Neural Programmer-Interpreters

Abstract

We propose the neural programmer-interpreter (NPI): a recurrent and compositional neural network that learns to represent and execute programs. NPI has three learnable components: a task-agnostic recurrent core, a persistent key-value program memory, and domain-specific encoders that enable a single NPI to operate in multiple perceptually diverse environments with distinct affordances. By learning to compose lower-level programs to express higher-level programs, NPI reduces sample complexity and increases generalization ability compared to sequence-to-sequence LSTMs. The program memory allows efficient learning of additional tasks by building on existing programs. NPI can also harness the environment (e.g. a scratch pad with read-write pointers) to cache intermediate results of computation, lessening the long-term memory burden on recurrent hidden units. In this work we train the NPI with fully-supervised execution traces; each program has example sequences of calls to the immediate subprograms conditioned on the input. Rather than training on a huge number of relatively weak labels, NPI learns from a small number of rich examples. We demonstrate the capability of our model to learn several types of compositional programs: addition, sorting, and canonicalizing 3D models. Furthermore, a single NPI learns to execute these programs and all 21 associated subprograms.

ICLR 2016 conference submission

References in corpus (7)

• Sequence to Sequence Learning with Neural Networks
• Learning to Execute
• Neural GPUs Learn Algorithms
• Reinforcement Learning Neural Turing Machines - Revised
• Neural Programmer: Inducing Latent Programs with Gradient Descent
• Learning Simple Algorithms from Examples
• Building Program Vector Representations for Deep Learning

Cited by in corpus (89)

• Deep Reinforcement Learning: An Overview
• CodeXGLUE: A Machine Learning Benchmark Dataset for Code Understanding and Generation
• Adaptive Computation Time for Recurrent Neural Networks
• Modular Multitask Reinforcement Learning with Policy Sketches
• Learning to Optimize
• Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions
• Born to Learn: the Inspiration, Progress, and Future of Evolved Plastic Artificial Neural Networks
• Programming with a Differentiable Forth Interpreter
• Learning to Infer and Execute 3D Shape Programs
• A Semantic Loss Function for Deep Learning with Symbolic Knowledge
• Neural Programmer: Inducing Latent Programs with Gradient Descent
• Dynamic Neural Program Embedding for Program Repair
• Learning Continuous Semantic Representations of Symbolic Expressions
• Compositional generalization through meta sequence-to-sequence learning
• Learn&Fuzz: Machine Learning for Input Fuzzing
• Neural Symbolic Machines: Learning Semantic Parsers on Freebase with Weak Supervision
• Generative Temporal Models with Memory
• Automated Curriculum Learning for Neural Networks
• A Neural Transducer
• Online Continual Learning on Sequences
• Meta-learning of Sequential Strategies
• AMPNet: Asynchronous Model-Parallel Training for Dynamic Neural Networks
• Learning Efficient Algorithms with Hierarchical Attentive Memory
• Improving the Neural GPU Architecture for Algorithm Learning
• A Comprehensive Review of State-of-The-Art Methods for Java Code Generation from Natural Language Text
• Learning Compositional Neural Programs with Recursive Tree Search and Planning
• Learning compositionally through attentive guidance
• Strong Generalization and Efficiency in Neural Programs
• Neural Machine Translation for Query Construction and Composition
• An Introduction to Deep Learning for the Physical Layer
• Coordination Among Neural Modules Through a Shared Global Workspace
• Estimate and Replace: A Novel Approach to Integrating Deep Neural Networks with Existing Applications
• Neural Program Search: Solving Programming Tasks from Description and Examples
• Learning Scalable and Precise Representation of Program Semantics
• Show Your Work: Scratchpads for Intermediate Computation with Language Models
• Neural Shuffle-Exchange Networks -- Sequence Processing in O(n log n) Time
• Fact-based Text Editing
• Object Files and Schemata: Factorizing Declarative and Procedural Knowledge in Dynamical Systems
• Transforming task representations to perform novel tasks
• Learning abstract structure for drawing by efficient motor program induction
• Discrete-Valued Neural Communication
• Inferring Logical Forms From Denotations
• Neural network gradient-based learning of black-box function interfaces
• Why Build an Assistant in Minecraft?
• Learning to Synthesize Programs as Interpretable and Generalizable Policies
• Can Neural Networks Understand Logical Entailment?
• Learning to Execute Programs with Instruction Pointer Attention Graph Neural Networks
• Robot Program Parameter Inference via Differentiable Shadow Program Inversion
• Transformers with Competitive Ensembles of Independent Mechanisms
• Disentangled Representations in Neural Models
• Compositional Generalization via Neural-Symbolic Stack Machines
• Neutaint: Efficient Dynamic Taint Analysis with Neural Networks
• AutoCorrect: Deep Inductive Alignment of Noisy Geometric Annotations
• Controllable Text Simplification with Explicit Paraphrasing
• A Multi-Type Multi-Span Network for Reading Comprehension that Requires Discrete Reasoning
• Synthesizing Action Sequences for Modifying Model Decisions
• REAS: Combining Numerical Optimization with SAT Solving
• Learning Differentiable Programs with Admissible Neural Heuristics
• Compositional Generalization with Tree Stack Memory Units
• Learning Task-General Representations with Generative Neuro-Symbolic Modeling
• Genetic algorithms with DNN-based trainable crossover as an example of partial specialization of general search
• NEUZZ: Efficient Fuzzing with Neural Program Smoothing
• Neural CRF Model for Sentence Alignment in Text Simplification
• Predicting Missing Information of Key Aspects in Vulnerability Reports
• REALab: An Embedded Perspective on Tampering
• Learning Fitness Functions for Machine Programming
• Modeling Past and Future for Neural Machine Translation
• Learning an Executable Neural Semantic Parser
• Learning Compositional Neural Programs for Continuous Control
• Avoiding Tampering Incentives in Deep RL via Decoupled Approval
• Operational Calculus for Differentiable Programming
• Applications of Probabilistic Programming (Master's thesis, 2015)
• Neural Program Synthesis By Self-Learning
• Explaining Transition Systems through Program Induction
• Using Program Induction to Interpret Transition System Dynamics
• Amanuensis: The Programmer's Apprentice
• Hierarchical Variational Imitation Learning of Control Programs
• pix2rule: End-to-end Neuro-symbolic Rule Learning
• Understanding in Artificial Intelligence
• Meta-Learning an Inference Algorithm for Probabilistic Programs
• An Empirical Comparison of Syllabuses for Curriculum Learning
• Learning Solving Procedure for Artificial Neural Network
• Progress Extrapolating Algorithmic Learning to Arbitrary Sequence Lengths
• Neural Programming by Example
• Neural Architecture for Question Answering Using a Knowledge Graph and Web Corpus
• Type-driven Neural Programming by Example
• Reinforcement Learning of Implicit and Explicit Control Flow in Instructions
• Sequential Coordination of Deep Models for Learning Visual Arithmetic
• Exploiting Deep Learning for Secure Transmission in an Underlay Cognitive Radio Network