The data learning problem in cognitive architectures☆
Section snippets
The data learning problem
The data learning problem is a phenomenon that arises when an agent faces the task of acquiring declarative information provided by an external source. Paradigmatic tasks are, in the laboratory, paired associates, where in response to a stimulus word such as shelf, the agent has to learn to say the digit eight; or in everyday life, learning the correct answer to a question, such as that the capital of France is Paris. In such situations, both the question and the answer (or the “stimulus” and
The data learning problem in Soar
The data learning problem was first noticed and discussed (and termed the data chunking problem) by Rosenbloom et al., 1987, Newell, 1990 in early work with Soar, and later substantially re-analysed by others (Vera et al., 1993, Young and Lewis, 1999). All these authors trace the problem back to the learning mechanism employed by Soar.
In Soar, all knowledge – both declarative and procedural – is encoded as if ⇒ then associations in production rules. These production rules implement a search in
Generality of the data learning problem
Although the data learning problem has to date been documented and analysed only for Soar, our account of its origins suggests that the problem is potentially of wide generality and should arise also for other, or perhaps even all, cognitive architectures. After all, the only assumption needed to carry the argument is that learned associations incorporate the information attended to by the agent – and that would seem to be a reasonable tenet of most architectures and associative theories.
So, is
The data learning problem in ACT-R
ACT-R (Anderson and Lebiere, 1998, Anderson et al., 2004) stores long-term knowledge in two different forms. Declarative memory (DM) holds information in the form of simple relational units called chunks, while production memory holds knowledge in procedural form as production rules. By default, facts are stored as chunks in DM. However, retrieval of chunks from DM is a potentially slow and error-prone process. Retrieving an item accessed only infrequently can take an appreciable fraction of a
The data learning problem in connectionist architectures
At first sight it seems implausible that connectionist models would suffer from the data learning problem, since such models are generally regarded as epitomising the mechanism needed for associating a given input with a specific output. However, closer examination raises doubts as to whether the usual story about associative learning and connectionism is indeed sufficient to enable networks to learn associations directly without encountering the data learning problem.
We consider first the
Concluding discussion
This paper has defined and explored the data learning problem and the way the problem manifests itself in a symbolic cognitive architecture, Soar; in a hybrid architecture, ACT-R; and in connectionist architectures. Despite their marked differences at the implementation level, the problem arises in a strikingly similar way in all three families of architecture. The paper further argues that the problem should in principle occur for all cognitive architectures.
The core of the phenomenon is that,
References (15)
- et al.
Eliciting self-explanations improves understanding
Cognitive Science
(1994) - et al.
An integrated theory of the mind
Psychological Review
(2004) - et al.
The atomic components of thought
(1998) - et al.
Stimulus sampling theory
Two-phase model for prompted recall
Psychological Review
(1970)Classification and cognition
(1994)- et al.
Why there are complementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory
Psychological Review
(1995)
Cited by (0)
- ☆
An earlier version of this paper was presented at the International Conference on Cognitive Modelling, in Pittsburgh PA, in July 2004 (Young, 2004). I am grateful to Anna Cox, Frank Ritter, and three anonymous referees for constructive comments on the original version.