Webpage of Amiram )Ami) Moshaiov
(Jan2020)
_files/image003.gif)
_files/image005.jpg)
_files/image007.gif)
_files/image009.gif)
_files/image011.gif)
Amiram (Ami) Moshaiov
1.
School of Mechanical Engineering, Iby and Aladar Fleischman
Faculty of Engineering
2.
Sagol School of Neuroscience
Tel-Aviv University
Ramat-Aviv, Tel-Aviv 69978, Israel
e-mail: moshaiov@tauex.tau.ac.il , Tel. (work): 972-3-6407098
Interactive Concept-based Search HHAI2024 Tutorial Amiram (Ami) Moshaiov, Tel-Aviv University
I.
Our Computational Intelligence Research Group
A.
What is
Computational Intelligence (CI)?
CI
deals with bio-inspired computational methods. Its roots are in Cybernetics and
in Soft-computing. See:
(http://www.asc-cybernetics.org/foundations/definitions.htm ) (https://en.wikipedia.org/wiki/Soft_computing).
While soft computing has focused on
Artificial Neural Networks (ANNs), Fuzzy Logic (FL), and Evolutionary
Computation (EC), CI has a much wider scope. In addition to the above three
building blocks, CI deals with other bio-inspired methods such as Swarm Intelligence
and Artificial Immune Systems. CI encompasses not only such bio-inspired computational
techniques, but also their hybridizations.
Although
developed somewhat separately from Artificial Intelligence (AI), CI may be
considered as a branch of AI.
CI methods are currently in use in
many technical systems and are consistently explored by researchers all over
the world. Such activities are in the focus of the IEEE Society of
Computational Intelligence.
CI
applicability to robotics and engineering design makes it a valid topic for
teaching mechanical engineering students. Such a course is provided by our
school for both undergraduate and graduate students.
B.
Overview of
our Research
Our research group focuses on Evolutionary
Computation (EC) methods, and in particular on Evolutionary Multi-objective
Optimization (EMO). We also deal with hybridizations of ANNs and FL with EC.
For example, we deal with Neuroevolution, which is also known as Evolutionary
Neural-Networks.
We develop generic methods and
algorithms with a focus on multi-objective problems. Such problems are common
in engineering and other application areas (See https://en.wikipedia.org/wiki/Multi-objective_optimization ). Clearly, in the design of technical systems one must
face with tradeoffs that are imposed by the existence of conflicting design objectives.
Evolutionary algorithms are known to be most useful for finding the
Pareto-optimal solutions to such problems. Such a solution approach, to
multi-objective problems, reveals the performance tradeoffs, which supports a
rational selection of a solution.
We have extended single and
multi-objective optimization to Multi-Concept Optimization, which has a wide
scope of applications as explained in Subsection C.1 below.
In 2014 we have extended our
interest in multi-objective optimization to the related problem of non-cooperative
multi-objective games, which are also known as multi-payoff Games (see Subsection
C.2 below). Other related problems that we have dealt with are described
in Subsections C.3-C.5) below. Related publications can be found
in Sections II.A-II.D, below.
C.
Uniqueness
of our Research
Note:
See the “List of Recent Articles” in
Section II below for exemplary articles on the following topics
1.
Multi-Concept Search and Optimization
Traditional optimization aims at
finding optimal solutions. In contrast, the original aim of multi-concept search
and optimization has been to support the selection of conceptual solutions. Its
development has been strongly connected with Pareto-based multi-objective
optimization and multi-criteria decision-making.
For over a decade we have been
leading the development of multi-concept optimization. We have uniquely
developed a relaxed-Pareto version of the concept-based approach. It was
motivated by the need to overcome the logical flaw that we have found in the
original s-Pareto version of the concept-based approach.
Our relaxed-Pareto version of the concept-based
approach can be used not only for concept selection, but also for the exploration
of design spaces, both at the conceptual and particular design levels, and as
an alternative to multi-modal optimization.
In collaboration with Dr. Ohad Gur
of the Israel Aerospace Industries, we have demonstrated the practicality of
our concept-based approach to the industry.
