The Future of ACT-R

Revisited

 

Fourth Annual ACT-R Workshop

Carnegie Mellon University

August 5, 1997

 

Christian Lebiere

cl+@cmu.edu

 

Last Year’s Changes

 

• Eliminate lists as built-in type

unbounded span?

chunks as cons cells

 

• Eliminate generalized negation

exhaustive search? latency?

conflict resolution can handle it

 

• Eliminate instantiation-specific PG-C

q, a, r, b only as (learned) numbers

can still introduce an arbitrary evaluation function through the :value parameter

 

• Disallow bindings to nil?

 

Plus

 

• New dependency-based analogy

Better syntax

More flexible, general mechanism

 

Conflict Resolution

Currently:

 

Instantiations generated in parallel and evaluated in order of increasing latency.

 

Problems:

 

• Cutoff latency is NOT matching latency.

 

• Assumes excessive parallelism.

 

Proposal:

 

• Productions that match the goal are

instantiated sequentially.

 

• Highest PG-C considered first.

 

• No multiple instantiations.

 

• Production latencies are cumulative.

 

• Latency of retrieval failure is latency

of activation threshold (0.0 by default).

Q & A & R & B

 

Currently:

 

Q&A and R&B are the probability of success and cost of the action of the present production and all productions until ultimate success or failure, resp.

The value of a subgoal is the PG-C of the instantiation, with multiple subgoals getting half the value of the previous one.

 

Problems:

 

• Need to estimate the success of the

matching part of the production.

 

• Arbitrary termination points for success

and failure leads to improper credit

assignment and lack of generalization.

 

• Subgoals are penalized multiple times.

 

• Subgoal value gets exponentially low.

 

Q & A & R & B cont.

 

Proposal:

 

• Q&A apply to the matching, action and

subgoals of a production.

 

• R&B apply to future productions until

the current goal is popped.

 

• When learning R&B, a goal fails if the

popping production is marked as :failure

or if the goal is popped by default.

Otherwise the goal succeeds.

 

• When learning Q&A, a production fails if

the matching or action part fails or any

of its subgoals fail.

 

• The value of a subgoal is equal to rG-b

of the creating instantiation. For nth of

many subgoals, divide rG-b by n.

 

• For analogized productions,

estimate Q&A from the matching part.

 

Popping upon Failure

Currently:

 

The best instantiation considered is fired. If none is found, the system stops.

 

Problems:

 

• Sometimes gets lost considering

instantiations with very negative PG-C.

 

• Sometimes gets stuck in subgoals.

 

Proposal:

 

• Pop the goal when no instantiation can

be found with a positive PG-C.

 

• Return a :failure value to signal the

failure to parent goals.

 

• Provides an estimate of the time to

create (memorize) a wme by subgoaling.

 

• Replaces usual Give Up production.

 

Chunks

 

Currently:

 

Multiple copies of same chunk (goal) are created.

 

Problems:

 

• Subgoals have to be repeatedly

recomputed because they cannot be

retrieved or rehearsed.

 

Proposal:

 

• When a goal is popped, it is merged with

any identical wme, which is reinforced.

 

• Does not apply to rhs creation or

modification of wmes.

 

Activation Noise

Currently:

 

New Gaussian noise of mean 0 and variance AN is added to each wme at each iteration.

 

Problems:

 

• Too high: wide fluctuation from cycle to

cycle in ability to recall chunks.

 

• Too low: not enough randomness in

which chunks are recalled or not.

 

Proposal:

 

• Introduce a Permanent Activation Noise

added to a wme’s base level at creation.

Defaults to NIL (current situation).

 

• High PAN and low (T)AN provides for

randomness but consistency in recall.

 

• P/TAN ~ Representation/Processing.

 

The ACT-R System

 

 

(p future

 

• ACT-R 3.0

 

• development environment

 

• visual interface

 

• analogy authoring tools

 

• experimental paradigm software

 

& models models models

 

==>

 

integrate system components

 

integrate independent models

)