Compositional Attention Networks for Machine Reasoning
arXiv:1803.03067
Abstract
We present the MAC network, a novel fully differentiable neural network architecture, designed to facilitate explicit and expressive reasoning. MAC moves away from monolithic black-box neural architectures towards a design that encourages both transparency and versatility. The model approaches problems by decomposing them into a series of attention-based reasoning steps, each performed by a novel recurrent Memory, Attention, and Composition (MAC) cell that maintains a separation between control and memory. By stringing the cells together and imposing structural constraints that regulate their interaction, MAC effectively learns to perform iterative reasoning processes that are directly inferred from the data in an end-to-end approach. We demonstrate the model's strength, robustness and interpretability on the challenging CLEVR dataset for visual reasoning, achieving a new state-of-the-art 98.9% accuracy, halving the error rate of the previous best model. More importantly, we show that the model is computationally-efficient and data-efficient, in particular requiring 5x less data than existing models to achieve strong results.
arXiv commentPublished as a conference paper at ICLR 2018
References in corpus (9)
- Adam: A Method for Stochastic Optimization
- Hierarchical Question-Image Co-Attention for Visual Question Answering
- Dynamic Memory Networks for Visual and Textual Question Answering
- A simple neural network module for relational reasoning
- Towards Deep Symbolic Reinforcement Learning
- Learning to Compose Neural Networks for Question Answering
- Learning to Reason: End-to-End Module Networks for Visual Question Answering
- CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
- Survey of Visual Question Answering: Datasets and Techniques
Cited by in corpus (55)
- ViLT: Vision-and-Language Transformer Without Convolution or Region Supervision
- Large-Scale Adversarial Training for Vision-and-Language Representation Learning
- Transparency by Design: Closing the Gap Between Performance and Interpretability in Visual Reasoning
- Improving Multi-hop Knowledge Base Question Answering by Learning Intermediate Supervision Signals
- Visual Entailment: A Novel Task for Fine-Grained Image Understanding
- Object Level Visual Reasoning in Videos
- Compositional generalization in a deep seq2seq model by separating syntax and semantics
- Talk2Nav: Long-Range Vision-and-Language Navigation with Dual Attention and Spatial Memory
- Trends in Integration of Vision and Language Research: A Survey of Tasks, Datasets, and Methods
- Unshuffling Data for Improved Generalization
- CLOSURE: Assessing Systematic Generalization of CLEVR Models
- Multimodal Unified Attention Networks for Vision-and-Language Interactions
- RAVEN: A Dataset for Relational and Analogical Visual rEasoNing
- Discrete and continuous representations and processing in deep learning: Looking forward
- Dynamic Graph Attention for Referring Expression Comprehension
- Understanding the Role of Scene Graphs in Visual Question Answering
- MoVie: Revisiting Modulated Convolutions for Visual Counting and Beyond
- Variational Cross-Graph Reasoning and Adaptive Structured Semantics Learning for Compositional Temporal Grounding
- Visual Entailment Task for Visually-Grounded Language Learning
- VideoNavQA: Bridging the Gap between Visual and Embodied Question Answering
- DeepLogic: Towards End-to-End Differentiable Logical Reasoning
- Learning Visual Question Answering by Bootstrapping Hard Attention
- Think before you act: A simple baseline for compositional generalization
- VIOLIN: A Large-Scale Dataset for Video-and-Language Inference
- Induce, Edit, Retrieve: Language Grounded Multimodal Schema for Instructional Video Retrieval
- Measuring non-trivial compositionality in emergent communication
- SceneGATE: Scene-Graph based co-Attention networks for TExt visual question answering
- Attention on Abstract Visual Reasoning
- On transfer learning using a MAC model variant
- Attention over learned object embeddings enables complex visual reasoning
- Zero-Shot Transfer VQA Dataset
- A Study of Compositional Generalization in Neural Models
- How to Make a BLT Sandwich? Learning to Reason towards Understanding Web Instructional Videos
- Analogical Reasoning for Visually Grounded Language Acquisition
- KANDINSKYPatterns -- An experimental exploration environment for Pattern Analysis and Machine Intelligence
- Visual Question Answering with Prior Class Semantics
- Giving Commands to a Self-driving Car: A Multimodal Reasoner for Visual Grounding
- Visual Superordinate Abstraction for Robust Concept Learning
- NAAQA: A Neural Architecture for Acoustic Question Answering
- New Ideas and Trends in Deep Multimodal Content Understanding: A Review
- Question Guided Modular Routing Networks for Visual Question Answering
- AiR: Attention with Reasoning Capability
- Linguistically Driven Graph Capsule Network for Visual Question Reasoning
- Few-Shot Abstract Visual Reasoning With Spectral Features
- Improved RAMEN: Towards Domain Generalization for Visual Question Answering
- Robust Question Answering Through Sub-part Alignment
- Holistic Multi-modal Memory Network for Movie Question Answering
- Compositional Attention: Disentangling Search and Retrieval
- Cascaded Mutual Modulation for Visual Reasoning
- Adventurer's Treasure Hunt: A Transparent System for Visually Grounded Compositional Visual Question Answering based on Scene Graphs
- Neuro-Symbolic AI: An Emerging Class of AI Workloads and their Characterization
- Calibrating Concepts and Operations: Towards Symbolic Reasoning on Real Images
- Transformation Driven Visual Reasoning
- Reasoning Over History: Context Aware Visual Dialog
- Interpretable Neural Computation for Real-World Compositional Visual Question Answering