning---a core ability within human intelligence---presents some unique challenges as a domain: we do not come to understand and solve mathematical problems primarily on the back of experience and evidence, but on the basis of inferring, learning, and exploiting laws, axioms, and symbol manipulation rules. In this paper, we present a new challenge for the evaluation (and eventually the design) of neural architectures and similar system, developing a task suite of mathematics problems involving sequential questions and answers in a free-form textual input/output format. The structured nature of the mathematics domain, covering arithmetic, algebra, probability and calculus, enables the construction of training and test splits designed to clearly illuminate the capabilities and failure-modes of different architectures, as well as evaluate their ability to compose and relate knowledge and learned processes. Having described the data generation process and its potential future expansions, we conduct a comprehensive analysis of models from two broad classes of the most powerful sequence-to-sequence architectures and find notable differences in their ability to resolve mathematical problems and generalize their knowledge.

References in corpus (2)

• Inferring Algorithmic Patterns with Stack-Augmented Recurrent Nets
• Annotating Derivations: A New Evaluation Strategy and Dataset for Algebra Word Problems

Cited by in corpus (16)

• The Pile: An 800GB Dataset of Diverse Text for Language Modeling
• Scaling Laws for Autoregressive Generative Modeling
• Generative Language Modeling for Automated Theorem Proving
• Neural Symbolic Regression that Scales
• Emergent Symbols through Binding in External Memory
• Towards Question Format Independent Numerical Reasoning: A Set of Prerequisite Tasks
• AR-LSAT: Investigating Analytical Reasoning of Text
• KANDINSKYPatterns -- An experimental exploration environment for Pattern Analysis and Machine Intelligence
• Human-like machine thinking: Language guided imagination
• Learning to solve arithmetic problems with a virtual abacus
• Towards Coinductive Models for Natural Language Understanding. Bringing together Deep Learning and Deep Semantics
• Math Operation Embeddings for Open-ended Solution Analysis and Feedback
• Adversarial Examples for Evaluating Math Word Problem Solvers
• Machine Number Sense: A Dataset of Visual Arithmetic Problems for Abstract and Relational Reasoning
• SCIMAT: Science and Mathematics Dataset
• Disambiguating Symbolic Expressions in Informal Documents