Machine Learning from Verbal User Instruction

Explore machine learning from verbal user instruction in this lecture by Tom Mitchell from Carnegie Mellon University. Delve into the future of machine learning, focusing on sensor-effector system learning through human instruction. Examine the potential of learning within the sensor-effector closure of smartphones and discover the philosophy behind learning by instruction. Investigate natural language approaches, including CCG parsing, and learn how semantics for commands like "Tell" are derived. Analyze the teaching of conditionals and their impact on machine learning systems. Consider the implications of every user becoming a programmer and discuss the theoretical foundations needed for this emerging field of interactive learning.

Syllabus

Intro
The Future of Machine Learning
Sensor-Effector system learning from human instruction
Within the sensor-effector closure of your phone
Learning for a sensor-effector system
Our philosophy about learning by instruction
Machine Learning by Human Instruction
Natural Language approach: CCG parsing
CCG Parsing Example
Semantics for "Tell" learned from "Tell Tom I am late."
Outline
Teach conditionals
Teaching conditionals
Experiment
Impact of using advice sentences
Every user a programmer?
Theory needed