Google’s artificial intelligence team seeks to find solutions to computational problems, in theory, algorithms, journalism, machine learning, dialogue and other data streams that impact Google products and scientific progress. It focuses on two tools – software libraries that transform research results into products and services, and publications that communicate the results of his work to the community.
Real-world graph-based applications contain a variety of information about data relationships. The team’s main goal is to extend machine learning (ML) methods to better model relationships that are used in many Google products.
Google has a long history of creating and using machine learning techniques, having previously created the Google Core API application programming interface for supervised machine learning. Google’s AI team is actively working with other Google products such as Docs, Search and Ads to implement ML-based solutions for high-quality search.
It also includes supervised learning and partial/unsupervised learning. Google’s artificial intelligence team has developed a rule-based approach and successfully applied it to Google Search and advertising products, YouTube and Google Shopping.
The online clustering team provides clustering of datasets that can contain billions of data points, calibrating results from thousands of points per second. The goal is to provide scalable non-parametric clustering without assumptions. The team has developed design techniques to deal with biased information in the data.
Another exciting area of research is cross-lingual and cross-modal access to dynamically organized information, making writing, browsing and reading interesting. The Co-author team is integrating web content with Google Docs and has yet to propose other new applications.
Google’s artificial intelligence team filters data to detect, understand and model hidden user behavior. Structured data is important for all Google products such as event checking, search and Q&A. It uses a wide range of techniques including machine learning, search data mining and information extraction. The team also develops fast inference techniques for ML models that increase speed by more than 50 times and provide accurate solutions.
It designs automata, grammars and other models for speech and writing, translation and text mining. These can be combined and optimized to provide high accuracy, efficient speech recognition, text normalization and other functions. Sensitive content detection creates a comprehensive set of classifiers to detect any kind of offensive content, images or videos. Google’s artificial intelligence team has achieved this through various techniques such as ML models trained on images and text from the web.
Multiple Google AI teams have developed machine learning algorithms and systems that learn user preferences through personalized and targeted experiences. Google AI is developing systems to transform cloud-based ML models into models that work on resource-constrained mobile devices. It is also enriching online conversations with media understanding by using multimodal signals from images, video, text and the web.
Glassbox Learning conducts research and development to make machine learning more understandable without compromising accuracy. It also provides comprehensive assurances on the relationship between inputs and outputs. The team implements AdaNets, which adaptively learns network structure and weights.