R&D and Innovation sessions

To unleash the full potential of AI, Intuit envisions a future that melds the best capabilities of machines and humans to deliver personalized customer experiences, all on one secure platform. Bharath Kadaba examines how Intuit combines rules-based knowledge engineering with data-driven machine learning and natural language processing to build the human-expert-in-the-loop AI systems of the future. Read more.

Natural language understanding (NLU) underlies a wide range of applications and services. Rich resources available for English do not exist for most other languages, but the questions of how to expand these resources without duplicating effort and if it's possible to develop language-agnostic NLU-dependent applications remains. Huaiyu Zhu, Dulce Ponceleon, and Yunyao Li believe the answer is yes. Read more.

Statistical manufacturing has remained largely unchanged since postwar Japan. AI and DL allow for nonlinear feedback and feed-forward systems to be integrated for real-time monitoring and evolution of each part assembly. Vadim Pinskiy explores a system capable of detecting, classifying, and automatically correcting for manufacturing defects in a multinodal process. Read more.

Cell shaping is used to configure radio antenna parameters to improve the service quality. Julien Forgeat explores a reinforcement learning (RL) approach to configuring radio antenna parameters using industry-leading radio simulators from Ericsson and UC Berkeley RISELab's Ray distributed compute framework together with its built-in RL algorithm in RLlib. Read more.

Robot technologies are becoming more capable and affordable. Yet even though technologies like natural language processing, mapping, and navigation are becoming more mature and standardized, it's often difficult to quantify human social behavior with algorithms. Dylan Glas and Phoebe Liu highlight some of the ongoing research to enable human-robot interaction. Read more.

Tzvika Barenholz and Induprakas Keri detail Intuit’s efforts to deploy fully homomorphic encryption (FHE) in production, which allows models to be trained and run on encrypted data, and supporting Intuit’s commitment to the highest standard in data stewardship. You'll take a sneak peak at some of the optimizations and tricks that make FHE practical. Read more.

Using deep learning models to perform natural language inference (NLI) is a fundamental task in natural language processing. Chaitanya Shivade introduces a recently released dataset, MedNLI, for this task in the clinical domain, describes state-of-the-art models, explores how to adapt these into the healthcare domain, and details applications that can leverage these models. Read more.

Dejan Milojicic examines a software stack designed for the special-purpose machine learning accelerator. The software stack improves usability and programmability of the accelerator, making it accessible from common machine learning frameworks. The software toolchain also exposes the intricacies of the parallelism of the accelerator while hiding its complexities. Read more.