Scaling can lead to compositional generalization
This addresses a fundamental challenge in AI for improving model reliability in tasks requiring systematic reasoning, though it is incremental by building on existing scaling trends.
The paper tackles the problem of whether neural networks can achieve compositional generalization, finding that scaling data and model size leads to such generalization across tasks with shared structure, and proves that multilayer perceptrons can approximate compositional task families with linear neuron scaling.
Can neural networks systematically capture discrete, compositional task structure despite their continuous, distributed nature? The impressive capabilities of large-scale neural networks suggest that the answer to this question is yes. However, even for the most capable models, there are still frequent failure cases that raise doubts about their compositionality. Here, we seek to understand what it takes for a standard neural network to generalize over tasks that share compositional structure. We find that simply scaling data and model size leads to compositional generalization. We show that this holds across different task encodings as long as the training distribution sufficiently covers the task space. In line with this finding, we prove that standard multilayer perceptrons can approximate a general class of compositional task families to arbitrary precision using only a linear number of neurons with respect to the number of task modules. Finally, we uncover that if networks successfully compositionally generalize, the constituents of a task can be linearly decoded from their hidden activations. We show that this metric correlates with failures of text-to-image generation models to compose known concepts.