At its core, Google’s mission is to organize the world’s information and make it universally accessible and useful. The challenges of realizing this mission span the breadth of Computer Science and Engineering – including the study of algorithms, artificial intelligence, computer vision, cryptography, data mining, distributed and parallel computing, human-computer interaction, hypertext and the Web, information retrieval, machine learning, machine translation, market algorithms/economics, natural language processing, networks, optimization, programming languages, robotics, security and privacy, software engineering, speech and auditory processing, video processing, and virtual reality. The enormous scale of Google’s operation also leads to fundamental questions in the development, deployment and evolution of planetary-scale systems. The technical excitement of all this and the real and practical benefit we can bring to Google’s users contribute to the excitement of doing research at Google.
Within Google’s overall mission, Google’s research mission is to innovate, to catalyze innovation, and to learn in ways that collectively help Google achieve its goals. Over the past decade, Google has redefined the state-of-the-art in many of the technical areas in which we work - our culture is one that emphasizes rapid and significant innovation. In order to sustain this, we keep our research and engineering efforts integrated, with research activities located in almost every engineering group as well as in the research group. This decentralized approach helps us to bring the best possible technology to our users by putting outstanding engineers and researchers in direct contact with the most relevant problems, and empowering them to deliver solutions to our end users directly. At Google, research ideas can immediately influence engineering products, and product experience can directly motivate and shape our research agenda. This is not to say we focus only on shorter term projects; rather, we focus on long term, extremely challenging problems but in a way where we can leverage Google’s strengths (usage, processing data, and the strength of Google engineering) and contribute to Google as a business. At the same time, we are enthusiastic participants in the global community. We contribute to open source activities, work with our academic colleagues, serve as program chairs and journal editors, and aggressively publish our work in academic literature. We are proud of the breadth of our external publications, and of the transformations we have helped to engender through key publications on Algorithms and Theory, AI, Economics and Market Algorithms, HCI, IR, Machine Learning, Machine Translation, NLP, Security, Speech, Systems, Vision, and many more.
Here are just a few examples of things we’ve accomplished.
Under this decentralized research model, we’ve been able to build an industry-leading systems infrastructure, while allowing the same engineers that have contributed to those innovations to remain members of the research community. The impact of their research is amplified by the fact that their ideas have usually been tested through real product implementation by the time of publication. Google’s systems research portfolio has to a significant degree established the state-of-the-art for how cloud computing should be architected (e.g., map-reduce and very large scale cluster computing), with publications that range from cluster architecture and energy efficiency to distributed file systems and cluster-level programming models.
HCI researchers at Google have enormous potential to impact the experience of Google users as well as conduct innovative research. Grounded in user behavior understanding and real use iteration, Google’s HCI researchers invent, design, build and trial real-scale interactive systems in the real world, typically exploring areas where products and features may not yet exist. We declare success only when we impact the well-being of our users and user communities, often through new and improved Google products.
HCI research has fundamentally impacted the design of Gmail and Android, and we are engaged in a variety of HCI areas such as mobile and ubiquitous computing, social computing, interactive visualization and visual analytics, as well as search and browsing interfaces. Projects include gesture and touch interaction; activity-based and context-aware computing; recommendation of social and activity streams; analytics of social media engagements, and end-user programming. As Google at large we follow a simple but vital premise: “Focus on the user and all else will follow.”
We view translation of human languages is a key part of Google’s mission to make information accessible across language boundaries. Our approach to machine translation is data-driven and empirical, partially due to the scale at which we must operate. We employ statistical techniques to learn translation models from very large quantities of parallel and monolingual text relying heavily on our large computing clusters and corresponding systems infrastructure. Our learning process is largely language-independent which allows us to build machine translation systems for many languages (assuming the availability of training data) very efficiently. On translate.google.com we allow Google users to translate text and web pages between more than 41 different languages, or perform web search in foreign languages. Research challenges abound as we aim for ever higher quality and more languages.
Google’s speech research has two aims: 1) to enable simple mobile access to the web and to Google’s services by making spoken input and output ubiquitously available, and 2) to facilitate search and organization of spoken data, which constitutes a large portion of the world’s information. Achieving these goals will require dramatic improvements in core speech technologies as well as a deepened understanding of the integration of speech technology into effective user interfaces. Our approach is based on building statistical models of speech at Google scale, taking advantage of massive amounts of data and computational power. Our research focuses on improving our algorithms for automatic learning from data, and on building bigger and richer models to better capture the complexities of speech. The speech group is involved in both research and the development of services. A number of services have been deployed, including GOOG-411 (over-the-phone business finder), Google Search by Voice (spoken input of search queries when mobile), and voicemail transcription for Google Voice.