Since 2018, we have presented a Tutorial on Multi-concept Optimization at international conferences including: IEEE-CEC 2018 and IEEE-SMC 2019. Next presentation is expected at GECCO-2020. See:
( https://gecco-2020.sigevo.org/index.html/Tutorials#id_Multi-concept%20Optimization ).
2. Non-cooperative Multi-objective Games
In collaboration with Dr Gideon Avigad, a former PhD student
of our group, and others, we have developed a unique non-utility approach to
multi-objective games, which are also known as multi-payoff games. At present
we are investigating the capability of our non-utility approach to support
decision-makers in various application areas such as defense (e.g., cyber
security), operation research, economics and biology.
3. Neuro-fuzzy Learning of the Behavior
of Biological Systems
In collaboration with Prof. Amir Ayali, we have developed a
neuro-fuzzy approach that can uniquely support the understanding of biological
systems.
4.
Family Bootstrapping by Genetic Transfer Learning
Commonly,
evolutionary computation process is initiated with a random population of
solutions. Yet, for complex problems, such an approach may fail to get started
without taking proper measures. This phenomenon is known as the bootstrap
problem. To overcome the problem, we have suggested a genetic transfer learning
approach, which we have termed family bootstrapping.
While other methods exist to cope
with the bootstrap problem, our unique family bootstrapping approach may
facilitate revolutionizing the evolutionary design of robots. This is due to
the use of relations among tasks. Family bootstrapping may lead to the creation
of databases of useful ancestries, which will eliminate, or at least reduce,
the need to initiate evolution processes from random.
5.
Multi/Many-objective Neuroevolution
We have been one of the first groups
to consider the application of evolutionary multi-objective optimization for
neuroevolution, and in particular for robot neuro-controllers. With this respect, we have been the
first to propose an algorithm for multi-objective evolution of CPNs, i.e., counterpropagation
networks. We
have also developed a unique MO-TWEANN, i.e., Multi-Objective version of the
famous NEAT algorithm, for Topology and Weight Evolving Artificial Neural
Network, also known as Neural Architecture Search (NAS). Recently we have
developed a unique many-objective TWEANN algorithm to cope with an increasing
number of objectives (4 and beyond).
6.
Multi-Competence Cybernetics
We have been the first to postulate
on the significance of Pareto-optimality to the study of biological systems (See:
https://link.springer.com/chapter/10.1007/978-3-540-72964-8_14 ). A few years passed and in 2012 researchers from the
Weizmann Institute of Science published a paper in Science confirming our suggestion
J. (See Shoval et al., Science 336, 1157, 2012).
D. List of Application Areas (for which we are
currently applying our CI tools)
Engineering
Design, Robotics, Control, Aeronautics, Cyber Security, Communication, Agriculture,
Computer Vision and Classification
Operation Research
and Management, Economy
Ecology, Zoology
E. Current International and National Collaboration
1. AI to the Rescue: Life-and-Death
Decision-Making under Conflicting Criteria – Funded by the Volkswagen
Foundation.
In
collaboration with S. Mustaghim, Otto von Guericke University, Germany and B. Adini,
Faculty of Medicine at Tel-Aviv University.
2. Drought-related Functional Tradeoffs
of Trees – In collaboration with the Int. Institute for Applied System
Analysis (IIASA), Austria.
3. Multi-concept Optimization of
Robotic Manipulators – In collaboration with the Agricultural Research
Organization, Volcani Center, Israel.
4. Non-cooperative Multi-payoff Games
in Economics – In collaboration with the Economic Department, LUISS
University, Italy.
F. Other Research Projects/Research Topics of Interest
1. Adversarial Robotics under
Conflicting Objectives – Funded by the Israeli Ministry of Science.
2. Communication Network Design by
Multi-Concept Optimization under Conflicting Objectives – Funded by the
Israeli Defense Ministry.
3. Deep ANNs and Ensembles of ANNs by
Many-objective Evolution of the Networks'
Architecture and Weights.
