AD-DROP: Attribution-Driven Dropout for Robust Language Model
Fine-Tuning
- URL: http://arxiv.org/abs/2210.05883v1
- Date: Wed, 12 Oct 2022 02:54:41 GMT
- Title: AD-DROP: Attribution-Driven Dropout for Robust Language Model
Fine-Tuning
- Authors: Tao Yang, Jinghao Deng, Xiaojun Quan, Qifan Wang, Shaoliang Nie
- Abstract summary: We find that dropping attention positions with low attribution scores can accelerate training and increase the risk of overfitting.
We develop a cross-tuning strategy to alternate fine-tuning and AD-DROP to avoid dropping high-attribution positions excessively.
- Score: 24.028662731799127
- License: http://creativecommons.org/licenses/by-nc-sa/4.0/
- Abstract: Fine-tuning large pre-trained language models on downstream tasks is apt to
suffer from overfitting when limited training data is available. While dropout
proves to be an effective antidote by randomly dropping a proportion of units,
existing research has not examined its effect on the self-attention mechanism.
In this paper, we investigate this problem through self-attention attribution
and find that dropping attention positions with low attribution scores can
accelerate training and increase the risk of overfitting. Motivated by this
observation, we propose Attribution-Driven Dropout (AD-DROP), which randomly
discards some high-attribution positions to encourage the model to make
predictions by relying more on low-attribution positions to reduce overfitting.
We also develop a cross-tuning strategy to alternate fine-tuning and AD-DROP to
avoid dropping high-attribution positions excessively. Extensive experiments on
various benchmarks show that AD-DROP yields consistent improvements over
baselines. Analysis further confirms that AD-DROP serves as a strategic
regularizer to prevent overfitting during fine-tuning.
Related papers
- Low-rank finetuning for LLMs: A fairness perspective [54.13240282850982]
Low-rank approximation techniques have become the de facto standard for fine-tuning Large Language Models.
This paper investigates the effectiveness of these methods in capturing the shift of fine-tuning datasets from the initial pre-trained data distribution.
We show that low-rank fine-tuning inadvertently preserves undesirable biases and toxic behaviors.
arXiv Detail & Related papers (2024-05-28T20:43:53Z) - Improving Adversarial Training using Vulnerability-Aware Perturbation
Budget [7.430861908931903]
Adversarial Training (AT) effectively improves the robustness of Deep Neural Networks (DNNs) to adversarial attacks.
We propose two simple, computationally cheap vulnerability-aware reweighting functions for assigning perturbation bounds to adversarial examples used for AT.
Experimental results show that the proposed methods yield genuine improvements in the robustness of AT algorithms against various adversarial attacks.
arXiv Detail & Related papers (2024-03-06T21:50:52Z) - Perturbation-Invariant Adversarial Training for Neural Ranking Models:
Improving the Effectiveness-Robustness Trade-Off [107.35833747750446]
adversarial examples can be crafted by adding imperceptible perturbations to legitimate documents.
This vulnerability raises significant concerns about their reliability and hinders the widespread deployment of NRMs.
In this study, we establish theoretical guarantees regarding the effectiveness-robustness trade-off in NRMs.
arXiv Detail & Related papers (2023-12-16T05:38:39Z) - Feature Separation and Recalibration for Adversarial Robustness [18.975320671203132]
We propose a novel, easy-to- verify approach named Feature Separation and Recalibration.
It recalibrates the malicious, non-robust activations for more robust feature maps through Separation and Recalibration.
It improves the robustness of existing adversarial training methods by up to 8.57% with small computational overhead.
arXiv Detail & Related papers (2023-03-24T07:43:57Z) - Proposal Distribution Calibration for Few-Shot Object Detection [65.19808035019031]
In few-shot object detection (FSOD), the two-step training paradigm is widely adopted to mitigate the severe sample imbalance.
Unfortunately, the extreme data scarcity aggravates the proposal distribution bias, hindering the RoI head from evolving toward novel classes.
We introduce a simple yet effective proposal distribution calibration (PDC) approach to neatly enhance the localization and classification abilities of the RoI head.
arXiv Detail & Related papers (2022-12-15T05:09:11Z) - Domain Adaptation with Adversarial Training on Penultimate Activations [82.9977759320565]
Enhancing model prediction confidence on unlabeled target data is an important objective in Unsupervised Domain Adaptation (UDA)
We show that this strategy is more efficient and better correlated with the objective of boosting prediction confidence than adversarial training on input images or intermediate features.
arXiv Detail & Related papers (2022-08-26T19:50:46Z) - Domain-Adjusted Regression or: ERM May Already Learn Features Sufficient
for Out-of-Distribution Generalization [52.7137956951533]
We argue that devising simpler methods for learning predictors on existing features is a promising direction for future research.
We introduce Domain-Adjusted Regression (DARE), a convex objective for learning a linear predictor that is provably robust under a new model of distribution shift.
Under a natural model, we prove that the DARE solution is the minimax-optimal predictor for a constrained set of test distributions.
arXiv Detail & Related papers (2022-02-14T16:42:16Z) - Boosting Unsupervised Domain Adaptation with Soft Pseudo-label and
Curriculum Learning [19.903568227077763]
Unsupervised domain adaptation (UDA) improves classification performance on an unlabeled target domain by leveraging data from a fully labeled source domain.
We propose a model-agnostic two-stage learning framework, which greatly reduces flawed model predictions using soft pseudo-label strategy.
At the second stage, we propose a curriculum learning strategy to adaptively control the weighting between losses from the two domains.
arXiv Detail & Related papers (2021-12-03T14:47:32Z) - Exploiting Position Bias for Robust Aspect Sentiment Classification [10.846244829247716]
We propose two mechanisms for capturing position bias, namely position-biased weight and position-biased dropout.
Our proposed approaches largely improve the robustness and effectiveness of current models.
arXiv Detail & Related papers (2021-05-29T04:41:09Z) - Robustified Domain Adaptation [13.14535125302501]
Unsupervised domain adaptation (UDA) is widely used to transfer knowledge from a labeled source domain to an unlabeled target domain.
The inevitable domain distribution deviation in UDA is a critical barrier to model robustness on the target domain.
We propose a novel Class-consistent Unsupervised Domain Adaptation (CURDA) framework for training robust UDA models.
arXiv Detail & Related papers (2020-11-18T22:21:54Z) - Adversarial Distributional Training for Robust Deep Learning [53.300984501078126]
Adversarial training (AT) is among the most effective techniques to improve model robustness by augmenting training data with adversarial examples.
Most existing AT methods adopt a specific attack to craft adversarial examples, leading to the unreliable robustness against other unseen attacks.
In this paper, we introduce adversarial distributional training (ADT), a novel framework for learning robust models.
arXiv Detail & Related papers (2020-02-14T12:36:59Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.