Language translation and speech recognition, as well as Web search, rely on the statistics of natural language to infer the best responses to user requests. More data and better algorithms and systems have enabled our researchers to develop statistical models that capture in greater detail the structure and meaning of a growing range of linguistic expressions to support more accurate inferences of user intent, whether in analyzing sentences in English to better translate them into Japanese, or in inferring the semantic class and typical attributes of an entity in context to better find important information about the entity. The overall research challenge for natural language processing is to be able to combine multiple sources of information to carry out reliable inferences across many domains, genres, and user needs.
Most of the world’s data resides in pixels, but its usefulness resides in our capacity to infer what the pixels are about. Our efforts in computer vision focus on three main areas: the meaning of still and moving images, developing semantic similarity measures for visual objects, and synthesizing meaningful composites for visualization and browsing of large image collections. Overall our approach is data-driven and specifically uses large datasets on parallel computing clusters to directly solve the problem at the target scale. Our ability to mine meaningful information from multimedia positively influences several Google products including Google Image Search, YouTube, Google Maps, Google Earth, Google News, Orkut, Picasa, and more.
Much of our work on language, speech, translation, and visual processing relies on machine learning. In all of those tasks and many others, we gather large volumes of direct or indirect evidence of relationships of interest, and we apply learning algorithms to generalize from that evidence to new cases of interest. Machine learning at Google raises deep scientific and engineering challenges. Contrary to much of current theory and practice, the statistics of the data we observe shifts very rapidly, the features of interest change as well, and the volume of data often precludes the use of standard single-machine training algorithms. When learning systems are placed at the core of interactive services in a rapidly changing and sometimes adversarial environment, statistical models need to be combined with ideas from control and game theory, for example when using learning in auction algorithms. Google Research is at the forefront of innovation in machine learning with one of the most active groups working on virtually all aspects of learning and a strong academic presence through technical talks and publications at major conferences and journals.
From an Algorithms research point of view, Google’s mission presents many exciting optimization challenges, all of great interest. These include optimizing internal systems such as scheduling the machines that power the numerous computations done each day, as well as optimizations that affect the core products and users, from determining the auctions for online sponsored search and offline TV media, to automatic management of ad campaigns and developing statistical estimates of ad performance, finding best paths, determining reputation scores, and many others. Google researchers routinely innovate in these problems areas with both papers in top conferences, and product impact in Google Maps, Local Search, Ads, and more …
While Google has made its mark on searching for unstructured information, there is a significant amount of structured data on the Web that is extremely valuable to our users. Our first main direction of research has been on tapping the content stored in databases behind forms, known as the deep web (or the invisible web). Our research has developed methods for automatically analyzing web forms and submitting meaningful queries to them, to obtain HTML pages that can be inserted into the Google index. As a result, we are able to show deep-web results to our users in over 1000 queries per second (as of January 2008). A second line of research investigates the collection of HTML tables on the Web that contain valuable data. We extracted a corpus of over 150 million high-quality tables and have developed methods for searching tables in response to queries. Our long-term goal is to develop search techniques that blend results from any relevant structured data source on the Web, whether it is behind a form, in a table or a mashup, with other search results. Recently, we have also created a new type of database services, called Fusion Tables, to enable storage, sharing, and collaboration based on relational data.
We also actively engage in the scientific community, publishing our results in top journals and conferences and contributing to open source software projects. Google started out as our founders’ graduate school research project; since those humble beginnings, we’ve supported the teaching and learning experience, and continued to maintain strong relations with leading academic institutions worldwide. The Google Research Awards program aims to identify and support world-class, full-time faculty pursuing innovative research in areas relevant to Google’s mission. We encourage award recipients to participate in Google’s own academic environment by giving talks, engaging in discussions with our researchers, and sharing ideas and insights. The Visiting Faculty program enables leading academics to work at Google for periods of 6-12 months and take advantage of our challenging research problems, our wealth of data, our world-class computing infrastructure, and the opportunity to deploy research in a forum that will be used by millions of people. The Google Fellowship program recognizes and supports graduate students doing exceptional work in Computer Science as part of their quest to discover and achieve great things. Read more about all our University collaborations here.
We could discuss other work in privacy and security, human-computer-interaction, computer vision and very many other areas and you can check this out for a list of recent papers. If Google meets with success in its research mission, it will be making substantial progress on many of the grand challenge problems in computer science and artificial intelligence.
At Google, our spirit of collaboration, innovation, and of rapid launch and iteration, permeates everything we do and means we're able to generate great research results AND surprise and delight millions of users with technology that they would not have dreamed possible.