| |
Workshop on Development
and Learning
Topics
Each session has a list of topics that the talks and discussion
in that session should address. The topic questions following each topic are
suggestive and are not meant to be exclusive. The following is a list of
sessions and the list of assigned topics.
Session 1: Child Mental Development. Topics: 1.b, 1.c, 1.e,
2.d, 2.e
Session 2: Networks. Topics: 1.a, 1.b, 1.d, 2.b, 2.c, 2.d
Session 3: Neuroscience: Topics: 1.d,
2.a, 2.b, 2.c, 2.d, 2.e
Session 4: Computation: Topics:
1.a, 1.b, 1.e, 2.a, 2.b, 2.d
Session 5: Future Directions and Applications: Topics: 3.a, 3.b, 3.c,
3.d, 3.e, 4.a, 4.c
Session 6: How to Proceed? Topics: 4.a, 4.b, 4.c, 4.d, 4.e
The above topic numbers correspond to the following index
numbers:
The role of development and learning in human intelligence and
the role they can play in making intelligent machines.
- Relationships between development and learning.
Does automatic development require a paradigm change from the current machine
learning (or engineering) paradigm? How do children develop to learn things
their parents do not understand? Can a human make a machine to learn to understand a
language that the human maker does not understand?
What is development and what is learning? What are the fundamental differences
between human learning and current machine learning techniques?
Assuming that development includes learning, what are the major capabilities that a human
developmental system has but a currently typical machine learning system does not?
- Mechanisms that enable environment/learner interactions in development
and learning.
What mechanisms/processes enable interaction between given (changing) learner architecture
and the external world to which learner must adapt and through which the learner develops
new skills?
How does a human child gradually (through development) make sense out of what is sensed
from the environment and what he acts upon in the environment?
What is the role of the somatosensory and body senses in perceptual/cognitive
organization?
- Self-regulation or autonomous systems of development and learning.
How does a learner learns from its own internally generated inputs (e.g., through
proprioceptive sensors) and from its own exploration of the environment and not just
passively from externally driven input?
How does the process of development gradually enable such an active learning capability?
How do internal regulation (learning automation)and external regulation (environmental
feedback)interact?
- Representation issues in development.
Does a
human developmental algorithm enables automatic derivation of representation for all the
tasks to be learned in the life time?
What are the possible mechanisms that enable a developmental system to autonomously build
representation for ever changing,hardly predictable tasks that an individual (natural or
artificial)has to tackle over his/her/its lifespan?
- Scaling-up of capabilities, i.e., development as the scaling up of
capabilities from ground, with genetically coded or human programmed bias.
What are the computational models of development, for a subpart of human/agent
or for an overall system?
If human development is controlled by a human developmental algorithm that is inherited
from the parents' genes, what are the possible mechanisms that enable autonomous and
automated development?
How does a developmental system scale up its cognitive and behavioral capabilities?
Common developmental principles that are shared by very diverse
cognitive and behavioral capabilities such as vision, speech, language, understanding,
reasoning, planning, decision making, navigation, object manipulation and other motor
actions.
- Commonality among different domains of capabilities.
Traditionally, capabilities developed in various performance domains were considered very
different and thus have been studied separately with different representations and
methodologies. What are the benefits and what are the drawbacks of this paradigm of study?
If the unit of representation goes down to millisecond (or finer) level of sensory signal
stream, neural signals in the brain and muscle contraction signal stream, is there a
computational view that characterizes all these capabilities?
What are the state-of-the-art results from studies about plasticity of human brain, such
as those through varying extent of sensory input, redirecting input, transplanting cortex,
etc?
What do the results from those studies suggest?
- Illustrations of common principles that cross domains in development and
learning.
What are the common principles shared by visual language (e.g., American Sign Language)
understanding and spoken language understanding?
What are the common principles shared by speech production and building a Lego toy?
What kind of granularity of representation will allow cross-modality sensor-integration
and effector-cooperation?
What neural mechanisms enable an organism to distinguish between aversive and
appetitive stimulus?
- Constraints on development and learning for different cognitive and
behavioral capabilities.
Is there evidence for or modeling of a constrained modularity or constrained nonmodularity
approach?
