Skip to main content Skip to navigation

Feedback from WMB on WEB-EM-3 abstracts

General Advice

  • Think about diagrams for the paper - visualisation to convey the effect of models
  • Don't leave the paper-writing till the end unless you are quite sure of its structure, extent and content. You'll find it helpful to build the model in conjunction with conceiving and writing your paper, as EM will allow you to adapt your models to different themes and illustrative purposes. One of our concerns last year was that models and papers were not always sufficiently well-integrated - how your paper refers to your model is an examinable characteristic.
  • Think about a story and an intended audience. This needn't be the orthodox "I'm telling you about how and why I built this EM artefact" - it could be about how a model could be used in the classroom; in what respects it was appropriate to have built a model using EM principles; whether a projected implementation of an ambitious modelling task is feasible; what Eden developers need to know about how EM is practised and how tools could be improved etc.
  • Check out what has already been done - research / think of related studies / references outside EM - contextualise your study: don't be reluctant to invoke 'non-academic' software, such as SimCity.
  • Important to have references and relevant models in mind at an early stage.
  • Consider what is distinctive about application of EM in your context.
  • Beware that modelling behaviours / systems of interacting independent objects won't necessarily bring out the qualities of modelling with definitive scripts.
  • Be prepared to be critical of EM principles/tools / don't accept what you've been told uncritically - don't make claims without finding / citing evidence.
  • Consider using special-purpose notations to simplify model-building.
  • Be alert to the potential problems of limited stack size in Eden models - these can be addressed by recompiling tkeden.
  • Don't forget the value of keeping versions - sometimes especially helpful if you need to make a point about your modelling exercise in the context of your paper.
  • Good to consider the mode of presentation of state very prominently, as this will be critical in determining the potential scope and focus of your model. For instance, is a model of an evironment going to involve spatial realism, or will it be concerned with the visualisation of abstract quantities? If the former, what relevance does this have for observation and agency? if the latter, then where will agency and observation be invoked? Connected with this issue is whether your modelling will involve thinking in terms of LSD accounts, or dependency-based interaction between processes, without explicit agency being ascribed.
  • Beware / be suspicious of thinking about constraints, functionalities and clock-driven behaviours too prematurely. These aspects of the model are rarely conducive to thinking in an EM idiom.
  • Be clear about which kind of construal of a situation you are targetting in a particular aspect of the modelling e.g. are you considering the mechanisms by which you presume things happen, or simply modelling observation of the results?
  • Evaluate your work mentally against the criteria for assessment.
  • Be careful how you use the terms 'empirical modelling' and 'empirical model' where you mean 'EM' and 'EM model / artefact'
  • Don't underestimate the importance of a good title, both as a pointer to the direction of your own thinking throughout the model-building and paper-writing and as a guide to the reviewer / reader. Where possible, ensure that your title gives specific insight into your chosen topic, and is not so general that the reviewer begins in complete ignorance of what you're actually thinking about. You are of course free to change your title at any stage prior to submission.

A pure definitive language for Empirical Modelling

Nicolas Pope

About abstract:

purely definitive language as a possible EM tool

[is it really appropriate to have a 'purely definitive language for EM', as opposed to (say) a purely definitive platform for representing state over which EM can then be carried out? - quality of this sentiment has implications for the title?]

a more comprehensive model --> more comprehensive models?

user-interface comments - concern re implied orthodox CS stance that misses the point re EM

About model:

recreating --> re-creating

another model -> other models

"system could be used outside of EM"

Beware! be clear about where credit can be given for your work.

The primary modelling work involved in this project will be the creation of the tool itself, this presents many complex issues and requires significant time but does introduce me to the issues relating to Empirical Modelling"
Vital to focus on parallel-to-existing-models model-building and EM issues,

General comments: I think Gehring's paper is rather negative and limited in its vision, and possibly not as well-informed as it might be. In my view, it doesn't really buy into the radical shift of outlook that EM proposes, and identifies rather spurious problematic ingredients as a result.

AMULET (Brad Myers et al, CMU) is a very good interface source - better aligned to EM than Borning's work in my view - need to be careful about conflating constraint satisfaction with dependency maintenance.

Definitely worth looking into whether / if so how the virtual machine architecture that underlies tkeden could be directly reprogrammed in DOSTE - if that makes sense. This would be a key source of insight onto how DOSTE can serve in a different way as a EM tool.

Excellent new reference in Garrett and Foley

Complex issues surrounding other students working in parallel.

Definitive Logic Simulation

Nicholas Lee

About abstract:

authors work --> author's work, thrid --> third, depencies --> dependencies, empirical modelling --> Empirical Modelling, seperate --> separate

in dealing with feedback, necessarily must introduce some notion of current value of observable being defined in terms of its value "just previously" - as discussed in one of the lectures on the ADM

definitely worth exploring possibilities associated with clocking and use of Eden todo() feature - this could offer an alternative way to address the buffering issues

notation is a good idea, but quite ambitious

Where the interface for such an environment is concerned, you may want to consider using GEL in conjunction with Scout - for which there would be additional credit, because of its novelty

About model:

Good to put your work together with OLS into the empublic repository?