4. Multi-objective Genetic Transfer for
Optimization.
5. Metaheuristics and Adaptive Parameterization
for Multi-objective Optimization and Multi-concept Optimization.
6. Model Learning for Multi-concept
Optimization.
7. Machine Learning with Many-objective
Optimization as Applied to Big Data.
8. Revisiting Multi-concept
Optimization and Multi-objective Games for the understanding of Biological
Systems
G. For Prospective MSc/PhD Students & Post-docs
If
you are interested in topics such as: Systems (technical or biological),
Engineering Design, Robotics, Mechtronics, Control, Artificial Intelligence,
Bio-inspired Computational Methods such as Evolutionary Computations, Neural
Networks, and Fuzzy Logic, then you may want to consider joining our research
group.
In
such a case, please read the above description on the uniqueness of our
research, and see if it fits your interests.
H. List of Current Students
Meir
Harel (PhD)
Eliran
Farhi (PhD)
Adham
Salih (PhD)
Ofer
Weiss (PhD)
Yoni
Maor (PhD)
Joseph
Gabbay (PhD)
Stav
Bar-Sheshet (PhD)
Tamir
Mhabari (MSc)
Yossef
Braslev (MSc)
Yuval
Alon (BSc)
I. List of Recent Graduates
Erella
Matalon-Eisenstadt (PhD, 2019)
David
Alkaher (PhD, 2019)
Barak
Samina (MSc, 2019)
Roi
Chananel (MSc, 2019)
Adham
Salih (MSc, 2018)
Gil
Segal (MSc, 2017)
Omer
Abramovich (MSc, 2017)
II.
List of Recent Articles
Note: For copies of reports/papers, which
are not linked here, please contact moshaiov@tauex.tau.ac.il
A.
Articles on Multi-objective
Optimization and Multi-Concept Optimization
1.
Farhi, E. and Moshaiov, A. Window-of-Interest-based Multi-objective Evolutionary
Search for Satisficing Concepts
Proceedings
of the IEEE Conference on Systems, Man and Cybernetics, 2017.
2.
Moshaiov, A. The Paradox of Multimodal
Optimization: Concepts vs. Species in Single and Multi-objective
Problems
Proceedings
of the IEEE Congress on Evolutionary Computation, 2016.
3.
Snir, A., Samina, B. and Moshaiov, A. Concept-based Evolutionary
Multi-Criteria Exploration of Design Spaces under Run-time Limitation
Proceedings
of the IEEE Symposium Series on Computational Intelligence, 2015.
4.
Moshaiov, A., Snir, A. and Samina, B. Concept-based Evolutionary
Exploration of Design Spaces by a Resolution-Relaxation-Pareto Approach,
Proceedings
of the IEEE Congress on Evolutionary Computation, 2015.
B.
Articles on Multi-objective
Neuroevolution
5.
Salih, A. and Moshaiov, A. Many-objective Topology and Weight Evolution of
Neural-Networks,
Submitted
to the International Joint Conference on Neural Networks, 2020.
6.
Abramovich, O. and Moshaiov, A. Multi-objective Topology and Weight
Evolution of Neuro-controllers,
Proceedings
of the IEEE Congress on Evolutionary Computation, 2016.
7.
Salih, A. and Moshaiov, A. Multi-objective Neuroevolution:
Should the Main Reproduction Mechanism be Crossover or Mutation?
Proceedings of the IEEE Conference on Systems, Man and
Cybernetics,
2016.
8.
Moshaiov,
A. and Tal, A. Family Bootstrapping: a Genetic Transfer Learning Approach
for Onsetting the Evolution for a Set of Related Robotic Tasks. Proceedings
of the IEEE Congress on Evolutionary Computation, 2014.
9.
Moshaiov,
A. and Abramovich, O. Is MO-CMA-ES Superior to NSGA-II
for the Evolution of Multi-objective Neuro-controllers?
Proceedings of the IEEE Congress on Evolutionary Computation, 2014.
C.
