The language prior problem in VQA makes the model directly predictis based on questions, causing the model’s performance to drop sharply outside the distribution. Current debiased methods often achieve good out-of-distribution generalization capabilities at the expense of in-distribution performance degradation. We propose a novel method combining Robust knowledge distillation and self-contrast Reasoning (RR-VQA) to solve the language prior problem in VQA. We propose the QAS module to select reasonable questions for images, perform knowledge distillation through the ID and OOD teacher models, and obtain pseudo answers after passing through the QAS module. The CRSG module we propose synthesizes four visual language positive and negative samples for contrastive reasoning, which effectively increases the semantic dependency of the images while avoiding performance degradation due to spurious correlations between questions and answers. Our method is model-agnostic and achieves state-of-the-art performance on VQA-CP v2 dataset while maintaining performance on VQA v2 dataset.

Visual Question Answering (VQA) is a critical multimodal task. Many VQA models have been shown to rely on language prior. To this end, we propose a new unbiased VQA method to solve the question of bias from the perspective of internal and external interaction of modal information and transformation of language question. In the part of internal and external interaction of modal information, we design a module that interacts with questions and images and another module that only interacts with itself within the image to generate advanced visual features. In the part of linguistic prior capture branch, We transform the original question into three sub-questions according to a certain strategy, and input the three sub-questions to participate in the training process. Extensive experiments show that our method achieves very competitive performance (61.45%) on the VQA-CP v2 dataset.

In order to detect a person's individual momentary thermal sensation and comfort, an increased number of models are developed that involve physiological data, especially skin temperatures and features based on the heart activity. In this paper, we investigate the feasibility of Machine Learning (ML) models which include physiological data based on two data sources: (1) a smartwatch and (2) a portable chest belt device. Further, we investigate the difference between thermal sensation and comfort votes and propose a new combined ground truth label. Data were collected in lab-like studies in a single-bed hotel room. Our study focuses on the detection of cold-induced thermal discomfort by varying the room air temperature between 24 °C and 20 °C. Results show that ML based approaches lead to accuracies up to 83.1% when using a chest belt device. Using a smartwatch, accuracies drop down to 79.8%. Our investigation shows that even though Heart Rate (HR) based features improve the prediction accuracy, the highest feature importances have distal skin temperature features.