Building large learning models with herbal

Jaehyon Paik; Jong W. Kim; Frank E. Ritter; Jonathan H. Morgan; Steven R. Haynes; Mark A. Cohen

Building large learning models with herbal

Jaehyon Paik, Jong W. Kim, Frank E. Ritter, Jonathan H. Morgan, Steven R. Haynes, Mark A. Cohen

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

In this paper, we describe a high-level behavior representation language (Herbal) and report new work regarding Herbal's ACT-R compiler. This work suggests that Herbal reduces model development time by a factor of 10 when compared to working directly in Soar, ACT-R, or Jess. We then introduce a large ACT-R model (541 rules) that we generated in approximately 8 hours. We fit the model to learning data. The comparison indicates that humans performing spreadsheet tasks appeared to start with some expertise. The comparison also suggests that ACT-R, when processing tasks consisting of hundreds of unique memory elements over times spans of twenty to forty minutes, may have problems accurately representing the learning rates of humans. In addition, our study indicates that the spacing between learning sessions has significant effects that may impact the modeling of memory decay in ACT-R.

Original language	English (US)
Title of host publication	Proceedings of the 10th International Conference on Cognitive Modeling, ICCM 2010
Pages	187-192
Number of pages	6
State	Published - 2010
Event	10th International Conference on Cognitive Modeling, ICCM 2010 - Philadelphia, PA, United States Duration: Aug 5 2010 → Aug 8 2010

Other

Other	10th International Conference on Cognitive Modeling, ICCM 2010
Country/Territory	United States
City	Philadelphia, PA
Period	8/5/10 → 8/8/10

All Science Journal Classification (ASJC) codes

Artificial Intelligence
Modeling and Simulation

Cite this

@inproceedings{7576d9737a6e43f39685b89c94523b02,

title = "Building large learning models with herbal",

abstract = "In this paper, we describe a high-level behavior representation language (Herbal) and report new work regarding Herbal's ACT-R compiler. This work suggests that Herbal reduces model development time by a factor of 10 when compared to working directly in Soar, ACT-R, or Jess. We then introduce a large ACT-R model (541 rules) that we generated in approximately 8 hours. We fit the model to learning data. The comparison indicates that humans performing spreadsheet tasks appeared to start with some expertise. The comparison also suggests that ACT-R, when processing tasks consisting of hundreds of unique memory elements over times spans of twenty to forty minutes, may have problems accurately representing the learning rates of humans. In addition, our study indicates that the spacing between learning sessions has significant effects that may impact the modeling of memory decay in ACT-R.",

author = "Jaehyon Paik and Kim, {Jong W.} and Ritter, {Frank E.} and Morgan, {Jonathan H.} and Haynes, {Steven R.} and Cohen, {Mark A.}",

year = "2010",

language = "English (US)",

pages = "187--192",

booktitle = "Proceedings of the 10th International Conference on Cognitive Modeling, ICCM 2010",

note = "10th International Conference on Cognitive Modeling, ICCM 2010 ; Conference date: 05-08-2010 Through 08-08-2010",

}

TY - GEN

T1 - Building large learning models with herbal

AU - Paik, Jaehyon

AU - Kim, Jong W.

AU - Ritter, Frank E.

AU - Morgan, Jonathan H.

AU - Haynes, Steven R.

AU - Cohen, Mark A.

PY - 2010

Y1 - 2010

N2 - In this paper, we describe a high-level behavior representation language (Herbal) and report new work regarding Herbal's ACT-R compiler. This work suggests that Herbal reduces model development time by a factor of 10 when compared to working directly in Soar, ACT-R, or Jess. We then introduce a large ACT-R model (541 rules) that we generated in approximately 8 hours. We fit the model to learning data. The comparison indicates that humans performing spreadsheet tasks appeared to start with some expertise. The comparison also suggests that ACT-R, when processing tasks consisting of hundreds of unique memory elements over times spans of twenty to forty minutes, may have problems accurately representing the learning rates of humans. In addition, our study indicates that the spacing between learning sessions has significant effects that may impact the modeling of memory decay in ACT-R.

AB - In this paper, we describe a high-level behavior representation language (Herbal) and report new work regarding Herbal's ACT-R compiler. This work suggests that Herbal reduces model development time by a factor of 10 when compared to working directly in Soar, ACT-R, or Jess. We then introduce a large ACT-R model (541 rules) that we generated in approximately 8 hours. We fit the model to learning data. The comparison indicates that humans performing spreadsheet tasks appeared to start with some expertise. The comparison also suggests that ACT-R, when processing tasks consisting of hundreds of unique memory elements over times spans of twenty to forty minutes, may have problems accurately representing the learning rates of humans. In addition, our study indicates that the spacing between learning sessions has significant effects that may impact the modeling of memory decay in ACT-R.

UR - http://www.scopus.com/inward/record.url?scp=78149394173&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=78149394173&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:78149394173

SP - 187

EP - 192

BT - Proceedings of the 10th International Conference on Cognitive Modeling, ICCM 2010

T2 - 10th International Conference on Cognitive Modeling, ICCM 2010

Y2 - 5 August 2010 through 8 August 2010

ER -

Building large learning models with herbal

Abstract

Other

All Science Journal Classification (ASJC) codes

Other files and links

Fingerprint

Cite this