1. Arrowhead framework
1.1. Showcase
1.1.1. Robotic showcase
1.1.1.1. Edge computing
1.1.1.1.1. AI applications
1.1.1.2. Automatic orchestration
1.1.1.2.1. Reinforcement learning
1.1.1.2.2. Evolutionary learning
1.2. Problems
1.2.1. Local clouds
1.2.1.1. Security
1.2.1.2. interoperability
1.2.1.2.1. Current background
1.2.1.3. automatic orchestration
1.2.1.3.1. Question
1.2.2. Inter clouds
1.2.2.1. Security
2. How can ML help?
2.1. Learning ability
2.1.1. What
2.1.1.1. The ability to infer information from data
2.1.1.2. The ability to learn more abilities to learn from previous experience
2.1.2. Why
2.1.2.1. To reduce human's interference as possible
2.1.2.1.1. Fully automatic reduce human's labor
2.1.2.1.2. Can do it faster
2.1.2.1.3. Can discover more than human can (maybe)
2.1.3. How
2.1.3.1. Current state-of-the-art
2.1.3.1.1. Deep Artificial Neural Network
2.1.3.1.2. Traditional ML
2.1.3.2. No idea actually
2.1.3.2.1. Possibility to look into how human's brains work
2.1.3.2.2. Thoroughly literature review, maybe can invent a new learning algorithm
2.1.3.3. Need for research
2.2. Edge computing
2.2.1. Detecting security threat
2.2.2. Other applications depending on study cases
3. How to express itself that is comprehensive to human
3.1. Why?
3.1.1. For the love of God, humans like taking control of everything (or having the idea that everything is under control)
3.1.1.1. Increase trust
3.1.1.2. Safer, if humans can intervene when something goes wrong
3.1.1.3. Easier to shift the blame
3.2. What?
3.2.1. Visualization?
3.2.2. Feedback
3.3. How?
3.3.1. Need for research
4. General Intelligence
4.1. Ability to reason
4.2. plan
4.3. problem solving
4.4. think abstractly
4.5. comprehend complex ideas
4.6. learn quickly
4.7. learn from experience
5. Dynamic aware systems
5.1. What
5.1.1. Can recognize changes in the system
5.1.1.1. Adjust the system accordingly (automatically)
5.1.2. Can apply changes without interrupting the running system
5.1.3. Can evolve itself to adapt to new settings
5.2. Why
5.2.1. Critical systems (nuclear, hospital) need to be running all the time
5.2.2. Optimize the system when something happens
5.2.2.1. Chaining in the system itself
5.2.2.1.1. New services coming
5.2.2.1.2. Current services die out
5.2.2.2. Changing in the environment
5.3. How is it linked to arrowhead framework
5.3.1. Interoperability
5.3.2. Dynamic orchestration
6. Local cloud as a brain
6.1. Memory
6.1.1. Data
6.1.2. Patterns infer from data
6.1.3. Ability to infer more patterns from learnt pattern
6.2. Learning functionality
6.3. Sensors
6.3.1. "Driver": convert data collected from sensors
6.3.1.1. Save useful data to memory
6.3.1.2. Input to other functions
6.4. Functions that run the system
6.4.1. Unconscious functions
6.4.2. Conscious functions
6.4.2.1. Manipulate reality
6.4.2.1.1. Collect more data
6.4.2.1.2. Advance the system
7. Cognitive psychology refers to all processes by which the sensory input is transformed, reduced, elaborated, stored, recovered, and used
7.1. Each sensor has their own sensory memory (for example visual memory)
7.2. Pattern recognition
7.2.1. Pattern recognition that can be done based on experiences is done by long term memory
7.2.2. Learn new pattern from sensory input out of context, has no experience before hand
7.2.3. Sensory memory uses template theories
7.2.4. Feature theories identify patterns by listing their parts
7.3. "Attention" functions as a filter for sensory input
7.3.1. Many different visual objects appear before us but we can only "focus" on one events at a time
7.3.2. At least in DNN algorithms, this wall has been broken since DNN algorithms can detect multiple objects/events at a same time. There is limit to number of events, but the threshold is much higher than human's capacity
7.3.3. This is not a requirement, this is a technical issue