1 Episodic Memory for Soar Agents Andrew Nuxoll 11 June 2004.

Slides:



Advertisements
Similar presentations
PS2015 Lecture 2 Cognitive Models of Memory. Cognition Lecture 2 l Key issues where cognitive psychology parts from common sense »1. Deterministic (by.
Advertisements

Learning To Use Memory Nick Gorski & John Laird Soar Workshop 2011.
A LLISON S EIBERT & A LEXANDRA W ARLEN Efficient Episode Recall and Consolidation E MILIA V ANDERWERF & R OBERT S TILES.
Michael Alves, Patrick Dugan, Robert Daniels, Carlos Vicuna
Integrated Episodic and Semantic Memory in Robotics Steve Furtwangler, with Robert Marinier, Jacob Crossman.
Intelligent Agents Russell and Norvig: 2
1 Pertemuan 25 Learning Matakuliah: T0264/Intelijensia Semu Tahun: Juli 2006 Versi: 2/1.
Outline Introduction Soar (State operator and result) Architecture
File Management Chapter 12. File Management A file is a named entity used to save results from a program or provide data to a program. Access control.
Artificial Intelligence MEI 2008/2009 Bruno Paulette.
A Query Mechanism for Inspecting Behavior History Tolga Konik University of Michigan.
1 Episodic Memory for Soar Andrew Nuxoll 15 June 2005.
Chapter 9 DATA WAREHOUSING Transparencies © Pearson Education Limited 1995, 2005.
Focusing Your Evaluation Activities Chapter Four cont…
Case Based Reasoning Melanie Hanson Engr 315. What is Case-Based Reasoning? Storing information from previous experiences Using previously gained knowledge.
Impact of Working Memory Activation on Agent Design John Laird, University of Michigan 1.
1 Soar Semantic Memory Yongjia Wang University of Michigan.
DATA WAREHOUSING.
A Soar’s Eye View of ACT-R John Laird 24 th Soar Workshop June 11, 2004.
Marakas: Decision Support Systems, 2nd Edition © 2003, Prentice-Hall Chapter Chapter 7: Expert Systems and Artificial Intelligence Decision Support.
MICHAEL T. COX UMIACS, UNIVERSITY OF MARYLAND, COLLEGE PARK Toward an Integrated Metacognitive Architecture Cox – 8 July 2011.
Contrasting Examples in Mathematics Lessons Support Flexible and Transferable Knowledge Bethany Rittle-Johnson Vanderbilt University Jon Star Michigan.
Psychology of Music Learning Miksza Memory Processes.
Overview of Long-Term Memory laura leventhal. Reference Chapter 14 Chapter 14.
Soar-RL: Reinforcement Learning and Soar Shelley Nason.
© 2004 Soar Technology, Inc.  July 13, 2015  Slide 1 Thinking… …inside the box Soar Workshop Presentation Presented on 10 June 2004 by Jacob Crossman.
Sepandar Sepehr McMaster University November 2008
Databases & Data Warehouses Chapter 3 Database Processing.
Digital Object: A Virtual Online Storage Solution 598C Course Project Huajing Li.
Copyright R. Weber Machine Learning, Data Mining ISYS370 Dr. R. Weber.
February 1 & 31 Csci 2111: Data and File Structures Week4, Lectures 1 & 2 Fundamental File Structure Concepts & Managing Files of Records.
A Multi-Domain Evaluation of Scaling in Soar’s Episodic Memory Nate Derbinsky, Justin Li, John E. Laird University of Michigan.
Interpreted Declarative Representations of Task Knowledge June 21, 2012 Randolph M. Jones, PhD.
MEMORY & INTELLIGENCE.
Information Processing. History In response to Behaviorism, a cognitive model of mind as computer was adopted (1960’s, 70’s) Humans process, store, encode,
Dr Carolyn Mair Southampton Solent University.  Bit about me  Motivation  Aims  Study  Future work  Feedback please.
SARTRE: System Overview A Case-Based Agent for Two-Player Texas Hold'em Jonathan Rubin & Ian Watson University of Auckland Game AI Group
Integrating Background Knowledge and Reinforcement Learning for Action Selection John E. Laird Nate Derbinsky Miller Tinkerhess.
Inability to retrieve information previously stored in LTM
Current issues in Brazil Objectives: students will discuss a current topic in Brazil. Work cooperatively in small groups. Organize a three-to-five minute.
© 2014 Cengage Learning. All Rights Reserved. May not be scanned, copied or duplicated, or posted to a publicly accessible website, in whole or in part.
File Organization Lecture 1
Future Memory Research in Soar Nate Derbinsky University of Michigan.
Decision Making Chapter 7. Definition of Decision Making Characteristics of decision making: a. Selecting a choice from a number of options b. Some information.
Strategies for Distributed CBR Santi Ontañón IIIA-CSIC.
ACIS Introduction to Data Analytics & Business Intelligence Database s Benefits & Components.
Unit 3 – Neurobiology and Communication
MEMORY & INTELLIGENCE. MEMORY: The input, storage, and retrieval of what has been learned or experienced.
Augmenting (personal) IR Readings Review Evaluation Papers returned & discussed Papers and Projects checkin time.
RULES Patty Nordstrom Hien Nguyen. "Cognitive Skills are Realized by Production Rules"
IMS 4212: Database Implementation 1 Dr. Lawrence West, Management Dept., University of Central Florida Physical Database Implementation—Topics.
Beyond Chunking: Learning in Soar March 22, 2003 John E. Laird Shelley Nason, Andrew Nuxoll and a cast of many others University of Michigan.
Autonomous Mission Management of Unmanned Vehicles using Soar Scott Hanford Penn State Applied Research Lab Distribution A Approved for Public Release;
Understanding Naturally Conveyed Explanations of Device Behavior Michael Oltmans and Randall Davis MIT Artificial Intelligence Lab.
Brunning – Chapter 5 Retrieval Processes. Encoding Specificity Tulving & Osler (1968) –Encoding is enhanced when conditions at retrieval match those present.
Competence-Preserving Retention of Learned Knowledge in Soar’s Working and Procedural Memories Nate Derbinsky, John E. Laird University of Michigan.
3/14/20161 SOAR CIS 479/579 Bruce R. Maxim UM-Dearborn.
Learning from Bare Bones
Modeling Primitive Skill Elements in Soar
Topic 2 – Cognitive Psychology
Topic 2 – Cognitive Psychology
View Integration and Implementation Compromises
Module 11: File Structure
3.3. Case-Based Reasoning (CBR)
Architecture Components
Soar 9.6.0’s Instance-Based Model of Semantic Memory
John Laird, Nate Derbinsky , Jonathan Voigt University of Michigan
Playing with Semantic Memory
Bryan Stearns University of Michigan Soar Workshop - May 2018
Bryan Stearns University of Michigan Soar Workshop - May 2018
Presentation transcript:

