case-by-case problem solving pei wang temple university philadelphia, usa

7
Case-by-Case Case-by-Case Problem Problem Solving Solving Pei Wang Temple University Philadelphia, USA

Upload: randall-wilcox

Post on 03-Jan-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Case-by-Case Case-by-Case Problem Problem SolvingSolving

Pei WangTemple UniversityPhiladelphia, USA

Page 2: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Algorithmic Problem Solving

Use a computer to solve a problem: Problem is a class, solution is an algorithm

e.g., “sorting” to “quicksort” Problem is an instance, solution is a result

e.g., “sort [3, 2, 4, 1]” to “[1, 2, 3, 4]”

The former is done by human, the latter is done by computer following the algorithm

Page 3: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

No algorithm for it?

What if the computer has no algorithm for a problem instance?

Use a general-purpose algorithm

e.g., state-space search Find an algorithm first

e.g., machine learning

Page 4: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Solving it without algorithm?!

How about to directly solve the problem instance without following an algorithm?

“Nonsense! How can a computer run without algorithms?”

“But this process can still be carried out by algorithms not defined for this problem. An algorithm for problem P is not an algorithm for problem Q, right?”

Page 5: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Case-by-case problem solving

NARS represents a problem (instance) as an inference task, to be processed by a set of general-purpose inference rules

Rule selection is knowledge-driven, rather than algorithm-guided

Knowledge selection is context-sensitive Inference process is resource-restricted

Page 6: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Scopes of input-output

Each operation in NARS is controlled by certain algorithm, with fixed input-output mapping (see code)

The lifelong experience of the system fully determines its lifelong behaviors (see examples)

However, there is no function that maps “problem” to “solution”

Page 7: Case-by-Case Problem Solving Pei Wang Temple University Philadelphia, USA

Properties of CPS

CPS and APS are suitable for different (knowledge/resources) situations

In CPS, the following notions are different: Problem Solution Solvable problems Resource cost of a problem Scaling up

… …