Dear Gavin,
Congratulations! Your paper, 5 (A Logic for Specifying Agent Actions and Observations with Probability), has been
accepted for poster presentation at STAIRS 2012. A total of 48 papers
were submitted, of which 32 were accepted (67% acceptance rate).
In addition to presenting a poster, you will have 5 minutes to
give a brief summary of your work orally. Please limit yourself
to 5 slides, in pdf format, that you are supposed to send
to us via email a week before STAIRS.
The camera-ready version of your paper is due
*** June 24. ***
That is in less than a week. Yes, this is a tight schedule but
this is due to an internal deadline given to STAIRS by the
printer. Sorry!
We require *both* the PDF as well as the complete Latex/Word
sources (as some archive such as zip or tgz) of your paper
in IOS Press' FAIA format:
http://www.frontiersinai.com/?q=templates
Please carefully consider the reviews of your
paper (see below) when preparing your final version. You
can upload the camera-ready copy (PDF + zipped sources) via
the EasyChair system at
http://www.easychair.org/conferences/?conf=stairs2012
The registration page as well as details on the accomodation
can be found at the general ECAI hompage:
http://www2.lirmm.fr/ecai2012/
We're looking forward to a great STAIRS, due to your
contribution and the invited speakers:
Alan Bundy (University of Edinburgh)
Gemma C. Garriga (INRIA Lille Nord Europe)
Malte Helmert (University of Basel)
Andreas Krause (ETH Zurich)
Michele Sebag (Université Paris Sud)
See you in Montpellier!
Sincerely yours,
Kristian Kersting and Marc Toussaint
STAIRS 2012 Programme Co-Chairs
----------------------- REVIEW 1 ---------------------
PAPER: 5
TITLE: A Logic for Specifying Agent Actions and Observations with Probability
AUTHORS: Gavin Rens, Gerhard Lakemeyer and Thomas Meyer
SUMMARY:
The submitted paper investigates a formal modal logic proposed in a
previous paper of the authors. This logic allows to represent and reason
probabilistically with uncertain actions and observations. A logic
reasoning procedure called "tableau method" is proposed and conjectures
about formal guarantees are discussed. The proposed logic is set into
relation with partially observable Markov decision processes (POMDPs),
but no theoretical results with respect to equivalence or expressivity
are given. No experimental results are reported.
COMMENTS:
Reasoning about uncertain actions and observations is a core topic in
AI. Whether logic reasoning is an appropriate approach for this purpose
is unclear; hence, the investigation of the paper to answer this
question is well motivated. However, the paper does not answer,
neither in theory nor in experiments, whether the proposed logical
formalism is superior to existing formalisms (such as POMDPs) in
expressivity or efficiency. In particular, if the authors' goal is to
show in the future (as they state) that their is logic is decidable,
then strong restrictions must have been made somewhere on the
expressivity of the logic. It would be great if this was discussed more
clearly.
The only contribution of the paper seems to be the proposed tableau
method for reasoning which is exposed in great detail and might indeed
be a suitable reasoning approach. Formal guarantees are not given,
however.
The paper is often hard to follow: plenty of formal concepts and symbols
are presented, often without intuition. Likewise, the presentation of
the tableau method in form of a very long list of logical rules without
explanation does not increase comprehensibility.
Theorem 3.1: "Soundness has been proved". Where? By whom? If this
theorem has been proved somewhere else, why is it stated as a theorem
here (which suggests a contribution of the submitted paper)?
Theorem 3.1: "From our research, termination seems achievable". Which
research? Why is this statement justified?
Sec. 4 examples: It is nice to have many examples. However, it might be
better to focus on a single example and explain it in more detail on an
intuitive level: what does the statement to be proven mean? What's the
intuition behind the individual steps?
Akronyms should be spelled out before their first usage. For example,
what is LUAP in the introduction?
Introduction, definition of "cognitive robots" as "field of formal
logic dedicated to knowledge repsrentation": in contrast, the reviewer
is used to the usage of "cognitive robots" as a fuzzy term to subsume
all sorts of "intelligent" reasoning and behavior of real-world robots
(not restricted to logic).
Definition 2.2.: Why does it make sense to formally distinguish between
rewards and costs?
The caligraphic symbols, for example in the introduction for POMDPs,
can be hard to read and distinguish.
Sec. 2.1 "contains three sorts": these are four; and what are "sorts"
anyway?
Sec. 2.1, definition of a POMDP reward function: the expected rewards
for future states is captured by the value of states, not by their
reward.
Sec. 2.1, end: the concept of "dynamic terms" is not used thereafter.
Sec. 2.2: "Because N is a bijection, it follows..." It is not clear why
this follows.
Tableau method: ternary relation "skeleton" -- why does it make sense
to specify a relation where the first argument must always be 0?
----------------------- REVIEW 2 ---------------------
PAPER: 5
TITLE: A Logic for Specifying Agent Actions and Observations with Probability
AUTHORS: Gavin Rens, Gerhard Lakemeyer and Thomas Meyer
The paper describes a language based on modal logic for representing probabilistic information about effects and observations of actions.
The paper is very detailed and is mostly concerned with defining the semantics and inference rules of the formalism. I think it will be hard for most readers to understand the rules and the language is also quite challenging.
It would be very beneficial to include some meaningful examples and showing how they can be encoded in your language. Currently the concrete examples of the language are purely abstract.
----------------------- REVIEW 3 ---------------------
PAPER: 5
TITLE: A Logic for Specifying Agent Actions and Observations with Probability
AUTHORS: Gavin Rens, Gerhard Lakemeyer and Thomas Meyer
This paper describes a formal language for POMDPs arising in and
aimed towards Cognitive Robotics applications where there is uncertainty
in perceptions. The authors describe the Specification for Logic of Actions
and Observations with Probability (SLAOP), a modal logic with actions
and observations as first-class objects specifically tailored towards
cognitive robotics.
- The authors intuit that SLAOP is sound based on a tableau method.
However, the work presented here does not actually prove decidability or
completeness with respect to the semantics of the outlined formalism.
The authors discuss their approach toward the proof and motivate their
intuitions with examples.
- The work is reasonably dense, and overall, is difficult to read. However,
that seems unavoidable given the subject under consideration.
--
This message is subject to the CSIR's copyright terms and conditions, e-mail legal notice, and implemented Open Document Format (ODF) standard.
The full disclaimer details can be found at http://www.csir.co.za/disclaimer.html.
This message has been scanned for viruses and dangerous content by MailScanner,
and is believed to be clean.
Please consider the environment before printing this email.