These are the year's most interesting AI research articles. It combines breakthroughs in artificial intelligence (AI) with data science. It is chronologically organized and includes a link to a longer article.

Accurate and Scalable Gaussian Processes for Fine-Grained Air Quality Inference

Air pollution is a global problem with significant health consequences. Monitoring air quality (AQ) at a finer scale is essential for reducing air pollution. Existing AQ station deployments are, however, few. Conventional interpolation methods must be more capable of understanding the complicated AQ phenomena. For AQ modelling, physics-based models require domain knowledge and pollution source data.

The authors present an AQ estimation method based on Gaussian processes in this study.

a) a non-stationary (NS) kernel to allow input-dependent smoothness of ft;

b) a Hamming distance-based kernel for categorical features; and

c) a locally periodic kernel to capture temporal periodicity are the critical aspects of our technique.

Using batch-wise training, the researchers expand their approach to a significant amount of data. Their method exceeds conventional baselines and is the most advanced neural attention-based method.

Anatomizing Bias in Facial Analysis

Existing facial analysis techniques have been demonstrated to produce biased results towards particular demographic subgroups. Due to their impact on society, it has become essential to ensure that these systems do not discriminate against individuals based on their gender, identity, or skin tone. It has prompted studies into identifying and mitigating bias in artificial intelligence systems.

In this study, the authors outline algorithms for facial analysis bias detection/estimation and mitigation. Their primary contributions consist of a thorough review of proposed algorithms for understanding bias and a taxonomy and comprehensive overview of existing algorithms for bias mitigation. In addition, they explore open issues within the field of biassed facial analysis.

DrawMon: A Distributed System for Detection of Atypical Sketch Content in Concurrent Pictionary Games

Pictionary, the popular sketch-based guessing game, allows examining cooperative gameplay with a common objective in environments with restricted communication. However, some players occasionally produce sketches with unexpected content. While such content is sometimes related to the context of the game, it frequently violates the rules and degrades the playing experience. DrawMon is a revolutionary distributed framework for automatically detecting anomalous sketch content in concurrent Pictionary gaming sessions.

The researchers create specific online interfaces to collect game session data and annotate atypical sketch content, resulting in AtyPict, the world's first dataset of atypical sketch content. In addition, they use AtyPict to train CanvasNet, a deep neural network detecting unusual material. CanvasNet is a vital component of DrawMon, utilized by researchers. Their analysis of post-deployment gaming session data demonstrates the efficiency of DrawMon for scalable monitoring and detection of abnormal sketch content. Beyond the game of Pictionary, our contributions also serve as a design reference for customized atypical content response systems incorporating interactive whiteboards.