One of successful new programming methods in robotics is done in two steps: (1) creating a blank robot with some useful manipulators and (2) teaching the robot a given skill or process by -- figuratively -- walking it through the steps. The method has been successfully applied to dull routines like packing, carrying and/or sorting stuff, etc. Note that the method is attractive mostly because the robot can always be reset and re-taught another skill. The key feature is that humans are an integral part of the technology. In this paper, the same fundamental approach is applied to management of large sets of unstructured content where raw information is provided by folksonomies and structure is added gradually or on-demand via the proposed ReBot which requires human participation to function efficiently.
3. .
.
Social Robotics is ...
• ... when
robots replace people in social networks
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
3 /27
3/27
4. .
.
.... that was a JOKE!
Marat Zhanikeev -- maratishe@gmail.com --
(in case you have not figured it out)
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
4 /27
4/27
5. .
.
(the real) Social Robotics
02 E.Berger+4 "Inferring Guidance Information in Cooperative Human-Robot Tasks" HUMANOIDS (2013)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
5 /27
5/27
6. .
.
Social Robotics in a Nutshell
.
The Problem...
.
...
. is that
search space in a robot's decision making process is too big
.
... The Solution...
.
... is to
search space via
robots and humans
.
reduce
social interactions between
• a fundamentally distinct approach to robotics 02 04 05
• social robotics is a new area 03 07 , but tradition is still strong 06
• this paper: the same basic idea applied to software automation
02 E.Berger+4 "Inferring Guidance Information in Cooperative Human-Robot Tasks" HUMANOIDS (2013)
04 A.Thomas+1 "Reinforcement Learning with Human Teachers..." 21st AAAI (2006)
05 H.Suay+1 "Effect of Human Guidance and State Space Size ...." IEEE RO-MAN (2011)
03 W.Knox+2 "Training a Robot via Human Feedback: A Case Study" Social Robotics (2013)
07 B.Mutlu+4 "Social Robotics" Springer LNAI vol.7072 (2011)
06 P.Corke+0 "Robotics, Vision and Control" Springer (2011)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
6 /27
6/27
8. .
.
Software Robots: the Missing Topic
01
"Software Robots: The Long Tail of Automation" http://www.wired.com (2013)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
8 /27
8/27
9. .
.
Robots and ReBots as Complexity
•
•
software robots (rebots) are all about knowledge management
note: the same basic approach can be applied both to hardware and software
Search
the space
Robot
Tell
what
to do
A very
complex
system
Marat Zhanikeev -- maratishe@gmail.com --
Human
What
should
I do?
Robot
Search
the space
A less
complex
system
Human
Guide through
feedback
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
9 /27
9/27
10. .
.
The Hardware and Software Stories
1.
hardware: we always know exactly what the job is
◦
2.
software: we know the data, but we do not know what knowledge it contains
hardware: training is difficult because robots make decisions
(humans are guides)
◦ software: training it easy since humans make decisions
◦ ... but robots have to visualize the outcome properly
3.
hardware: long-term jobs are rare -- tasks are normally specific and now
◦ software: long-term, even continuous, is the intended mode of operation
◦ hence the ReBot -- recommendation bot is possible
◦ recommendations can take time to be generated
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
10 /27
10/27
11. .
.
Similarities and Differences
•
ReBot is centered around a classification problem 08
• ReBots do not have or need eyes
• ReBots are dumb,
Social
Robotics
The Rebot
(proposal)
humans make decision
Generic
Use
Wide range of
behavior
Any kind of
context
Teaching,
Guidance
Reinforced
Learning
Bayesian
Classification
Reasoning
YES. Vision,
recognition
NO
Not needed
Human
Role
Guide only
Guide and
decision maker
08 C.Manning+3 "Introduction to Information Retrieval" Cambridge University Press (2008)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
11 /27
11/27
13. .
.
ReBot versus Tradition
• in a way,
•
ReBot is an ideal context manager -- compared to existing tools 09
note:
multiple items is when more than one item are linked by some kind of
structure -- ReBots use metromaps 11
Traditional
Search
Facetted
Search
Advanced
Tools / Others
Rebot
(the proposal)
Multiple
Items?
Visual?
Underlying
Structure?
Learning?
NO
NO
NO
NO
NO
NO
NO
NO
YES
YES
YES
YES
YES
YES
YES
YES
09 M.Mas+0 "Faceted semantic search for personalized social search" Cornell University (2011)
11 K.Nesbitt+0 "Getting to more abstract places using the metro map metaphor" IV Conf. (2004)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
13 /27
13/27
15. .
.
ReBot: Where is The Originality?
1.
folksonomy-style input is OK
2.
