From improving the user experience to making the computational infrastructure more effective, AI is a crucial aspect of making current software systems and products perform as well as possible. AI is often more effective than even precisely developed human-crafted heuristic tactics today, whether it’s reducing latency, boosting the quality of a video stream, or streamlining the interfaces to match a specific person’s demands. But, to use AI more effectively in various products, several challenges must be addressed: the system must accommodate software engineers without machine learning backgrounds; it must provide mechanisms to optimize for a variety of product goals, which may differ from closed-form machine learning loss functions; it must distinguish causal connections from data correlations; and it must scale efficiently to train, host, and monitor vast numbers of AI models. 

Meta Researchers Develop ‘Looper,’ an end-to-end AI platform that has been designed with easy-to-use APIs for optimization, personalization, and feedback collecting to answer these needs. Looper may be used to support the entire machine learning lifecycle, from model training to deployment and inference to product evaluation and optimization. Looper allows us to modify the existing products to leverage AI for personalized optimizations rather than having to rebuild them around AI models. Currently, the Looper platform hosts 700 AI models and produces 4 million AI outputs every second.

Every day, billions of individuals use Meta’s various services, each with their interests and preferences. 

Looper enables us to adapt many of them “out of the box” at an unparalleled scale while avoiding the need for complex, specialized code. 

Providing a user with dozens of options in a UI menu can make a product unpleasant, regardless of its value. However, various people have distinct menu preferences. Similarly, opportunistically prefetching items likely to be viewed by a user to a mobile device may considerably improve the product’s user experience, but without overwhelming the device’s hardware capabilities involves correctly forecasting what would be of most interest. 

Unlike heavyweight AI models for vision, voice, and natural language processing, which prefer offline inference and batch processing, Looper uses models that can quickly be retrained and deployed in large numbers on shared infrastructure. The software translates metadata from user and system interactions into supervised learning labels or reinforcement learning rewards.

Looper aims for quick onboarding, stable deployment, and low-effort maintenance of various intelligent techniques, with positive benefits evaluated and optimized directly in the application. Looper uses current horizontal AI platforms, such as PyTorch and Ax, with interchangeable models for machine learning tasks, and it separates application code from platform code.  

The vertical machine learning platform supports moderate-sized models from horizontal platforms to improve many aspects of software systems. These models are easily deployed and maintained without the need for model-specific infrastructure. At Meta, Looper is used by more than 90 product teams, with 690 models making 4 million predictions per second. 

The range of AI competence among product teams ranged from novices to experienced AI engineers, with AI engineers accounting for only 15% of teams using the Looper platform. An easy-to-use AI platform is generally the decisive factor for adoption for teams with no prior production AI experience, and AI investment continues once the utility is demonstrated. Behind high-level services, Meta’s platform handles concerns about software upgrades, logging, monitoring, and other issues, unlocking significant productivity gains. A smart-strategies platform boosts productivity for experienced AI developers by automating time-consuming tasks like writing database queries, constructing data pipelines, and setting up monitoring and alerts. It allows product creators to deploy more AI use cases than narrow-focus solutions. Regardless of past AI experience, platform adopters set up the first machine learning models in just a few days, acquired training data rapidly, updated their models, and launched new products in just a few months. 

There are significant opportunities to embed self-optimizing innovative product choice strategies into software systems to improve user experience, optimize resource use, and support additional functions. Looper, the AI platform, simplifies the deployment of smart strategies at scale by addressing the difficulties of product-driven end-to-end machine learning systems. It provides immediate, measurable benefits regarding data availability, ease of configuration, prudent use of available resources, reduced engineering effort, and product impact assurance. The broad support for product effect evaluation via causal inference and overhead resource measures is particularly appealing to platform adopters. 

Looper makes smart tactics more available to software engineers, allowing product teams to self-serve, design, deploy, and upgrade AI-driven capabilities.

Paper: https://arxiv.org/pdf/2110.07554.pdf

Source: https://ai.facebook.com/blog/looper-meta-ai-optimization-platform-for-engineers/

Marktechpost is a California based AI News Platform providing easy-to-consume, byte size updates in machine learning, deep learning, and data science research

© 2021 Marktechpost LLC. All Rights Reserved. Made with ❤️ in California

No thanks, I’m not interested!