[2509.24653] Identity Bridge: Enabling Implicit Reasoning via Shared Latent Memory
Abstract:Despite remarkable advances, large language models often fail at compositional reasoning tasks, a phenomenon exemplified by the ``curse of two-hop reasoning''. This paper introduces the Identity Bridge, a simple yet powerful mechanism that resolves this compositionality gap by supervising the model on a zero-hop identity task. We demonstrate empirically that this addition enables models to successfully perform out-of-distribution two-hop reasoning, a task they otherwise completely fail. To explain this phenomenon, we provide a theoretical analysis using a simplified Emb-MLP model, proving that identity supervision reshapes the model's latent geometry. We show this alignment is induced by an implicit nuclear-norm regularization during optimization, which favors low-rank solutions that share structure across tasks. For complex tasks, we use small initialization or weight decay to enhance the regularization effect, which enhances the latent space alignment effect and slows down the generalization decay. Finally, we extend our investigation to large-scale models, observing that they still achieve two-hop reasoning through the latent memory, which provides crucial inspiration for enhancing their implicit reasoning abilities.
View PDF HTML (experimental) Abstract:Large Language Models (LLMs) excel at multi-hop reasoning in distribution, yet fail on unseen compositions, a phenomenon known as the curse of two-hop reasoning. In this work, we argue that this phenomenon can be attributed to a missing supervision on the bridge entity. We formalize this gap by introducing identity bridge, a minimal supervision that enforces a identity mapping on bridge tokens. Under this supervision, even a one-layer transformer with uniform attention (Emb-MLP) can achieve out-of-distribution (OOD) two-hop generalization. We provide a…
saved by
related reading
- Reasoning as Trajectoriesslhleosun.github.io
- Explore | alphaXivalphaxiv.org
- A Mathematical Framework for Transformer Circuitstransformer-circuits.pub
- 2305.18654arxiv.org
- Transformer Circuits Threadtransformer-circuits.pub
- [2305.18654] Faith and Fate: Limits of Transformers on Compositionalityarxiv.org
- 2305.18654.pdfarxiv.org
- Transformer Explainer: LLM Transformer Model Visually Explainedpoloclub.github.io
- [2603.07267] How to Steal Reasoning Without Reasoning Tracesarxiv.org
- Language model harnesses are compositional generalizersalexzhang13.github.io
- Training Large Language Models to Reason in a Continuous Latent Spacearxiv.org
- Explore | alphaXivalphaxiv.org