Jacob Valdez, Chance Huddleston, Kennedy Ratemo, Payton Field, Adam Mazen
Faculty Advisor(s):
Nadra Guizani
Department:
Computer Science & Engineering
Project Type:
Other Undergraduate
Title:
An Open Domain Conversation Chatbot
Abstract:
Open domain conversation represents significant mastery over both the mechanics of language production and the high-level and long-range interactions of social dynamics. Meeting these demands is no trivial task, and forgetting large and general multi-million dollar ML systems, the solution for college students during a resource-constrained pandemic is not as simple as scaling parameters and training data. Additionally, we seek to implement the entire inference training pipeline directly on mobile devices. However, in informal settings and when restricted to one user, open domain conversation may actually be closed. We believe a few well chosen priors can rapidly accelerate convergence such that few shot training can adapt an ML model to individualized user-agent interactions. The result we hope to achieve is a simple, yet powerful open domain chatbot on mobile Android phones.