General comments:

Should find out about the Virtual Electronic Laboratory - d'Ornellas and Sheth old MSc project

Robots in the Home: Learning to do everyday tasks.

Andrew Giles

About abstract:

problem is 'lots of lines of code'?

an excellent idea to think of what it would mean for / what is involved in "getting a robot to create an EM construal"

word 'mental model' is slippery - references and clarification about how you are interpreting this would be valuable. Perhaps arriving at such an understanding of what is plausible and feasible would be sufficient in itself as an objective for a project on this scale

About model:

'added to the robot's mental model view' sounds rather facile

Also some danger of a simplistic perspective in "I will just assume it has the capability of detecting objects a certain distance from itself" - if you really mean 'objects' here!

General comments:

Title is rather broad and doesn't convey some essential aspects of proposal

This project sounds unrealistically ambitious ... definitely the model would be worth more than 50% if you could build it. Important to focus on the conceptual issues, and beware of facile assumptions that sometimes afflict traditional AI models.

Dautenhahn's papers are not a good guide to what is practical in robotics - more of interest in relation to the potential implications of robots if we could actually build them [references from the EU project in which she, Nehaniv and others are involved would be relevant - in particular, connected with house cleaning]

You might like to contrast with out-and-out logicist approaches such as Matthew L Ginsberg and David E Smith's Reasoning about Action I and II and dip into Smith's The Origin of Objects just to appreciate the potential conceptual difficulties

The Applications of Empirical Modelling within the scope of Engineering Design.

About abstract:

"This paper intends to ..."

really difficult issue about finding the language to talk about the context for the comparison

"prove that the system exhibits the behaviour stated by the theorem"

"that both methods cannot solve the frame problem"

"lacks the next step of taking the model and implementing it / creating additional value from it"

About model:

"Between the states of locked and unlocked there needs to be at least one check."

"[in context of "If the car is locked", and an unlock request is made] At this point, if the locks are up (unlocked) ... should revert to the locked state" - so what is the status of the precondition "If the car is locked ..." in this case? Shouldn't there be reference to observables - e.g. independent ways in which the car might 'know' that it is locked? - suprised that there is so little explicit reference to the LSD / ADM framework in the model account - sounds rather like a conventional perspective on specification activity where the physicality of an EM interactive situation model (ISM) is being set aside

General comments:

Would favour a more informative and focused title?

You might find the digitalwatchRoe... model to be of interest

Beware of the out-of-date terminology etc used in [3]

An Investigation into the Sasami Notation

Andrew Knight

About abstract:

Looks potentially like an excellent balance between new research and review. Paper will serve a very good purpose if it does shed light on the various ways in which different EM approaches to graphics are interrelated. Also useful to document what has been done already in ways that expose its potential and limitations.

About model:

An existing model? - meaning one not currently visualised in 3D?

Might be useful to examine / refer to the planimeter model in this context (see Charlie Care).

Like the overall structure of your description which suggests a good staged approach to the model-building.

General comments:

References could include more details of candidate models - no mention of pre-existing Sasami models for instance

A rather generic title - can it be made more informative?

Distributed Model of Early Naval Gun Control

Michael de Selincourt

About abstract:

This is well-written and conceived, and looks like the basis for a feasible and interesting project.

About model:

I like the concept behind the project very much, as it resonates so well with the notion of Human Computing, and the transition from human to automated agency. It may not be necessary to make a highly sophisticated "family of models" (of "trajectory of models") in order to deliver a highly original and effective outcome. I don't personally find the actual topic of the model-building (efficient gunnery control!) so congenial, but that can hardly be helped.

From a technical perspective, it may be that working with dtekeden is too ambitious here ... but merely conceptualising the transition from manual to automated procedures with reference to EM principles and concepts would be interesting. You could always include model fragments to support such an account.

General comments:

I have a close musician friend in the Netherlands who has done some significant work on databases etc that have been deployed in similar (contemporary) military applications, I believe. His name is Maarten Boasson, and you may be able to find one or references to his work on his own system "SPICE". This could be totally irrelevant to your proposal, but might possibly be worth a brief mention if it doesn't take you too far out of the way. His approach is not conventionally object-oriented, I believe: a characteristic underlying theme in his research is: Architectural Support for Integration in Distributed Reactive Systems (25th Annual Int Computer SW and Applications conference 2001). There is a paper critiquing OO in relation to embedded systems by him at:

Distances between cars in traffic - influences, dependencies and outcomes

Sandro Krumbein

About abstract:

definitive principals --> definitive principles

non public --> non-public

like the idea of your model proposal very much

would like to throw in my favourite idea - speedos that are calibrated so as to display the square of the speed (possibly multiplied by a factor to reflect the mass of the [loaded] vehicle), which give a more effective notion of the real danger as measured by kinetic energy - perhaps other measure such as momentum also significant

About model:

the challenge will be to keep the model to a realistic scope and size whilst doing justice to your vision

good to keep the merits of 2D versus 3D models in mind (not just a matter of time restrictions) - our implicit models of what must be taken account when driving past

a large vehicle parked in a busy shopping street include the invisible potential presence of a pedestrian about to cross the road in front of the vehicle for instance - something that can be more easily conveyed in 2D than 3D

General comments:

You might find paper no 084 of some interest in relation to relevant issues for perception.