(1) many contributors, (2) no rules, (3) careless input 10
metromaps as the best visual interface 12
◦ traditional research only uses graphs,, metromaps are very very different
◦ folksonomy means
◦ new layout engines are proposed -- out of scope here 13
3.
recommendation engine based on classification -- hence the ReBot
◦ classes are based on trains and stations in metromaps
4.
social scalability 10
◦ allowing for large-scale social collaboration on publically shared metromaps
10 myself+0 "MetroMaps versus Facets: What Exactly is the Ontological Context?" IEICE AI研究会 (November 2013)
12 myself+0 "...the Most Cost Effective Method in Browser-Based Visualizations..." IIEEJ Visual Computing Workshop (2013)
13 myself "Rebot Project Page" http://www.github.com/maratishe/rebot (current)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
15 /27
15/27
17. .
.
ReBot: Simple Implementation
• the simple design is:
1.
learning during input via recommendation and feedback
on input
•
, ReBot recommends which trains to assign your paper to
• you provide feedback by (possibly) picking an unexpected order
2.
classification design which would make sure that ReBot gets
better with time
• Naive Bayes classifier is used
• but the
-- one item can belong to multiple classes
•
shows that the engine gets better with time
3.
design is unusual
next page
build a great visual interface based on metromaps and rings
• metromaps: the traditional concept of metromaps
• rings: simpler 3D visual aggregates of items
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
17 /27
17/27
18. .
.
ReBot Part 1: Folksonomical Input
•
(1) read the item, (2) select tags from the order recommended by ReBot, (3) let
ReBot learn from your selection -- not all assignments need learning
1
3
3
2
2
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
18 /27
18/27
19. .
.
ReBot Part 2: Recommendation
Rebot Order of User Selection
20
16
12
8
4
0
0
5.2
10.4
15.6
20.8
26
Time/Learning Sequence
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
19 /27
19/27
20. .
.
ReBot Part 3: Visual Interface
•
•
note: layouts are non-traditional 13 -- more suited to browser viewing
note 2: rings are simplified metromaps -- like Tokyo city with its Yamanote ring
13 myself "Rebot Project Page" http://www.github.com/maratishe/rebot (current)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
20 /27
20/27
21. .
.
ReBot.Papers: A Working Example
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
21 /27
21/27
22. .
.
ReBot.Papers: The Objective
manage my papers
• a ReBot which would
◦ in submission, rejected, passed, ...
powerful visual browsing function -- metromaps
• ReBot would recommend new stuff
• a very
◦ similar papers in the Internet, references, etc.
◦ recommendation to add more papers to the same train
◦ recommendations for a more optimal metromaps for my papers
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
22 /27
22/27
23. .
.
ReBot.Papers: Smarter Interface
• sorry, will omit small details ... but will show a
demo!
Rebot
Human
Surface
Depth
Browse
Selectors
Write
Metromap
(trains)
(groupings)
The Raw Metromap
Knowledge Base
Marat Zhanikeev -- maratishe@gmail.com --
Create
(papers)
Maintain
(the data itself)
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
23 /27
23/27
25. .
.
ReBot in the Cloud
•
the screenshot:
ReBot implemented as
Chrome Extension
cloud storage there is no need for web servers -- APIs are
sufficient
makes large-scale collaboration possible -- same for recommendations, etc.
• with
•
Cloud
Storage
Knowledge
(in the wild)
Places/members
Crawl,
recommend,
maintain
Marat Zhanikeev -- maratishe@gmail.com --
{structure}
(shared access)
Visual
interface
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
25 /27
25/27
26. .
.
Wrapup
•
software robots are proposed as recommendation bots
social robotics
simple design and a more complex working example are presented
the socially scalable version is on its way
◦ share the basic idea with
•
•
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
26 /27
26/27
27. .
.
That’s all, thank you ...
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
27 /27
27/27
28. .
.
[01]
(2013)
Software Robots: The Long Tail of Automation
http://www.wired.com
[02] E.Berger+4 (2013)
Inferring Guidance Information in Cooperative Human-Robot Tasks
HUMANOIDS
[03] W.Knox+2 (2013)
Training a Robot via Human Feedback: A Case Study
Social Robotics
[04] A.Thomas+1 (2006)
Reinforcement Learning with Human Teachers...
21st AAAI
[05] H.Suay+1 (2011)
Effect of Human Guidance and State Space Size ....
IEEE RO-MAN
[06] P.Corke+0 (2011)
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
27 /27
27/27
29. .
.
Robotics, Vision and Control
Springer
[07] B.Mutlu+4 (2011)
Social Robotics
Springer LNAI vol.7072
[08] C.Manning+3 (2008)
Introduction to Information Retrieval
Cambridge University Press
[09] M.Mas+0 (2011)
Faceted semantic search for personalized social search
Cornell University
[10] myself+0 (November 2013)
MetroMaps versus Facets: What Exactly is the Ontological Context?
IEICE AI研究会
[11] K.Nesbitt+0 (2004)
Getting to more abstract places using the metro map metaphor
Marat Zhanikeev -- maratishe@gmail.com --
ReBoT: a Recommendation-Based Software Robot -- http://tinyurl.com/marat140213 --
27 /27
27/27