Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Elements of Active Continuous Learning and Uncertainty Self-Awareness: a Narrow Implementation for Face and Facial Expression Recognition

Stanislav Selitskiy

TL;DR

This work tackles reliable face recognition and facial expression recognition under makeup and occlusion OOD conditions by introducing a self-awareness mechanism: a meta-learning supervisor ANN monitors a CNN ensemble, derives an uncertainty descriptor from ensemble activations, and learns a trustworthiness threshold to decide when predictions should be trusted. When uncertainty is high, the system triggers active learning with occasional Oracle labeling and updates the CNN ensemble, while a lifetime SNN component enables continuous adaptation. A memory-augmented loss layer stores performance statistics to refine the trust threshold, yielding improved trusted accuracy for FR and FER and enabling substantial label-efficient learning. The approach affords interpretable trust decisions and practical gains in real-world, low-label regimes, pointing to scalable uncertainty-aware recognition in constrained settings.

Abstract

Reflection on one's thought process and making corrections to it if there exists dissatisfaction in its performance is, perhaps, one of the essential traits of intelligence. However, such high-level abstract concepts mandatory for Artificial General Intelligence can be modelled even at the low level of narrow Machine Learning algorithms. Here, we present the self-awareness mechanism emulation in the form of a supervising artificial neural network (ANN) observing patterns in activations of another underlying ANN in a search for indications of the high uncertainty of the underlying ANN and, therefore, the trustworthiness of its predictions. The underlying ANN is a convolutional neural network (CNN) ensemble employed for face recognition and facial expression tasks. The self-awareness ANN has a memory region where its past performance information is stored, and its learnable parameters are adjusted during the training to optimize the performance. The trustworthiness verdict triggers the active learning mode, giving elements of agency to the machine learning algorithm that asks for human help in high uncertainty and confusion conditions.

Elements of Active Continuous Learning and Uncertainty Self-Awareness: a Narrow Implementation for Face and Facial Expression Recognition

TL;DR

This work tackles reliable face recognition and facial expression recognition under makeup and occlusion OOD conditions by introducing a self-awareness mechanism: a meta-learning supervisor ANN monitors a CNN ensemble, derives an uncertainty descriptor from ensemble activations, and learns a trustworthiness threshold to decide when predictions should be trusted. When uncertainty is high, the system triggers active learning with occasional Oracle labeling and updates the CNN ensemble, while a lifetime SNN component enables continuous adaptation. A memory-augmented loss layer stores performance statistics to refine the trust threshold, yielding improved trusted accuracy for FR and FER and enabling substantial label-efficient learning. The approach affords interpretable trust decisions and practical gains in real-world, low-label regimes, pointing to scalable uncertainty-aware recognition in constrained settings.

Abstract

Reflection on one's thought process and making corrections to it if there exists dissatisfaction in its performance is, perhaps, one of the essential traits of intelligence. However, such high-level abstract concepts mandatory for Artificial General Intelligence can be modelled even at the low level of narrow Machine Learning algorithms. Here, we present the self-awareness mechanism emulation in the form of a supervising artificial neural network (ANN) observing patterns in activations of another underlying ANN in a search for indications of the high uncertainty of the underlying ANN and, therefore, the trustworthiness of its predictions. The underlying ANN is a convolutional neural network (CNN) ensemble employed for face recognition and facial expression tasks. The self-awareness ANN has a memory region where its past performance information is stored, and its learnable parameters are adjusted during the training to optimize the performance. The trustworthiness verdict triggers the active learning mode, giving elements of agency to the machine learning algorithm that asks for human help in high uncertainty and confusion conditions.

Paper Structure

This paper contains 9 sections, 7 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Proposed Solution
  3. Uncertainty Meta-learning
  4. Active CNN ensemble Learning and Life-time SNN Learning
  5. Data Set
  6. Experiments
  7. Trusted Accuracy Metrics
  8. Results
  9. Discussion, Conclusions, and Future Work

Figures (4)

  • Figure 1: Meta-learning Supervisor ANN over underlying CNN ensemble.
  • Figure 2: Examples of the uncertainty shape descriptors for 0, 4, and 6 correct CNN ensemble FER predictions.
  • Figure 3: left - trusted threshold learned during the training phase (blue, dashed line), online learning changes for grouped test images (green), and shuffled test images (red) for FR task. Right - trusted accuracy against the trusted threshold for grouped test images for the FR task.
  • Figure 4: Examples of images for FER (anger expression) with the low trusted threshold (bad acting) - left and high trusted threshold (better acting) - right.