Articles on Fuzzy and
Neuro-fuzzy Control
10. Gabbay, J.,
and Moshaiov, A.
Human-like Motion Control for Autonomous Driving in Rugged Terrains,
To
be submitted to the IEEE International Conference on Fuzzy Systems,
2020.
11.
Segal, G., Moshaiov, A., Amichay G. and Ayali, A. Neuro-fuzzy Learning of Locust's
Marching in a Swarm.
Proceedings
of the International Joint Conference on Neural Networks, 2016.
D. Articles
on Single and Multi-objective Games
12. Eisenstadt,
E. and Moshaiov, A.
Co-Evolutionary Algorithm for Solving Multi-Objective Games,
To
be submitted to Swarm and Evolutionary Computation
13. Eisenstadt,
E. and Moshaiov, A. Mutual
Rationalizability in Vector-payoff Games,
Proceedings
of the International Conference on Evolutionary Multi-Criterion Optimization, 2019.
14. Harel, M.,
Moshaiov, A. and Alkaher, D. Rationalizable Strategies for the Navigator-Target-Missile
Game,
Accepted
for publication in the AIAA Journal of Guidance, Control, and Dynamics,
2019.
15. Eisenstadt,
E. and Moshaiov, A.
Decision‐making in Non‐cooperative Games with Conflicting Self‐objectives,
Journal
of Multi‐Criteria Decision Analysis, 2018.
16. Alkaher, D.
and Moshaiov, A.
Non-dominated Strategies for Cautious to Courageous Aerial Navigation,
AIAA
Journal of Guidance, Control, and Dynamics, 2018.
17. Eisenstadt,
E. and Moshaiov, A.
Novel Solution Approach for Multi-objective Attack-Defense Cyber Games with
Unknown Utilities of the Opponent, IEEE Transactions on Emerging Topics in
Computational Intelligence, 2017.
18. Harel, M.,
Eisenstadt, E. and Moshaiov, A. Solving Multi-objective Games using A-priori Auxiliary
Criteria,
Proceedings
of the IEEE Congress on Evolutionary Computation, 2017.
19. Eisenstadt,
E., Moshaiov, A. and Avigad G. The Competing Travelling Salespersons Problem under
Multi-criteria,
Proceedings of the International Conference on Parallel
Problem Solving from Nature, 2016.
20.
Alkaher,
D. and Moshaiov, A. Game-based Safe Aircraft Navigation
in the Presence of Energy-Bleeding Coasting-Missile.
AIAA Journal
of Guidance, Control, and Dynamics,
2016.
21. Alkaher, D.
and Moshaiov, A.
Dynamic-Escape-Zone to Avoid Energy-Bleeding Coasting-Missile,”
AIAA
Journal of Guidance, Control, and Dynamics, 2015.
22. Eisenstadt,
E., Moshaiov, A. and Avigad, G. Co-evolution of Strategies for Multi-objective Games under
Postponed Objective Preferences,
Proceedings of the IEEE Conference on Computational
Intelligence and Games (CIG), pp. 461–468, 2015.
21.
Alkaher, D., Moshaiov, A., and Or, Y. Guidance Laws based on Optimal
Feedback Linearization Pseudo-Control with Time-to-go Estimation.
AIAA Journal of Guidance, Control and Dynamics, 2014.
III.
My
Professional Experience
A. Academic Experience
During
1985-88, Amiram Moshaiov was an Assistant Professor at MIT, USA.
Since
1988 A. Moshaiov has been at Tel-Aviv University (TAU).
He
is the founder and developer of the robotics and mechatronics section of
the Mech. Eng. Program at TAU.
B. Engineering Experience
Moshaiov
served for 5 years as a mechanical engineer in the Israeli Navy.
While
leading the robotics and mechatronics labs of the Mechanical Engineering Program
at TAU, during the 90's, he was involved in the design of robots and their
subsystems.
