# Piggyback: Adapting a Single Network to Multiple Tasks by Learning to Mask Weights

arXiv: https://arxiv.org/abs/1801.06519 · HTML: https://mlsanity.com/paper/1801.06519/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Jan 2018.

## Built on by (most cited first, 11 of 11)

- [PLATON: Pruning Large Transformer Models with Upper Confidence Bound of Weight Importance](https://mlsanity.com/paper/2206.12562/) (2022)
- [Transferability in Deep Learning: A Survey](https://mlsanity.com/paper/2201.05867/) (2022)
- [Less is More: Selective Layer Finetuning with SubTuning](https://mlsanity.com/paper/2302.06354/) (2023)
- [Task Affinity with Maximum Bipartite Matching in Few-Shot Learning](https://mlsanity.com/paper/2110.02399/) (2021)
- [Fisher Task Distance and Its Application in Neural Architecture Search](https://mlsanity.com/paper/2103.12827/) (2021)
- [Pruning Pre-trained Language Models Without Fine-Tuning](https://mlsanity.com/paper/2210.06210/) (2022)
- [Boosting Binary Masks for Multi-Domain Learning through Affine Transformations](https://mlsanity.com/paper/2103.13894/) (2021)
- [CFA: Constraint-based Finetuning Approach for Generalized Few-Shot Object Detection](https://mlsanity.com/paper/2204.05220/) (2022)
- [Class-Incremental Mixture of Gaussians for Deep Continual Learning](https://mlsanity.com/paper/2307.04094/) (2023)
- [RECAST: Reparameterized, Compact weight Adaptation for Sequential Tasks](https://mlsanity.com/paper/2411.16870/) (2024)
- [Distillation-Guided Structural Transfer for Continual Learning Beyond Sparse Distributed Memory](https://mlsanity.com/paper/2512.15267/) (2025)

## Abstract (arXiv)

This work presents a method for adapting a single, fixed deep neural network to multiple tasks without affecting performance on already learned tasks. By building upon ideas from network quantization and pruning, we learn binary masks that piggyback on an existing network, or are applied to unmodified weights of that network to provide good performance on a new task. These masks are learned in an end-to-end differentiable fashion, and incur a low overhead of 1 bit per network parameter, per task. Even though the underlying network is fixed, the ability to mask individual weights allows for the learning of a large number of filters. We show performance comparable to dedicated fine-tuned networks for a variety of classification tasks, including those with large domain shifts from the initial task (ImageNet), and a variety of network architectures. Unlike prior work, we do not suffer from catastrophic forgetting or competition between tasks, and our performance is agnostic to task ordering. Code available at https://github.com/arunmallya/piggyback.