Introspective Learning : A Two-Stage approach for Inference in Neural Networks

Mohit Prabhushankar, Ghassan AlRegib

Keywords: Reasoning, Knowledge Representation, Robustness, Recognition

Abstract: In this paper, we advocate for two stages in a neural network's decision making process. The first is the existing feed-forward inference framework where patterns in given data are sensed and associated with previously learned patterns. The second stage is a slower reflection stage where we ask the network to reflect on its feed-forward decision by considering and evaluating all available choices. Together, we term the two stages as introspective learning. We use gradients of trained neural networks as a measurement of this reflection. We perceptually visualize the explanations from both stages to provide a visual grounding to introspection. For the application of recognition, we show that an introspective network is $4\%$ more robust and $42\%$ less prone to calibration errors when generalizing to noisy data. We also illustrate the value of introspective networks in downstream tasks that require generalizability and calibration including active learning and out-of-distribution detection. Finally, we ground the proposed machine introspection to human introspection in the application of image quality assessment.

One-sentence Summary: The paper proposes a framework that implicitly introspects against all possible decisions to create robust and calibrated decisions.