1 Episodic Memory for Soar Agents Andrew Nuxoll 11 June 2004

2 What is Episodic Memory? Memories of specific events in our past Memories of specific events in our past  Example: Your last vacation

3 Working Memory Soar-EpMem in Action Cue Retrieved Episodes A simple example… A simple example…

4 Working Memory Soar-EpMem in Action Cue Retrieved Episodes New memories are recorded periodically New memories are recorded periodically

5 Working Memory Soar-EpMem in Action Cue Retrieved Episodes The agent creates a cue The agent creates a cue

6 Working Memory Soar-EpMem in Action The cue is matched to episodic memory The cue is matched to episodic memory Cue Retrieved Episodes

7 Working Memory Soar-EpMem in Action Cue Retrieved Episodes The best match is retrieved into WM The best match is retrieved into WM

8 Benefits of Episodic Memory Aids in decision making through predicting the outcome of possible courses of action Aids in decision making through predicting the outcome of possible courses of action A recorded history can be used to answer questions about the past A recorded history can be used to answer questions about the past To help keep track of progress on long-term goals To help keep track of progress on long-term goals Learn from past experience when new time/resources become available Learn from past experience when new time/resources become available Generalize knowledge by comparing multiple events simultaneously Generalize knowledge by comparing multiple events simultaneously

9 Previous Work Episodic Memory Episodic Memory  Psychology – Endel Tulving  Cognitive Modeling – Erik Altmann Case-Based Reasoning Case-Based Reasoning  Continuous CBR – Ram and Santamaría

10 Key Differences: Episodic Memory Research Episodic Memory Research  Architectural implementation  Domain independent Continuous Case-Based Reasoning Continuous Case-Based Reasoning  Qualitative vs. quantitative episode content  Differing scope of match and retrieval

11 Pilot Implementation: Encoding Encoding initiation: upon significant change in activation levels Encoding initiation: upon significant change in activation levels Episode determination: hand selected and domain specific Episode determination: hand selected and domain specific Feature selection: hand selected and domain specific Feature selection: hand selected and domain specific

