Insertion Based Sequence Generation with Learnable Order Dynamics

2026-02-20

AI Summary

This work addresses the challenge of training insertion-based sequence generation models, which suffer from an intractably large action space. The authors propose a novel approach that integrates a learnable sequential dynamics mechanism into the target rate of discrete flow matching, enabling joint optimization of the generator and the insertion order policy without requiring numerical simulation, thereby facilitating end-to-end training. The framework employs a variable-length masked diffusion model as the generator, achieving for the first time flexible and stable order modeling in insertion-based generation. Experiments on de novo small molecule generation demonstrate significant improvements over uniform ordering strategies in both the quantity of valid molecules and overall generation quality. Further graph traversal studies elucidate the performance trade-offs among different parameterized order policies.

Technology Category

• Search and Optimization: Learning to Search
• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

In many domains generating variable length sequences through insertions provides greater flexibility over autoregressive models. However, the action space of insertion models is much larger than that of autoregressive models (ARMs) making the learning challenging. To address this, we incorporate trainable order dynamics into the target rates for discrete flow matching, and show that with suitable choices of parameterizations, joint training of the target order dynamics and the generator is tractable without the need for numerical simulation. As the generative insertion model, we use a variable length masked diffusion model, which generates by inserting and filling mask tokens. On graph traversal tasks for which a locally optimal insertion order is known, we explore the choices of parameterization empirically and demonstrate the trade-offs between flexibility, training stability and generation quality. On de novo small molecule generation, we find that the learned order dynamics leads to an increase in the number of valid molecules generated and improved quality, when compared to uniform order dynamics.

Problem

• insertion-based sequence generation
• action space
• learnable order dynamics
• discrete flow matching
• variable length sequences

Innovation

• insertion-based generation
• learnable order dynamics
• discrete flow matching
• masked diffusion model
• variable-length sequence generation