CS 730/830: Intro AI Class Outro AI at UNH Wheeler Ruml (UNH) - - PowerPoint PPT Presentation

CS 730/830: Intro AI

Class Outro AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 1 / 12

Class Outro

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 2 / 12

The AI View of An Agent

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 3 / 12

percepts → → actions

Past

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 4 / 12

■

perception: supervising learning (handwriting recognition), unsupervised learning (shape finding) [ HMMs ]

■

reasoning: constraint satisfaction, propositional satisfiability, first-order logic theorem proving [ tree search, optimization ]

■

planning: state-space search, motion planning, domain-independent task planning, planning under uncertainty (MDPs) [ anytime and real-time planning, reinforcement learning ]

■

acting: filtering (MCL) [ control ] Not: cognitive modeling, ethics, NLP, vision, philosophy of mind

Present

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 5 / 12

■

Fri May 1: no recitation

■

Tue May 5 9-noon: project presentations 10+2 minutes/person

■

Mon May 11 2pm: final papers email PDF, tarball, HOWTO given tarball and HOWTO, raw results should be reproducible on agate

Tips for A Research Talk

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 6 / 12

■

problem (example!), approach, results, extensions

■

practice beforehand: word choice, timing

Tips for a Research Paper

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 7 / 12

■

use the standard form: introduction (motivate and define problem, summarize paper), previous work, your approach, experimental results, discussion, conclusion

■

write for someone who has taken an AI class but doesn’t know anything about your specific problem

■

don’t just plot results, explicitly describe what they show and the conclusions you draw from them

Future

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 8 / 12

■

UNH AI group: usually weekly (Google ‘UNH AI group’) sign up for the mailing list! Fall:

■

( Wheeler Ruml: CS 931 Planning for Robots )

■

( Momotaz Begum: CS 733/833 Mobile Robotics )

■

( Laura Dietz: CS 753/853 Information Retrieval )

■

Laura Dietz: CS 780/880 ML for Sequences and Text

■

Marek Petrik: CS 950 Reinforcement Learning Spring:

■

Marek Petrik: CS 750/850 Machine Learning

■

Marek Petrik and Mark Lyon: CS 757/857 Optimization

■

Momotaz Begum: CS 780/880 Computer Vision

■

Momotaz Begum: CS 933 Human-Robot Interaction

■

Laura Dietz: CS 953 Knowledge Graphs and Text

Evaluations

Class Outro ■ The AI View ■ Past ■ Present ■ Talk ■ Paper ■ Future ■ Evaluations AI at UNH

Wheeler Ruml (UNH) Lecture 27, CS 730 – 9 / 12

These are important! I take them seriously and so does my boss. For free response text, please address: 1. Things that were good about the class, things that need work. specific suggestions or general comments! 2. Things that I did well, things that I should work on. Things that Tianyi did well, things that Tianyi should work

• n

Thanks.

AI at UNH

Class Outro AI at UNH ■ AI at UNH ■ EOLQs

Wheeler Ruml (UNH) Lecture 27, CS 730 – 10 / 12

AI at UNH

Class Outro AI at UNH ■ AI at UNH ■ EOLQs

Wheeler Ruml (UNH) Lecture 27, CS 730 – 11 / 12

■

Marek Petrik: robust RL

■

Momotaz Begum: assistive robotics

■

Laura Dietz: Queripedia

■

Wheeler Ruml: heuristic search, planning

◆

rational real-time search (Tianyi)

◆

suboptimal and bounded suboptimal (William)

◆

real-time path coverage (Alex)

◆

• nline goal recognition design (Kevin)

◆

motion planning in a dynamic environment (Yi)

◆

group assignment (Brendan)

◆

ROP attack assembly (Daroc)

◆

physical TSP (Bryan, Lucas, Charles)

◆

situated temporal planning (Shahaf)

EOLQs

Class Outro AI at UNH ■ AI at UNH ■ EOLQs

Wheeler Ruml (UNH) Lecture 27, CS 730 – 12 / 12

Nope.