C. Activities in Editorial Boards, International Organizations
and Conferences
Associate
Editor of the IEEE Transaction on Emerging Topics in Computational Intelligence
Member
of the Editorial Board of the Journal of Memetic Computing and a reviewer of
many other scientific journals
Member
of the IEEE Task Force on Artificial Life and Complex Adaptive Systems
Member
of the IEEE Task Force on Evolutionary Deep Learning and Applications
Member
of the EURO Working Group on Multicriteria Decision Aiding
Ex-member
of the Management Board of the European Network of Excellence in Robotics
Moshaiov is/was a member of many
international program committees of conferences such as:
The
IEEE International Conference on Systems, Man, and Cybernetics
The
IEEE/RSJ International Conference on Intelligent Robots and Systems
The
IEEE Congress on Evolutionary Computation
The
IEEE World Congress on Computational Intelligence
The
IEEE Symposium on Artificial Life
The
IEEE Symposium on Comp. Intelligence for Security and Defense Applications
The
IEEE Symposium on Comp. Intelligence in Multicriteria Decision Making
The
International Conference on Parallel Problem Solving from Nature
The
International Conference on Simulated Evolution and Learning
The
European Robotic Symposium
The
International IFAC Symposium on Robot Control
The
International Symposium on Tools and Methods of Competitive Engineering
The
International Conference on Engineering Design
The
International Conference on Mechatronics
The
IEEE International Conference on Control Applications
The
IEEE International Conference on Computational Cybernetics
D. Organization of Workshops and Special Sessions
at International Conferences
Originator and Co-Chair of:
The
IEEE/RSJ IROS Workshop on Multi-Objective Robotics
The
IEEE/RSJ IROS Workshop on Multi-Competence Optimization and Adaptation in
Robotics and A-life
The
GECCO Workshop on the Evolution of Natural and Artificial Systems
The
Special Session on EC and Games at IEEE-CEC 2017
IV.
My Current Research Interests
A. Computational Intelligence Methods
Evolutionary
Computation (e.g., Genetic Algorithms, Evolution Strategies)
Neural
Networks (e.g., deep neural-networks)
Fuzzy
Logic
Hybrid
Methods (e.g., Evolutionary Neural Networks, Neuro-Fuzzy Inference Systems)
Interactive
Evolutionary Computation
Evolutionary
Multi-Objective Optimization
Hybrids
of Evolutionary Multi-Objective Optimization and Multi-criteria Decision-making
methods
Evolutionary
Multi-Objective Adaptation
Co-Evolution
Genetic
Transfer
Learning,
Adaptation, Development
B. Other Techniques
Pontryagin
Maximum Principle
Differential
Games
Feedback
Linearization
C. Application Areas and Problems
Robotics
(e.g., Behavioral, Cognitive, Developmental, Evolutionary)
Guidance
and Control
Mechatronics
(e.g., Simultaneous evolution of mechanics and control)
Complex
Adaptive Systems
Cybernetics
(e.g., Metaphors and analogies in natural and artificial Evolution,
Multi-competence cybernetics)
Artificial
Life (e.g., Bio-plausible simulations, System identification of the behavior of
animals)
Computer
Vision (e.g., Object detection and recognition, Bio-inspired vision systems, Attention)
Multi-Objective
Games (known also as multi-payoff games or vector payoff games)
Aeronautical
Games (e.g., Air combat problems, Pursuit-evasion)
Cyber
Security Games
Games
in Operation Research (e.g., Competitive travelling salespersons problems)
Engineering
Design (e.g., Conceptual, Multi-objective, Robust, Interactive, Design space
exploration)
Multi-objective
Planning
Multi-objective
Navigation
Machine
Learning (e.g., Supervised and unsupervised learning, Reinforcement learning, Deep
learning, Transfer learning)
Text
Analysis
Multi-Agent
Systems
Artistic
Design by Evolutionary Computation
V.
My Past Research Topics
A.
Moshaiov published during the 80’s and early 90’s in international scientific
journals on many topics in structural mechanics, and shipbuilding manufacturing
processes. Later he conducted research on topics such as MEMS manufacturing
processes, bio-mechanics of tennis elbow, and robot kinematics. For his current
research topics, see above.