AI DevOps (ModelOps) at Thomson Reuters
While it is no longer considered a new concept, ModelOps effective design, implementation, and best practices are still forming in the industry. There are many supporting technologies, options, and opinions on how to best execute ModelOps pipelines. Thomson Reuters is reaching a time when delivering AI in an accelerated and continuous fashion is an imperative at all levels of the organization. ModelOps is foundational for this endeavor.
We are actively exploring Machine Learning in the cloud and how Continuous Learning and Integration can be done efficiently and in a sustainable way for AI models. In collaboration with Amazon, we developed, for example, a Secure Content Workspace (SCW) providing researchers easy access to Thomson Reuters’ internal content and machine learning environments such as AWS SageMaker. We are also actively exploring new cloud services as they become available, researching, evaluating and standardizing their use within the context of research and of the broader technology organization.
ModelOps is an extension and adaptation of the DevOps principles for the artificial intelligence (AI) ecosystem. Its objective is to shorten the AI delivery lifecycle and ensure long term sustainability and quality levels of AI solutions. This is achieved through adoption of automation, continuous delivery and monitoring best practices. Our objectives within this research domain is to shorten the "AI Lifecycle", provide continuous delivery and increase quality. It focuses on exploring the methods and technology related to the emerging ModelOps space.
How Thomson Reuters accelerated research and development of natural language processing solutions with Amazon SageMaker. > Read More
Pesaranghader, Ali, Andrew Alberts Scherer, George Sanchez, and Saeed Pouryazdian. 2021. “Understanding Dataset Shift and Potential Remedies.” Vector Institute.
Related research areas
Multidisciplinary approach to the challenges we face in terms of AI adoption and building trust in our solutions. We explore concepts such as interpretability, explainability, transparency, fairness, privacy and security, and societal impact – central to our AI Principles and company purpose.
Natural Language Processing
Natural Language Processing (NLP) focuses on designing algorithms to parse, analyze, mine, and ultimately understand and generate human language. NLP with a focus on text data, is one of our core enabling technologies given our customers’ work in information heavy segments.
Information Retrieval and QA
Our customers need the right information, in the right context, and often under tight time constraints. We adopt a comprehensive approach to the information findability problem, using a combination of search technologies, recommendation systems, and navigation-based discovery.