Relational inductive biases, deep learning, and graph networks
By Peter W. Battaglia, Jessica B. Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinícius Flores Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Caglar Gulcehre and 16 more. 2018
3,656 citations recorded by the source corpus.
Artificial intelligence (AI) has undergone a renaissance recently, making major progress in key domains such as vision, language, control, and decision-making. This has been due, in part, to cheap data and cheap compute resources, which have fit the natural strengths of deep learning. However, many defining characteristics of human intelligence, which developed under much different pressures, remain out of reach for current approaches. In particular, generalizing beyond one's experiences--a hallmark of human intelligence from infancy--remains a formidable challenge for modern AI. The following is part position paper, part review, and part unification. We argue that combinatorial generalization must be a top priority for AI to achieve human-like abilities, and that structured representations and computations are key to realizing this objective. Just as biology uses nature and nurture cooperatively, we reject the false choice between "hand-engineering" and "end-to-end" learning, and instead advocate for an approach which benefits from their complementary strengths. We explore how using relational inductive biases within deep learning architectures can facilitate learning about entities, relations, and rules for composing them. We present a new building block for the AI toolkit with a strong relational inductive bias--the graph network--which generalizes and extends various approaches for neural networks that operate on graphs, and provides a straightforward interface for manipulating structured knowledge and producing structured behaviors. We discuss how graph networks can support relational reasoning and combinatorial generalization, laying the foundation for more sophisticated, interpretable, and flexible patterns of reasoning. As a companion to this paper, we have released an open-source software library for building graph networks, with demonstrations of how to use them in practice.
Subfields: Artificial Intelligence, Graph Learning
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yoshua Bengio
- Mohit Bansal
- Sergey Levine
- Percy Liang
- Jianfeng Gao
- Quoc V. Le
- Ruslan Salakhutdinov
- Luke S. Zettlemoyer
In Graph Learning
- Jure Leskovec
- William L. Hamilton
- Stephan Günnemann
- Matthias Fey
- Marinka Zitnik
- Michael M. Bronstein
- Rex Ying
- Jian Tang
Citations in the corpus
- Object-Centric Learning with Slot Attention (2020, background): “[57] Peter W Battaglia, Jessica B Hamrick, Victor Bapst, Alvaro Sanchez-Gonzalez, Vinicius Zambaldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner, et al.”
- Local Relation Networks for Image Recognition (2019, methodology): “Appearance composability We follow a general approach for relation modeling [1] to compute appearance composability Φ(fθq (xp′), fθk(xp)), where xp′ and xp are projected to a query (by a channel transformation layer fθq ) and key (by a channel transformation layer fθk ) embedding space, respectively”
- Environmental drivers of systematicity and generalization in a situated agent (2019, background): “…generalising systematically requires inductive biases that are specifically designed to support some form of symbolic computation (such as graphs (Battaglia et al., 2018), modular-components defined by symbolic parses (Andreas et al., 2016; Bahdanau et al., 2018), explicit latent variables…”
- Scalable agent alignment via reward modeling: a research direction (2018, background): “Examples include modularity (Anonymous, 2019b), recursion (Cai et al., 2017), graph structure (Battaglia et al., 2018) or natural language (Andreas et al., 2018) in the latent space, differentiable external memory (Graves et al., 2016), or neural units designed to perform arbitrary arithmetic…”
- Toward Causal Representation Learning (2021, background): “Machine learning models that incorporate or learn structural knowledge of an environment have been shown to be more efficient and generalize better [9], [11], [15], [16], [27], [58], [77], [84], [85], [141], [157], [177], [181], [197], [211], [212], [244], [258], [272], [274].”
- Fast Graph Representation Learning with PyTorch Geometric (2019, background): “Similar to the concepts of convolutional and pooling layers on regular domains, GNNs are able to (hierarchically) extract localized embeddings by passing, transforming, and aggregating information (Bronstein et al., 2017; Gilmer et al., 2017; Battaglia et al., 2018; Ying et al., 2018).”
- Attacking Graph-based Classification via Manipulating the Graph Structure (2019, methodology): “normal user. Roughly speaking, there are two types of graph-based classification methods, i.e., collective classification[1, 9, 10, 14, 23, 25, 36, 40, 47, 59, 64–67, 72, 74] and graph neural network [5, 38, 52, 69]. Collective classification defines apriorreputationscore to each node based on the t”
- Reconciling deep learning with symbolic artificial intelligence: representing objects and relations (2019, background): “Notwithstanding its undeniable success, critics have recently drawn attention to a number of shortcomings in contemporary deep learning [9 ,10 ,11,12].”
- Integrating Learning and Reasoning with Deep Logic Models (2019, background): “In spite of the incredible success of deep learning, many researchers have recently started to question the ability of deep learning to bring us real AI, because the amount and quality of training data would explode in order to jointly learn the high-level reasoning that is needed to perform complex”
- Learning to Simulate Complex Physics with Graph Networks (2020, background): “D ECODER details .”
- Recurrent Independent Mechanisms (2019, background): “…scenes, and motion-capture data, and multiagent systems (Scarselli et al., 2008; Bronstein et al., 2017; Watters et al., 2017; Raposo et al., 2017; Santoro et al., 2017; Gilmer et al., 2017; Van Steenkiste et al., 2018; Kipf et al., 2018; Battaglia et al., 2018; Tacchetti et al., 2018).”
- Are Disentangled Representations Helpful for Abstract Visual Reasoning? (2019, methodology): “The Relation Network implements a suitable inductive bias for (relational) reasoning [5].”
- COBRA: Data-Efficient Model-Based RL through Unsupervised Object Discovery and Curiosity-Driven Exploration (2019, methodology): “To take advantage of the slot-structured representation of objects in z t , we use a Relation Network (Santoro et al., 2017), a sub-type of graph network (Battaglia et al., 2018).”
- Structured agents for physical construction (2019, methodology): “, 2014), implemented with graph neural networks (Scarselli et al., 2009; Bronstein et al., 2017; Gilmer et al., 2017; Battaglia et al., 2018).”
- Stochastic Prediction of Multi-Agent Interactions from Partial Observations (2019, positive): “It is also possible to use graph attention networks (Veli ˇ ckovi ´ c et al., 2018) for this task (see e.g., Hoshen (2017); Battaglia et al. (2018)).”
- Causal Reasoning from Meta-reinforcement Learning (2019, background): “…2018) that allow us to train on longer more complex episodes, as well as models which are more explicitly compositional (e.g. Andreas et al., 2016; Battaglia et al., 2018) or have richer semantics (e.g. Ganin et al., 2018) that can more explicitly leverage symmetries in the environment and…”