Synesthesia For All ? Environmental Effects on Musical Experience

Robert Donovan

About abstract:

Promising title, but not sure whether the 'Environmental effcts on musical experience' quite captures the intended meaning. Overall, I like your idea a lot though.

[Not sure what the authentic spelling of 'synesthesia' is - some versions have an 'ae' in them I think.]

Not sure as to which of 'stave notation' and 'staff notation' is the more authentic, but both seem to be widely used - I have only used the latter. I think that likening a staff notation representation to a set of instructions for building the Cutty Sark is misleading in some ways - since 'building a ship' seems to me to be more analogous to 'composing a symphony' than performing it (and the score inform performance, but doesn't explicitly tell you anything about composition in general).

Must attribute the quotation from Cantwell-Smith:

'With four simple letters I can refer to a composer who hasn't existed for more than 200 years - not to a set-theoretic model of him, but to his very heart and soul'.

About model:

I initially found the discussion of the voices a little difficult to relate to Erlkoenig, since there is only one singer and hence no instance of two roles being played simultaneously. (A Mozart opera such as Figaro might provide better examples for the 'cocktail effect' concept - I can provide some examples if necessary.) Presumably, you could still imagine highlighting parts of the accompaniment so as to shift the listener's focus ... actually could be an excellent teaching aid where listening to music is concerned - especially, as you rightly invoke affect as well as mere focus of attention.

General comments:

It might be interesting to take on some other example as well as Erlkoenig - the extract from Beethoven's Pastoral symphony that I intended to play the other day is one candidate. Colours (as in storm clouds) and also dramatic visual and aural events, like lightning flashes and thunderclaps are very clearly identifiable in the 100 bars or so that I displayed - it would be possible to adapt the techniques used in Erlkoenig to drive a visualisation.

Investigation into the practicality using DOSTE for empirically modelling physical situations

Robert Newman

About abstract:

Title is quite broad - it should probably become more specific and informative once your case study has been developed more fully.

This looks like a simpler, and more appropriate, context in which to consider a DOSTE prototype than the Tamworth accident scenario you originally proposed to address.

The biggest danger here is that you get distracted by DOSTE considerations from addressing the most pertinent issues concerning applying EM and reflecting on the results.

dependancy --> dependency

About model:

Need to be careful here that you are not working with a misconception about EM: "DOSTE is rather unique in that it does not allow features such as subroutines; everything must be expressed as a dependancy." As the ADM framework illustrates, there is more to EM than modelling with dependency (perhaps look at the most recent paper [096] for some attempt to clarify this).

May be of course that you don't need to invoke agency in a significant way - consider the parallel with the lines model for instance.

How the model gets to be built and interacted with will be very important - good idea to highlight the process of model construction through an evolving definitive script also I think - to counteract any danger of lapsing into conventional constraint-based specification.

General comments: Alexander Repenning's Agentsheet software is an end-user oriented implementation that also features a bridge model and uses an event-driven rule-based approach that has some points of contact with EM. You should also check out and reference the model made 2 years ago in WEB-EM-01. Excellent that you have other references in mind of course, but important to think not only about how these are built (whether procedural or not), but about what implications this has when compared with using EM.

Potential for 3D empirical modelling in dost

James Dowling

About abstract:

empirical modelling --> Empirical Modelling, dost --> DOSTE

Dangers of the notion of 'currently under development by Nick Pope' don't need to be stressed, I'm sure - important that focus on stable parts of the DOSTE concept and issues that are well-aligned to EM principles. In particular, beware the trap of working within a constraint-satisfaction paradigm

Likewise "Was it as intuitive as tkeden?" begs many questions - about comparability of tools and objectives in particular. (I'm not sure that DOSTE and tkeden set out with the same agenda, though I am convinced that - suitably developed - DOSTE will be a far better underlying platform for EM than the hybrid low-level eden environment. In order to practice EM, it will be quite essential to support all the paraphernalia of observables, dependency, agency and the LSD/ADM elaboration of these however.)

About model:

The DOSTE 'system' rings slightly the wrong bell in connection with EM - so much of the effectiveness of DOSTE as a potential platform for EM relies upon the way in which it is made accessible to the modeller, and supplemented to give support for characteristic EM activities. I tend to think of DOSTE more as a machine wrapper that is well-suited to integrating with EM activities - and one that can stabilise into a new kind of operating system, rather than as a system that will need to be of itself extended in order to enable EM.

Am also unsure as to whether you have the right focus in describing your modelling exercise - putting modelling with time-dependent elements (like forces, velocity etc) upfront before discussing the modelling of state-as-perceived / experienced. Similar concerns affect: "The model will be based on what values the user sets for it to play out ..."

General comments:

Need to be very careful that you understand what is in keeping with EM principles - especially challenging in the context of working with DOSTE.