What constraints on neurobiological mechanisms of learning and developing are posed by our
understanding about cross-species genetics and behavioral performances?
What constraints on neurobiological mechanisms of human learning are posed by our
understanding of cross-cultural similarities in performances and knowledge?
What constraints on neurobiological mechanisms of human learning are posed by unusual
circumstances of development including precocious and impaired development?
- The role of context in learning and developing.
How does an infant cope with bombardment of continuous streams of sensory inputs while
learning?
How can the related context be formed through learning?
How does the general concept of "chunking" apply to development and learning
(human or machine)?
What are the developmental mechanisms that enable scaling up not only individual
capabilities but also their integration, from continuous, real-time, multimodal sensory
input streams and effector action streams?
- The role of attention in development and learning.
What is our knowledge about human attention selection?
What roles does attention play in context formation?
What are the mechanisms that enable a learner to develop capabilities of attention
selection, including intramodal selection (e.g., selecting a part of sensory signal in
sensed visual images) and intermodal selection (e.g., selecting the current visual input
but not the current auditory input)?
Important directions for future research on development and
learning.
- Terminology discussion.
If we assume that human has a developmental algorithm that starts to run at the conception
time of each new human life, what does the human developmental algorithm do?
How do we name the kind of machines that can automatically develop their cognitive and
behavioral capabilities? How do we define the term "automatically" here
with consideration of the human developmental process?
How do we define a developmental algorithm?
What basic functions should a developmental algorithm have?
- New and open research topics.
What research topics are raised by our discussion about the development and learning for
machines as well as about mechanisms of human development and learning?
In what ways the future studies of cognitive and behavioral development are likely to
address very challenging capabilities including vision, speech recognition, language
understanding, reasoning, planning, decision making, speech production, navigation, object
manipulation. etc?
- Milestones problems.
What milestone problems can we lay out for future research in the development and learning
for machines and in understanding of human development and learning?
Can we quantize those milestone problems in certain precise way so that we can measure the
"generation" or human-equivalent "age" of future machines that can
develop?
- Short-term and long-term tasks.
Why do some tasks that are considered "easy" for humans turn out to be very hard
for machines and vice versa?
With regard to development, is it true that what is hard for human infants is also often
hard for machines?
From our knowledge about what is hard and what is easy for infants, can we suggest easy
and hard tasks for machines in the process of automated machine development?
In the context of development, what tasks are likely to be demonstrated by machines in a
relatively short time period and what are likely to require more time for machines to
develop skills for them?
- Major breakthroughs envisioned.
What breakthroughs are possible along the direction of making machines that can develop
and learn autonomously?
What breakthroughs are possible in our understanding of human development and learning?
What are the implications of those breakthroughs?
- Short-term and long-term applications of results from research on development
and learning.
- Major capabilities to be identified.
What major new capabilities are likely to be realized for machines along the direction of
development and learning?
What benefits are such new machine capabilities likely to bring about?
What benefits are such capabilities likely to bring to us for our understanding of human
development and learning?
- Infra-structural requirements.
What are the major differences between traditional methods of computation and those for
machine development and learning?
What kinds of infrastructure are required from industry and government in supporting the
related research?
What kinds of new equipment are needed for future related research?
What kinds of advances are envisioned for the future computer industry and robotic
industry in meeting the need of related studies? Do humanoid robots with embedded computers help?
- Applications of foreseeable research results.
What applications are likely from predictable research results for development and
learning, such as understanding of human and development, understanding of consciousness,
human educational benefits, human-machine interfaces, multimedia based sensor-integration,
situation analysis and decision making, humanoid robots, service robots that work in human
environments, smart toys, software for education and entertainment.
- Opportunities for industrial entrepreneurs and venture capitalists.
What kinds of new industry, new products, new services, and new markets are likely to
result from research on development and learning by machines and humans?
- Funding implications to sciences and engineering including the
identification of potential funding agencies.
How does the current funding structure provide funding opportunities for development and
learning?
What suggestions can the workshop make to governmental and private funding agencies?
The above is a tentative list. Suggestions from the participants
are welcome.
|