Open Access

The robust generalization of models to rare, in- distribution (ID) samples drawn from the long tail of the training distribution and to out-of-training-distribution (OOD) samples is one of the major challenges of current deep learning methods. For image classification, this man- ifests in the existence of adversarial attacks, the perfor- mance drops on distorted images, and a lack of general- ization to concepts such as sketches. The current under- standing of generalization in neural networks is very lim- ited, but some biases that differentiate models from human vision have been identified and might be causing these lim- itations. Consequently, several attempts with varying suc- cess have been made to reduce these biases during training to improve generalization. We take a step back and sanity- check these attempts. Fixing the architecture to the well- established ResNet-50, we perform a large-scale study on 48 ImageNet models obtained via different training meth- ods to understand how and if these biases - including shape bias, spectral biases, and critical bands - interact with gen- eralization. Our extensive study results reveal that contrary to previous findings, these biases are insufficient to accu- rately predict the generalization of a model holistically

With the increasing popularity of voice user interfaces (VUIs), there is a growing interest in the evaluation of not only their usability, but also the quality of the user experience (UX). Previous research has shown that UX evaluation in human-machine interaction is significantly influenced by emotions. As a consequence, the measurement of emotions through the user’s speech signal may enable a better measure of the voice user experience and thus allow for the optimization of human-computer interaction through VUIs. With our study, we want to contribute to the research on speech emotion recognition in the context of voice user experience. We recorded 45 German participants while they were interacting with a voice assistant in a Wizard-of-Oz scenario. The interactions contained some typical user annoyances that might occur in voice-based human-computer interaction. Three analysis modules provided insight into the voice user experience of our participants: (1) a UX-questionnaire; (2) the UEQ+ scales for voice assistants; (3) speech emotion recognition with OpenVokaturi.

Artificial intelligence (AI) and Machine Learning (ML) are rapidly turning from trending topics to requirement for competitiveness for enterprises. For marketing departments, AI and ML offer potential for improvement of their processes such as optimizing user experience and personalizing campaigns for selected audiences. Nevertheless, the integration of new technologies such as AI and ML into the existing marketing mix portfolio means a great challenge for marketing managers as their implementation requires new skills and knowledge which is not always already developed. The objective of the paper is to demonstrate how an industry-university cooperation (IUC) can enable the adaptation to new business contexts. Thus, this paper proposes a framework on IUC involving different project phases. It describes the process for placing AI-generated individual content, recommendations and references for specific interests.

This study introduces EmbeddedTrain, an innovative algorithm optimized for on-device learning in deep neural networks, specifically designed for low-power microcontroller units. EmbeddedTrain refines sparse backpropagation by dynamically adjusting the level of sparity, including the ability to selectively skip training steps. This feature significantly lowers computational effort without substantially compromising accuracy. Our comprehensive evaluation across diverse datasets—CIFAR 10, CIFAR100, Flower, Food, Speech Command, MNIST, HAR, and DCASE2020—reveals that EmbeddedTrain achieves near-parity with full training methods, with an average accuracy drop of only around 1% in most cases. For instance, against full training, EmbeddedTrain’s accuracy drop is minimal, for example, only 0.82% on CIFAR 10 and 1.07% on CIFAR100. In terms of computational effort, EmbeddedTrain shows a marked reduction, requiring as little as 10% of the computational effort needed for full training in some scenarios, and consistently outperforms other sparse training methodologies. These findings underscore EmbeddedTrain’s capacity to efficiently manage computational resources while maintaining high accuracy, positioning it as an advantageous solution for advanced embedded device applications in the IoT ecosystem.