π€ AI Summary
This work addresses the limitation of standard attention mechanisms, which rely on the independence assumption inherent in softmax and thus struggle to model structured cooperative relationships among entitiesβsuch as those in retrosynthetic reaction center prediction. To overcome this, the authors propose Variational Ising Attention (VIA), the first approach to integrate the Ising model into attention mechanisms. VIA explicitly captures pairwise dependencies through learnable coupling parameters and employs variational mean-field inference to derive attention distributions from collective states, replacing the conventional softmax-based ranking of independent terms. Experimental results demonstrate that VIA significantly outperforms baseline methods on retrosynthesis tasks, while ablation studies and multiple model variants confirm its effectiveness and generalization potential.
π Abstract
Attention enables context modeling via query-key scoring with softmax normalization. Driven by industrial long-context demands, mainstream research has converged toward sparsity and efficiency--yet softmax's independence assumption persists. For scientific tasks unburdened by long-token constraints, however, richer structured coupling may often be essential, making tailored attention both viable and more appropriate. To this end, we propose Variational-Ising-Attention (VIA), which augments softmax normalization with an interacting Ising model; attention patterns emerge from learnable pairwise couplings via variational mean-field inference, redefining attention from a ranking over isolated items to a collective state over interacting entities. We instantiate VIA on retrosynthesis reaction center prediction, a task inherently governed by cooperative bond-breaking constraints. Comprehensive experiments across model variants, coupled with mechanistic analyses, demonstrate that VIA consistently and substantially outperforms standard softmax attention. More broadly, our findings suggest that for scientific problems, the optimal solution is not general-purpose efficiency, but appropriately tailored attention aligned with intrinsic domain structure. This work provides a theoretically grounded and empirically validated instantiation of this paradigm.