2014 SR&ED Claim

2014 Scientific Research and Experimental Development tax credit claim

T661 - Part 2 – Project Information

Section B – Project Description

Line 242 -– What scientific or technological uncertainties did you attempt to overcome- uncertainties that could not be removed using standard practice? (Maximum 350 words)

The project is attempting to develop general purpose learning and thinking software for autonomous control of robots. The approach is based on the development on a new artificial neural network (ANN).

Current robot control software is not general purpose. It is specifically developed to function in a predefined environment with predefined tasks to be performed. An example is Google’s self-driving car control software. Learning capability, if present, is limited to this environment. General purpose artificial intelligence architectures exist but are not designed for robot control i.e. they are not grounded on sensory input and device activation. Examples include ACT-R, SOAR, CLARION, LIDA, SIGMA and HTM.  DRAMA, MAXSON and Brook’s COG are experimental architectures designed for robot control but have not been applied with any general purpose success. When such architectures are grounded on real sensory input they also make heavy use of stochastic inference which becomes complex and is a heavy computational overhead. These architectures are complex because they do not use the same structures for pattern recognition and motor action control.

This research is developing a new architecture to solve these problems. However there are no simple hierarchical ANNs based on binary nodes that are feed-forward and use reinforcement learning. There are also no ANNs that grow nodes in a hierarchy for both recognizing objects / sequences and for learning new action habits.

More specifically the project is attempting to:

  • Develop a new hierarchical structure for ANN nodes.
  • Use binary neurons (binons) with reinforcement learning.
  • Grow the ANN by adding nodes for parallel and sequential pattern recognition.
  • Convert graduated sensor readings (sub-symbolic) into symbolic stimuli.
  • Transfer short term memory traces into long term memory.
  • Learn action habits and integrate them into the associated hierarchical behavior network.
  • Execute action habits subconsciously while thinking about (simulating) others.

Line 244 – What work did you perform in the tax year to overcome the scientific or technological uncertainties described in Line 242? (Summarize the systematic investigation) (Maximum 700 words)

Approach: Increase the complexity of the ANN structure and algorithms to process the requirements at greater complexity while still handling the lower complexity features. Run regression tests on already working lower complexity features. Determine its success at learning and thinking based on the observation of its actions in artificial test environments simulated in software and by inspection of its internal memory traces and processes. More details about the research are available at www.adaptroninc.com

Two pieces of software are being researched;

A) In the area of object and sequential pattern recognition from multi-sensor senses producing graduated (ratio scale) and discrete (symbolic / nominal) stimulus readings using a hierarchy of binons. Each new version of this software is then integrated into B.

B) In the area of short term and long term memory, learning, thinking and a hierarchical action habit structure.

In area A:

  1. Tried treating all sensors as independent in Perceptra (see line 244) and found it caused a major performance problem without any improvement in recognition success.
  2. Obtained better recognition performance by only combining property binons at the lowest level.
  3. Tried combining adjacent patterns (as opposed to just the overlapping ones) at higher levels in the hierarchy but found no noticeable improvement in recognition rate.
  4. Tried using only the pixels that changed from one frame to the next in recognition of handwritten digits but got a very low recognition rate.
  5. Added a pruning mechanism to remove less frequently used nodes without reducing prediction accuracy.
  6. Selected a segment of real Morse-code to use in the investigation of the ANN performing sequential pattern recognition.
  7. Tried another approach to finding, counting and representing repeating sequential patterns.
  8. Introduced habituation of repeated stimuli at the sensor level for sequential pattern recognition

In area B:

  1. Added three levels of concentration when performing actions, (learning, practicing and performing) based on the three levels of novelty (new, familiar and learnt) for reinforcement learning.
  2. Tried inhibiting expected stimuli to reduce their distraction while performing action habits.
  3. Used a partial match of the trigger stimulus for selection of the action habit to perform.

Line 246 – What scientific or technological advancements did you achieve as a result of the work described in line 244? (Maximum 350 words)

  1. Published the paper: Perceptra: A New Approach to Pattern Classification Using a Growing Network of Binary Neurons (Binons) in the 12th International Conference on Cognitive Modeling - 2013.
  2. Developed a hierarchical structure for integrating trigger stimuli, actions and goal stimuli based on ideomotor action control theory.
  3. Realized that the process of paying attention is learnt the same way as learning overt actions. It includes reflexive behavior (distraction) and learning on what to focus attention.