It is commonly assumed that language refers to high-level visual concepts while leaving low-level visual processing unaffected. This view dominates the current literature in computational models for language-vision tasks, where visual and linguistic inputs are mostly processed independently before being fused into a single representation. In this paper, we deviate from this classic pipeline and propose to modulate the entire visual processing by a linguistic input. Specifically, we introduce Conditional Batch Normalization (CBN) as an efficient mechanism to modulate convolutional feature maps by a linguistic embedding. We apply CBN to a pre-trained Residual Network (ResNet), leading to the MODulatEd ResNet (MODERN) architecture, and show that this significantly improves strong baselines on two visual question answering tasks. Our ablation study confirms that modulating from the early stages of the visual processing is beneficial.

Artificial intelligence (AI) has changed how modern applications interact with us, and machine learning has driven the modernization of various processes and systems. We all know that artificial intelligence has strengthened its hold on how we interact with computers. Siri, Alexa, Cortana and Hey Google are now natural parts of our life, and references to them regularly appear in mainstream entertainment, such as TV shows and advertisements. Artificial intelligence and machine learning are also a natural part of our evolving ecosystem. People understand, for example, that the "social media algorithm"--the infamous algorithm that decides who sees what social media content--is driven by artificial intelligence.

Over the past few years, many organisations have found that although they train great models, they don't always gain long-term value from them. The reason for this is deployment and monitoring. How long does inference take? You might need a GPU. Do you want to predict in batches?

Grok's approach is different from first generation AIOps solutions. We believe in real-time processing of data and not learning from historical data. We believe in unsupervised learning and building sophisticated machine learning models without sophisticated configuration and programming. We utilize minimal data sets to quickly form observations early but also to continuously learn, recognize patterns and become more intelligent over time. In addition, Grok is the only AIOps solution to also eliminate and reduce the time and effort spent on manual administrative tasks – no more static rules, cookbooks, or toolkits or CMDBs that you will need to rely on to reap the benefits of AIOps.