Presentation is loading. Please wait.

Presentation is loading. Please wait.

Fahiem Bacchus University of Toronto

Similar presentations


Presentation on theme: "Fahiem Bacchus University of Toronto"— Presentation transcript:

1 Fahiem Bacchus University of Toronto
What have we Learned Fahiem Bacchus University of Toronto URL for the location of these slides.

2 Many Commonalities The different communities are addressing many similar problems. Given a transition system I Searching for a state satisfying a particular property Verification, planning II Checking all states Ensuring that all states satisfy a particular property. Doing a computation that potentially requires examining all states. Verification, MDPs Planning with optimization? 11/18/2018 Fahiem Bacchus

3 Many Commonalities The different communities are addressing many similar problems. Given a transition system I Searching for a path satisfying a particular property Verification, planning II Checking all path Ensuring that all states satisfy a particular property. Doing a computation that potentially requires examining all states. Verification, MDPs Planning with optimization? 11/18/2018 Fahiem Bacchus

4 Differences Verification Planning MDPs
Typical types of state information Boolean Relations Numeric Techniques BDDs, SAT, Propositional Lifted reasoning, some conversion to the propositional Dynamic programming, Linear programming, Approximations “Complexity” High combinatorial complexity, Smaller sized spaces Low combinatorial complexity, large state spaces Low combinatorial complexity, smaller sized spaces, but added complexity of optimization 11/18/2018 Fahiem Bacchus

5 Differences Many of these differences are a result of the “dominant” problems in the field. As a result there are many useful techniques specific to each area. In the future most of these fields would like to expand their scope, this will eliminate many of these differences. 11/18/2018 Fahiem Bacchus

6 Differences MDPs there is a desire to deal with relational representations. Relations become useful in verification when dealing, e.g., with software. Planning would like to deal with numeric resources and optimization. 11/18/2018 Fahiem Bacchus

7 Potential for Cross fertilization
As a result there will be a growing need to understand, adapt, and apply techniques from other areas in your own area. E.g., Using CSPs to do verification of RTL specifications (Drechsler) Using planning techniques in test generation (Damm) 11/18/2018 Fahiem Bacchus

8 Hurdles to overcome Collaboration Accessibility
Some useful collaborations will hopefully arise from the workshop. Trying to address benchmark problems with different techniques. Accessibility It can be difficult to access the literature in other fields The domain details are not necessarily abstracted from the search problems. To bridge this gap is a challenge. 11/18/2018 Fahiem Bacchus


Download ppt "Fahiem Bacchus University of Toronto"

Similar presentations


Ads by Google