Notes from Zero program West-Coast Planning meeting
Held at SRI International, Menlo Park, 27 Jan 1998.
Bob Balzer, note taker and organizer 


Balancing  PUSH U PULL
				(not like webcast)

      See slide 3 of  zero (html) zero-hyper.ppt 
   
 Push to get as much potential information close to the point of need 
	including data that are of interest given alternative prediction of outcomes
	(but not to the consumer to avoid overload and misinformation)
 Pull based on workflow/model to filter the pushed information.
 Zero-latency of information availability / display when needed.
- Zero Latency is instant pull
 Negative Latency through Information Push

Prediction
 What
- Probable Task Sequences (flow)
- Information needs within task steps (speculative retrieval)
 semantic analysis
 population level correlation
- Information values in hypotetical states (speculative computation)
 How
- Statistical models as baseline
- rule-driven
- learning from observation (e.g. workflow details)
 Control/Manage
- manage meta-data needed for prediction
- Integration between Learning and Described models
 learning agents versus descriptive agents
 declarative specification of workflow
 deal with multiple hypotheses (models)
- data level
n intentional level - first class representation of hypotheses






Architecture
 Composing Latency Management Architectures
- Latency Management Frameworks
 API
 Plug and Play for composability, survivability, fault tolerance, etc.
 Auto-Wrapping
 Support decomposable data sources
- data analog to components
- Retain context when data subset is disseminated
- Latency Management Policy
 Control of complexity
 Latency Analogy to fault tolerance
- what is fall back if zero is unatainable
- Latency Manager
 Self-organizing underlying architecture
- cache information
- resource trading markets
 Resource Constrained Latency management
- reduced latency with limited resources
- specification of requirements
- Layered adaptive management
- choosing intermediate results to maintain (e.g. summary tables)
 Hypothetical Transaction Management Architecture
- Hypothetical Transaction Management API
- Cascaded Hypotheticals
- State deltas
 composition
 specification
- Quick Commit


Intelligent Nomadic Active Agents
 Agents exchange goals, resources, precision, timing
- Meta information passed, reasoned about, acted upon
- Estimates of communications required for agent coordination
 what resources are available
 Process Aware Agents
- Distributed and coordinated
 Loose federation of nomatic agents
- guided by feedback and tips


Model Construction and Evolution
 Event mining on products of workflow
- identify change
- Compound events (temporal aggregation)
- Patterns matched across event space
- Distinguish events in "information system" versus workflow events
 Development and management of probabilistic Models
- contents of information sources
- queries and relationships between quieries
- modeling predictive power of information and attributes
 Maintaining a materialized view
- Know how accurate view is over time
 Model of Information System
- Active Intelligent wrappers
 workflow awareness
 resource awareness
 priority awareness
- Support capability dissemination
 data + software + workflow
 Background building of better models
- zero latency versus background learning (e.g. sleep)
 Scalable formalisms for probabilistic models
- control and workflow


Controlling Information System
 Representation and management of information network
- set of information nodes and resources superimposed on communication network
 Adequateness, completeness, correctness of information
- can it be checked afterwards
- cost/benefit analysis
- automatic optimization of resources
 Management of sensor tasking
- driven by Latency management

Supporting Teams working on common task
 Heterogeneous skills, needs,  and resources
 Conflict resolution within team

--------------------------------------------------  fin -------------------------------------------