Automatically Composing Representation Transformations as a Means for Generalization

By Michael Chang, Abhishek Gupta, Sergey Levine, T. Griffiths. 2018

80 citations recorded by the source corpus.

A generally intelligent learner should generalize to more complex tasks than it has previously encountered, but the two common paradigms in machine learning -- either training a separate learner per task or training a single learner for all tasks -- both have difficulty with such generalization because they do not leverage the compositional structure of the task distribution. This paper introduces the compositional problem graph as a broadly applicable formalism to relate tasks of different complexity in terms of problems with shared subproblems. We propose the compositional generalization problem for measuring how readily old knowledge can be reused and hence built upon. As a first step for tackling compositional generalization, we introduce the compositional recursive learner, a domain-general framework for learning algorithmic procedures for composing representation transformations, producing a learner that reasons about what computation to execute by making analogies to previously seen problems. We show on a symbolic and a high-dimensional domain that our compositional approach can generalize to more complex problems than the learner has previously encountered, whereas baselines that are not explicitly compositional do not.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Representation & Self-Supervised 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

In Computer Vision and Pattern Recognition

In Representation & Self-Supervised Learning

Citations in the corpus

Continual Learning with Recursive Gradient Optimization (2022, background): “, 2021), add additional components as new tasks arrive (Rusu et al., 2016; Fernando et al., 2017; Alet et al., 2018; Chang et al., 2019; Li et al., 2019), or use larger networks to generate network parameters (Aljundi et al.”
Continual Learning via Local Module Composition (2021, background): “Modularity in neural networks is studied in the context of scalability [9], and more recently as a way to achieve compositionality and systematic generalization [6, 47, 15, 8, 27, 19] as well as for multi-task learning [65, 82].”
Wide Neural Networks Forget Less Catastrophically (2021, background): “Expansion-based methods (Rusu et al., 2016; Fernando et al., 2017; Aljundi et al., 2017; Rosenbaum et al., 2018; Chang et al., 2018; Xu and Zhu, 2018; Ferran Alet, 2018; Li et al., 2019; Veniat et al., 2021) add new network modules or experts for new tasks.”
Towards Continual Reinforcement Learning: A Review and Perspectives (2020, background): “More recent work (Devin et al., 2017; Frans et al., 2017; Fernando et al., 2017; Rosenbaum et al., 2017; Meyerson and Miikkulainen, 2017; Ramachandran and Le, 2018; Chang et al., 2018; Liang et al., 2018; Alet et al., 2018; Cases et al., 2019; Yang et al., 2020) has focused on training neural networ”
Continual Learning in Low-rank Orthogonal Subspaces (2020, neutral): “Second, modular approaches use different network modules that can be extended for each new task [Fernando et al., 2017, Aljundi et al., 2017, Rosenbaum et al., 2018, Chang et al., 2018, Xu and Zhu, 2018, Ferran Alet, 2018].”
Using Hindsight to Anchor Past Knowledge in Continual Learning (2019, methodology): “Second, modular approaches use different parts of the prediction function for each new task (Fernando et al., 2017; Aljundi et al., 2017; Rosenbaum et al., 2018; Chang et al., 2018; Xu & Zhu, 2018; Ferran Alet, 2018).”
Efficient Lifelong Learning with A-GEM (2018, methodology): “One approach to LLL uses modular compositional models (Fer-nando et al., 2017; Aljundi et al., 2017; Rosenbaum et al., 2018; Chang et al., 2018; Xu & Zhu, 2018; Ferran Alet, 2018), which limit interference among tasks by using different subset of modules for each task.”
Routing Networks and the Challenges of Modular and Compositional Computation (2019, methodology): “If the model finds a good ‘locality’ of approximations, routing can allow a model to achieve truly impressive levels of generalization (compare [Chang et al., 2019], and in particular the section on nested implicatives in [Cases et al.”
Measuring Compositional Generalization: A Comprehensive Method on Realistic Data (2019, background): “Chang et al. (2019) introduce the compositional recursive learner that “can generalize to more complex problems than the learner has previously encountered”.”
Lifelong Learning of Compositional Structures (2020, background): “Other approaches do not make such an assumption, and instead learn the structure directly from optimization of a cost function [26, 13, 19, 2, 6].”
How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Composition (2022, methodology): “Methods for tackling this variant of the problem have included using RL (Chang et al., 2019) or expectation maximization (Kirsch et al., 2018) as the optimization tool.”
Recursive Routing Networks: Learning to Compose Modules for Language Understanding (2019, background): “, 2017), which learn to construct a neural network from pre-defined modules; the Compositional Recursive Learner (Chang et al., 2018), a closely related approach that uti-”
Multi-Task Reinforcement Learning with Context-based Representations (2021, background): “Chang et al. (2018) consider a family of algorithmic tasks, with different levels of complexity.”
Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks (2020, background): “At ﬁrst glance, explicitly modular networks (Clune et al., 2013; Andreas et al., 2016; Kirsch et al., 2018; Chang et al., 2019; Bahdanau et al., 2019) could provide a solution for the discovered problems.”
Multi-Task Learning with Deep Neural Networks: A Survey (2020, background): “In Routing Networks and the CRL, any of the network modules can be placed into an instantiated network at any depth.”
Efficient Lifelong Learning with A-GEM (2018, methodology): “One approach to LLL uses modular compositional models (Fer-nando et al., 2017; Aljundi et al., 2017; Rosenbaum et al., 2018; Chang et al., 2018; Xu & Zhu, 2018; Ferran Alet, 2018), which limit interference among tasks by using different subset of modules for each task.”
Compositional Generalization via Neural-Symbolic Stack Machines (2020, background): “Some recent work also studies length generalization for other tasks, including relational reasoning [14], multi-task learning [7], and structure inference [31].”
Compositional Generalization for Primitive Substitutions (2019, unknown): “…many related tasks (Lake and Baroni, 2018; Loula et al., 2018a; Liˇska et al., 2018; Bastings et al., 2018; Lake et al., 2019) and methods (Bastings et al., 2018; Loula et al., 2018a; Kliegl and Xu, 2018; Chang et al., 2018) using a variety of RNN models and attention mechanism have been proposed.”
Sharing Less is More: Lifelong Learning in Deep Networks with Selective Layer Transfer (2020, methodology): “…compositional structure and learning the optimal modules (Rosenbaum et al., 2019), so algorithms such as meta-learning (Alet et al., 2018), expectation-maximization (Kirsch et al., 2018) and reinforcement learning (Rosenbaum et al., 2018; Chang et al., 2019) have been used to tackle this challenge.”