Complexity Control Facilitates Reasoning-Based Compositional Generalization in Transformers
Transformers have demonstrated impressive capabilities across various tasks, yet their performance on compositional problems remains a subject of debate. In this study, we investigate the internal mechanisms underlying Transformers' behavior in compositional tasks. We find that complexity control strategies significantly influence whether the model learns primitive-level rules that generalize out-of-distribution (reasoning-based solutions) or relies solely on memorized mappings (memory-based solutions). By applying masking strategies to the model's information circuits and employing multiple complexity metrics, we reveal distinct internal working mechanisms associated with different solution types. Further analysis reveals that reasoning-based solutions exhibit a lower complexity bias, which aligns with the well-studied neuron condensation phenomenon. This lower complexity bias is hypothesized to be the key factor enabling these solutions to learn reasoning rules. We validate these conclusions across multiple real-world datasets, including image generation and natural language processing tasks, confirming the broad applicability of our findings.
Code (1)
Tasks
Image GenerationSimilar Papers 제목 키워드 기반
Disentangling Reasoning Capabilities from Language Models with Compositional Reasoning Transformers
This paper presents ReasonFormer, a unified reasoning framework for mirroring the modular and compositional reasoning process of humans in complex decision-making. Inspired by dual-process theory in cognitive science, th…
Decision MakingCritical Windows of Complexity Control: When Transformers Decide to Reason or Memorize
Recent work has shown that Transformers' compositional generalization is governed by \emph{complexity control}, initialization scale and weight decay, which steers training toward low-complexity reasoning solutions rathe…
COVR: A test-bed for Visually Grounded Compositional Generalization with real images
While interest in models that generalize at test time to new compositions has risen in recent years, benchmarks in the visually-grounded domain have thus far been restricted to synthetic images. In this work, we propose …
From Data to Control: A Formal Compositional Framework for Large-Scale Interconnected Networks
We introduce a compositional data-driven methodology with noisy data for designing fully-decentralized safety controllers applicable to large-scale interconnected networks, encompassing a vast number of subsystems with u…
Compositional Attention Networks for Interpretability in Natural Language Question Answering
MAC Net is a compositional attention network designed for Visual Question Answering. We propose a modified MAC net architecture for Natural Language Question Answering. Question Answering typically requires Language Unde…
Logical ReasoningQuestion AnsweringVisual Question AnsweringVisual Question Answering (VQA)