Intuition – Pseudoinstinct

From http://searchwarp.com/swa299107.htm

The purpose of intuition is to guide us, to literally lead us, to self-actualization – the process of becoming all that we essentially are and nothing that we are not. It is, in a powerful sense, a guide to our becoming whole and real. I mean here not ‘real’ in terms of scientific real of see, hear, touch. I mean real in terms of authentic, truthful, fully embodying one’s heart and soul.

If you are a follower of your intuition you will know that it always leads you to opportunities – never to dead ends. It always leads you to personal growth – it never banks on what you’ve already done, already learned, without naturally stretching you to discover and to know more of your Self.

Intuition leads us to expansion, yet in a very organic and natural way. It is not the mind dreaming, or planning. It is the inner wisdom leading us to our own next step.

From http://www.thehindu.com/thehindu/edu/2002/03/12/stories/2002031200160203.htm:

An instinct is a reaction. An animal, as soon as it is born, instinctively gets up, walks to its mother, and drinks milk. There is no thinking involved here; it is something that the animal has been programmed to do. It seems as if it is automatic.

Application to Purrpuss:

Instinct – PP brain doesn’t seem capable of generating instinct. Long term memory is a collection of associations. Instinct is something defined by the robot component behaviour, like sensors or actuators. So, I guess in machine terms, instinct can be regarded as special kind of fast interrupt into PPBrain. It’s unlike a regular interrupt that specifies the limit of the sensors. It’s like an interrupt inside the sensors/actuators telling that the robot made should do that specific actions to prevent bad actions. I have no idea how to find these “bad actions” referring to previous robot experience.

Intuition – PP doesn’t really make a decision based on event planning, but it’s more on the association of events planning. I think it’s similar to human’s experience. We made our intuition based on what experience we will encounter later, not exactly the exact events. I think to some extend, human experience can be regarded as associations for PP brain. PP brain with better exposure of associations can probably produce better intuition. The flow of associations determined the actual actions. If we can organise the associations in Long Term (associations) memory of  Purrpuss, I think we can next associations (in which doesn’t necessarily same associations but same flow of associations). These flow of associations (similar concept of flow of events = associations) can be used to estimate possible new associations and thus possible actions towards the associations.

Meeting with John Andreae

I had a meeting with the person who invented PURR PUSS brain. Apparently, he is still in town working on his PURR PUSS project, specifically on robot planning and consciousness. He has been telling me about his system and the Squash-Pop Microworld program.

What makes PURR PUSS different from other reinforcement learning is the ability of the robot to set its own goal, called novelty. It is different from reward system whereas the robot try to find a path to get the reward. The problem with reward system is the events occurs after it get the rewards. The robot tends to follow the rigidness of the path to fetch the earlier rewards and makes it less reliable when something different occurs along the path.

In PURR PUSS, the novelty is the main concern, rather than the actual reward. To be honest, I not fully understand it until I managed to spend more time on researching PURR PUSS. However, the idea is to let the robot to set its own goal as novelty rather than reward. The robot will do series of events that will lead to a novelty, or a loop. Therefore, if it found new event associations, it will try to create a loop or a novelty to go back to the “original loop”. In simpler words, I can imagine that PURR PUSS is more adventurous than generic reward oriented robots. Go to a new world and discover all the possibilities.

We also discussed about John’s work on Squash-Pop Microworld. There are a fully aware robot (teacher) and new robot with PURR PUSS brain working together in a 3×3 grid. The objective is to trap a ball in a corner and squash it. Therefore, these robots need to communicate to achieve the objective.

Basically, John write his code in similar manner of writing for an embedded system. For example, he has a series of event type registers to represent the states and a polling system consisting different hex values of registers, similar to a finite state machine. I am certain that only he can effectively read his own code. So, my job is to redesign the PURR PUSS system into modular object system.

Ps:
I have some ideas running in my head. But, at the same time, I’m having a headache and I have many other works and those drunktards near my flat is partying after midnight. Perhaps I should get couple of Panadol tablets.

PURR-PUSS Learning and Human Learning

One interesting aspect of PURR-PUSS brain is the ability to learn from a teacher. Just like on your first day to school, learning alphabet and numbers. Just like Ayleen O Hanlon said, “Apabila belajar alphabet. A B C. Belajar hitung. Satu Dua Tiga. Tetapi selepas huruf C, kemudian huruf apa? Dan selepas nombor Tiga, kemudian nombor berapa?”

Although it would be trivial for a PURR-PUSS brain to learn alphabet, learning to count might be a bit nontrivial which might be a good option for me to consider to test this particular example in my PURR-PUSS object system later.

Of course the main reason for this post is to show the video. =D

Wonderfully soft-spoken Australian girl delivering her speech during Pertandingan Pidato Antarabangsa Bahasa Melayu Piala TPM. The participant from Malaysia and Holland is also worth to see.

PP Object System

2009-03-17

Created Stimulus and AssociationTemplate classes. Started on Association class.

The idea is to create AssociationTemplate class and Association class will follow the rules from AssociationTemplate. We’ll begin to assign (or possibly create) Stimulus into the associations.

About PURRPUSS, my final year project and This Blog

About PURR-PUSS

Details about PURRPUSS by John Andreae can be found at department website: http://www.elec.canterbury.ac.nz/purr_puss/

 

About my final year project

The purpose of this project is to design a collection of objects using C++ to carry out reinforcement learning experiments based on Purposeful Unprimed Real-world Robot with Predictors Using Short Segments (PURR-PUSS) suggested by John Andreae. The main target for the object system will be C++ developer who need to implement PURR-PUSS brain into their works. Statistical tools such as network description plan and reward graph will be implemented to help developers to evaluate their experiments. 

The PURR-PUSS object system will be used to implement an experiment suggested by John Andreae from his book, possibly BunPie Microworld or First Moment Experiment. This is mainly to verify the result validity and system performance. A new reinforcement learning experiment will be designed to evaluate the performance of long term memory PURR-PUSS brain.

About This Blog

Mainly serve as my journal to document progress made since today. I made it public view so that any steps taken (good or bad) will be benefited by public to a certain extend.