Description:
This class focuses on building agents that achieve human-level performance in specialized technical domains and are adept at collaborating with humans using natural language. We draw upon research in cognitive and systems neuroscience to take advantage of what is known about how humans communicate and solve problems in order to design advanced artificial neural network architectures. Introductory lecture material is available here and if you want to learn more about the class, visit the 2019 and 2020 websites in the course archives.
Students taking CS379C this year will develop strategies for exploring next generation AI systems that interact cooperatively with humans and contribute to solving complex tasks such as automated code synthesis. Students will collaborate in teams to develop curriculum learning protocols for training interactive systems that learn to communicate by grounding their acquisition of language in the domain in which they are learning skills, as well as help in developing component technologies for special cases of more complicated systems.
To assist students working on these projects, we have invited researchers at DeepMind, Google Brain, Stanford, and local startups who have volunteered to help out by providing expertise and project consultation in the areas of neural code generation, natural language processing, early child development and language acquisition. Course materials including supplementary readings, the schedule of invited talks, and project-related lectures from previous years and invited speakers are available here.
Prerequisites:
CS229, CS230, PSYCH209 or an equivalent introductory machine learning course is required. CS221 or an equivalent introductory artificial intelligence course is recommended but not required. CS224N is a plus if you are interested in the natural language processing aspects of building human-machine interfaces.
Time and Location:
TTh 4:30 - 5:50pm, Request Zoom Link
Staff:
Instructor: Thomas Dean
Email: tldean [at] stanford [dot] edu
Office hours: by appointment
Course Assistant: Lucas Sato
Email: satojk [at] stanford [dot] edu
Grading:
- Project proposal due around midterm (30%)
- Project report due around finals week (70%)