Related papers: Generalist: Decoupling Natural and Robust Generalization

Generalist: Decoupling Natural and Robust Generalization

URL: http://arxiv.org/abs/2303.13813v1
Date: Fri, 24 Mar 2023 05:24:23 GMT
Title: Generalist: Decoupling Natural and Robust Generalization
Authors: Hongjun Wang, Yisen Wang
Abstract summary: We propose a bi-expert framework called emphGeneralist where we simultaneously train base learners with task-aware strategies. Generalist achieves high accuracy on natural examples while maintaining considerable robustness to adversarial ones.
Score: 14.244311026737666
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Abstract: Deep neural networks obtained by standard training have been constantly plagued by adversarial examples. Although adversarial training demonstrates its capability to defend against adversarial examples, unfortunately, it leads to an inevitable drop in the natural generalization. To address the issue, we decouple the natural generalization and the robust generalization from joint training and formulate different training strategies for each one. Specifically, instead of minimizing a global loss on the expectation over these two generalization errors, we propose a bi-expert framework called \emph{Generalist} where we simultaneously train base learners with task-aware strategies so that they can specialize in their own fields. The parameters of base learners are collected and combined to form a global learner at intervals during the training process. The global learner is then distributed to the base learners as initialized parameters for continued training. Theoretically, we prove that the risks of Generalist will get lower once the base learners are well trained. Extensive experiments verify the applicability of Generalist to achieve high accuracy on natural examples while maintaining considerable robustness to adversarial ones. Code is available at https://github.com/PKU-ML/Generalist.

Related papers

Generalist++: A Meta-learning Framework for Mitigating Trade-off in Adversarial Training [105.74524789405514]
adversarial training (AT) is currently the most effective defense against neural networks.<n>We propose to partition the overall generalization goal into multiple sub-tasks, each assigned to a dedicated base learner.<n>In the later stages of training, we interpolate their parameters to form a knowledgeable global learner.<n>We term this framework Generalist and introduce three variants tailored to different application scenarios.
arXiv Detail & Related papers (2025-10-15T09:47:54Z)
Where to find Grokking in LLM Pretraining? Monitor Memorization-to-Generalization without Test [19.213961869113188]
We conduct the first study of grokking on checkpoints during one-pass pretraining of a 7B large language model (LLM), i.e., OLMoE.<n>Our study, for the first time, verifies that grokking still happens in the pretraining of large-scale foundation models.<n>We develop two novel metrics to quantify pathway distance and the complexity of a single pathway.
arXiv Detail & Related papers (2025-06-26T17:59:58Z)
Zero-Shot Generalization during Instruction Tuning: Insights from Similarity and Granularity [84.12126298229866]
We show that zero-shot generalization during instruction tuning happens very early. We also show that encountering highly similar and fine-grained training data earlier during instruction tuning, without the constraints of defined "tasks", enables better generalization. For the first time, we show that zero-shot generalization during instruction tuning is a form of similarity-based generalization between training and test data at the instance level.
arXiv Detail & Related papers (2024-06-17T16:40:21Z)
Improving Policy Optimization with Generalist-Specialist Learning [23.480173193633252]
Generalization in deep reinforcement learning over unseen environment variations usually requires policy learning over a large set of diverse training variations. We propose a novel generalist-specialist training framework. Specifically, we first train a generalist on all environment variations; when it fails to improve, we launch a large population of specialists with weights cloned from the generalist. We show that this framework pushes the envelope of policy learning on several challenging and popular benchmarks including Procgen, Meta-World and ManiSkill.
arXiv Detail & Related papers (2022-06-26T22:06:40Z)
Formulating Robustness Against Unforeseen Attacks [34.302333899025044]
This paper focuses on the scenario where there is a mismatch in the threat model assumed by the defense during training. We ask the question: if the learner trains against a specific "source" threat model, when can we expect robustness to generalize to a stronger unknown "target" threat model during test-time? We propose adversarial training with variation regularization (AT-VR) which reduces variation of the feature extractor across the source threat model during training.
arXiv Detail & Related papers (2022-04-28T21:03:36Z)
Understanding Robust Generalization in Learning Regular Languages [85.95124524975202]
We study robust generalization in the context of using recurrent neural networks to learn regular languages. We propose a compositional strategy to address this. We theoretically prove that the compositional strategy generalizes significantly better than the end-to-end strategy.
arXiv Detail & Related papers (2022-02-20T02:50:09Z)
When Is Generalizable Reinforcement Learning Tractable? [74.87383727210705]
We study the query complexity required to train RL agents that can generalize to multiple environments. We introduce Strong Proximity, a structural condition which precisely characterizes the relative closeness of different environments. We show that under a natural weakening of this condition, RL can require query complexity that is exponential in the horizon to generalize.
arXiv Detail & Related papers (2021-01-01T19:08:24Z)
Robustness, Privacy, and Generalization of Adversarial Training [84.38148845727446]
This paper establishes and quantifies the privacy-robustness trade-off and generalization-robustness trade-off in adversarial training. We show that adversarial training is $(varepsilon, delta)$-differentially private, where the magnitude of the differential privacy has a positive correlation with the robustified intensity. Our generalization bounds do not explicitly rely on the parameter size which would be large in deep learning.
arXiv Detail & Related papers (2020-12-25T13:35:02Z)
Adversarial Training for Large Neural Language Models [107.84290922621163]
We show that adversarial pre-training can improve both generalization and robustness. ALUM regularizes the training objective by applying perturbations in the embedding space that maximizes the adversarial loss. ALUM can be further combined with task-specific fine-tuning to attain additional gains.
arXiv Detail & Related papers (2020-04-20T00:07:18Z)
Attacks Which Do Not Kill Training Make Adversarial Learning Stronger [85.96849265039619]
Adversarial training based on the minimax formulation is necessary for obtaining adversarial robustness of trained models. We argue that adversarial training is to employ confident adversarial data for updating the current model.
arXiv Detail & Related papers (2020-02-26T01:04:38Z)

This list is automatically generated from the titles and abstracts of the papers in this site.