12 Pilot Implementation: Storage Episode structure: episodes are stored as Soar productions Episode structure: episodes are stored as Soar productions Episode dynamics: none Episode dynamics: none

13 Pilot Implementation: Retrieval Retrieval initiation: deliberate retrieval in an agent-selected substate Retrieval initiation: deliberate retrieval in an agent-selected substate Cue determination: agent-selected data Cue determination: agent-selected data Retrieval: exact match Retrieval: exact match Retrieved episode representation: direct modification of the agent-selected substate Retrieved episode representation: direct modification of the agent-selected substate Retrieval meta-data: unique sequential id (to provide an idea of temporal order) Retrieval meta-data: unique sequential id (to provide an idea of temporal order)

14 Pilot Implementation: Issues Exact match led to encoding specificity issues Exact match led to encoding specificity issues Problems from overwriting the sub-state Problems from overwriting the sub-state  Recursion  Spurious operator proposals  Requires that agent create a sub-state to do a retrieval Domain dependent Domain dependent

15 Current Implementation: Changes Partial match over a separate episodic memory Partial match over a separate episodic memory  Memories are no longer stored as rules Use of an architecture-specified buffer for query and retrieval (analogous to the ^io link) Use of an architecture-specified buffer for query and retrieval (analogous to the ^io link)

16 Current Implementation: Encoding Encoding initiation: one episode per agent action Encoding initiation: one episode per agent action Episode determination: all of working memory(!) Episode determination: all of working memory(!) Feature selection: the entire episode can affect retrieval Feature selection: the entire episode can affect retrieval

17 Current Implementation: Storage Episode structure: episodes are stored in an internal data structure Episode structure: episodes are stored in an internal data structure Episode dynamics: still none Episode dynamics: still none

18 Current Implementation: Retrieval Retrieval initiation: cue is constructed in an architecture-specified buffer Retrieval initiation: cue is constructed in an architecture-specified buffer Cue determination: agent selected data Cue determination: agent selected data Retrieval: exact match Retrieval: exact match Retrieved episode representation: the episode is recreated in an architecture-specified buffer Retrieved episode representation: the episode is recreated in an architecture-specified buffer Retrieval meta-data: agent can retrieve the next memory in temporal sequence Retrieval meta-data: agent can retrieve the next memory in temporal sequence

19 Working Memory Activation Extension of the memory decay work by Ron Chong Extension of the memory decay work by Ron Chong Reimplementation by Michael James: Reimplementation by Michael James:  Includes all of working memory  Improvements in efficiency

20 Activation & Matching Problem: All WMEs in an episode are weighted equally Problem: All WMEs in an episode are weighted equally Core Idea: The activation level of WMEs indicates their relevance to current task Core Idea: The activation level of WMEs indicates their relevance to current task Implementation: Use the activation levels of the WMEs in the episode to bias the match Implementation: Use the activation levels of the WMEs in the episode to bias the match

21 Evaluation using Eaters Pac-Man-like Pac-Man-like Two types of food Two types of food  Bonus food (10 pts)  Normal food (5 pts)

22 Create a memory cue (input-link + proposed direction) Create a memory cue (input-link + proposed direction) East South North Evaluate moving in each available direction Evaluate moving in each available direction An Episodic Memory Eater Episodic Retrieval Retrieve the best matching memory Retrieve the best matching memory Retrieve Next Memory Retrieve the next memory (in temporal order) Retrieve the next memory (in temporal order) Use the change in score to evaluate the proposed action Use the change in score to evaluate the proposed action Move North = 10 points

23 Initial Results

24 Problem #1: I-Support Masking Problem: Testing an i-supported WME provides no activation boost Problem: Testing an i-supported WME provides no activation boost Solution = Pay it Backward: Testing an i-supported WMEs boosts the activation level of its “set of o-support” Solution = Pay it Backward: Testing an i-supported WMEs boosts the activation level of its “set of o-support”

25 Problem #2: New WME Masking Problem: A new WME starts at a fixed activation level Problem: A new WME starts at a fixed activation level Solution = Pay it Forward: Activation of newly created WMEs is based upon those WMEs which were tested to create it Solution = Pay it Forward: Activation of newly created WMEs is based upon those WMEs which were tested to create it

26 Results

27 Current Challenge: Performance

28 Nuggets Coal Domain independent, architectural implementation Domain independent, architectural implementation Performance issues Still only tested in a single domain