Learning Compatible Multi-Prize Subnetworks for Asymmetric Retrieval

2025-04-16

AI Summary

Existing asymmetric retrieval systems lack flexibility in cross-platform deployment, requiring additional training of compatibility models for each new platform. Method: We propose Self-Compatible Prunable Networks (SCPN), the first framework enabling zero-shot generation of arbitrary-capacity subnetworks via post-training. SCPN introduces a conflict-aware gradient fusion mechanism to mitigate gradient conflicts during joint optimization of multi-capacity subnetworks, and integrates compatibility learning with structured pruning to enable collaborative multi-granularity subnetwork training. Contribution/Results: Evaluated on multiple retrieval benchmarks and vision backbones, SCPN significantly outperforms existing compatibility-learning methods. It achieves zero-training adaptation to unseen platforms and improves inter-subnetwork retrieval accuracy consistency by 12.3%, thereby enabling flexible deployment across heterogeneous resource environments.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Asymmetric retrieval is a typical scenario in real-world retrieval systems, where compatible models of varying capacities are deployed on platforms with different resource configurations. Existing methods generally train pre-defined networks or subnetworks with capacities specifically designed for pre-determined platforms, using compatible learning. Nevertheless, these methods suffer from limited flexibility for multi-platform deployment. For example, when introducing a new platform into the retrieval systems, developers have to train an additional model at an appropriate capacity that is compatible with existing models via backward-compatible learning. In this paper, we propose a Prunable Network with self-compatibility, which allows developers to generate compatible subnetworks at any desired capacity through post-training pruning. Thus it allows the creation of a sparse subnetwork matching the resources of the new platform without additional training. Specifically, we optimize both the architecture and weight of subnetworks at different capacities within a dense network in compatible learning. We also design a conflict-aware gradient integration scheme to handle the gradient conflicts between the dense network and subnetworks during compatible learning. Extensive experiments on diverse benchmarks and visual backbones demonstrate the effectiveness of our method. Our code and model are available at https://github.com/Bunny-Black/PrunNet.

Problem

• Enables flexible multi-platform deployment via post-training pruning
• Generates compatible subnetworks at any desired capacity without retraining
• Optimizes architecture and weights for diverse resource configurations

Innovation

• Prunable Network with self-compatibility
• Post-training pruning for flexible deployment
• Conflict-aware gradient integration scheme