LPLgrad: Optimizing Active Learning Through Gradient Norm Sample Selection and Auxiliary Model Training

TL;DR

A novel AL approach, Loss Prediction Loss with Gradient Norm (LPLgrad), designed to quantify model uncertainty effectively and improve the accuracy of image classification tasks, which outperforms state-of-the-art methods by order of magnitude.

Abstract

Machine learning models are increasingly being utilized across various fields and tasks due to their outstanding performance and strong generalization capabilities. Nonetheless, their success hinges on the availability of large volumes of annotated data, the creation of which is often labor-intensive, time-consuming, and expensive. Many active learning (AL) approaches have been proposed to address these challenges, but they often fail to fully leverage the information from the core phases of AL, such as training on the labeled set and querying new unlabeled samples. To bridge this gap, we propose a novel AL approach, Loss Prediction Loss with Gradient Norm (LPLgrad), designed to quantify model uncertainty effectively and improve the accuracy of image classification tasks. LPLgrad operates in two distinct phases: (i) {\em Training Phase} aims to predict the loss for input features by jointly training a main model and an auxiliary model. Both models are trained on the labeled data to maximize the efficiency of the learning process, an aspect often overlooked in previous AL methods. This dual-model approach enhances the ability to extract complex input features and learn intrinsic patterns from the data effectively; (ii) {\em Querying Phase} that quantifies the uncertainty of the main model to guide sample selection. This is achieved by calculating the gradient norm of the entropy values for samples in the unlabeled dataset. Samples with the highest gradient norms are prioritized for labeling and subsequently added to the labeled set, improving the model's performance with minimal labeling effort. Extensive evaluations on real-world datasets demonstrate that the LPLgrad approach outperforms state-of-the-art methods by order of magnitude in terms of accuracy on a small number of labeled images, yet achieving comparable training and querying times in multiple image classification tasks.

Figures (24)

• Figure 1: Typical active learning pipeline.
• Figure 2: A visualization of our LPLgrad approach. It highlights the two key components of LPLgrad: training and querying phases. The process involves 9 steps (represented in red circles): In steps 1 and 2, labeled images $\mathcal{L}$ are processed through the main model $\boldsymbol{w}_{\text{main}}$, where feature maps are extracted and subsequently fed into the auxiliary model $\boldsymbol{w}_{\text{aux}}$. This combination of models yields two distinct losses, $l_{\text{main}}$ and $l_{\text{aux}}$, which are then aggregated to compute the total loss $L_{\text{total}}$. From steps 3 to 9, samples from the unlabeled set $\mathcal{U}$ are processed, producing scores that are passed through a softmax classifier to obtain entropy values for each sample. These entropy values are back-propagated through $\boldsymbol{w}_{\text{main}}$ to compute the gradient. Then, LPLgrad calculates the gradient norm and selects the samples with the highest gradient norms for annotating and adding to $\mathcal{L}$.
• Figure 5: Classification performance comparison on Comprehensive Disaster Dataset.
• Figure 7: Confidence scores of $\boldsymbol{w}_{\text{main}}$ across various AL rounds on the CIFAR-10 dataset.
• Figure : (a) CIFAR-10 dataset.