1 / 24

Heuristic learning

Heuristic learning. Intelligent Information Analysis based on V-Graph knowledge representation. Speech Plot. Problem Redundancy / Insufficiency / Corruption of input data. Representation Learning Recode Intrinsic Confliction Heuristic Learning

uzuri
Download Presentation

Heuristic learning

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Heuristic learning Intelligent Information Analysis based on V-Graph knowledge representation

  2. Speech Plot • Problem • Redundancy / Insufficiency / Corruption of input data. • Representation Learning • Recode • Intrinsic Confliction • Heuristic Learning • Roaming, interactivity, reuse of learning achievement.

  3. Machine learning • Ideas: • Supervised, Unsupervised learning • Clustering • Dimension reduction • Structure learning • Framework • Artificial Neuron Network Series • Statistical Learning • Feature learning

  4. Deprecated methods • Dead/Stall Methods: • Expert System • Genetic Programming • Expert System: It’s impractical to establish a perfect knowledge system.(Academic Research stalled doesn’t mean no advantage for industrial use) • Software engineering problem • Un-model-able/Un-logic of Real world. • Genetic Programming: Evolution functions and target evaluation aren’t compatible. • Huge under-fitting/over-fitting problem

  5. Under/Over-Fit • Under-fit: Model is too simple. • Over-fit: • Assume input data is noiseless, there won’t be over-fitting problem. • Over-fit confused signal with noise. • Noise? Signal?– Noise could lead to important discovery, e.g. Discovery of Argon • My Theory: Fitting is verification/tool/part of chain/sample/… is not solution. Interactivity could be the solution.

  6. Software engineering problem • Industrialize usage and academic researches are very different. • For machine learning: • Data is precious • And worthless • The simpler/straightforward/constant/… the problem is, the better to use machine learning. • Otherwise we have all witnessed.

  7. Summary of commonly discussed machine learning problems • http://www.denizyuret.com/2014/02/machine-learning-in-5-pictures.html • Illustrated the common issues and description. • Major problem: • Under/Over-fitting • Software-Engineering problem

  8. Representation learning • Aims • Supervised <-> Semi-supervised: a balance between “aesthetics” and reality • Means • Recode of input data • Lessen Variable combination • Detect and remove irrelevant features • De-tangle, Re-tangle. (Very much alike Annealing Algorithms) • Frameworks: • RBMs

  9. Encoding • Deep Architecture • Unsupervised pre-training helps. • Aesthetics • I think it’s essentially “roaming”, in a narrow form • Supervised training has few spots. • Pre-training <-> Training • Cost function compatibility • Data required is huge.

  10. Major problem of current ml methods • Cost functions works well for SIMPLE CASES, for complex cases, dimension of evaluation explodes. i.e. Cost function cannot express choices. • Training data could be very redundant, while insufficient. The achievement by using statistical analysis highly depends on how well the application data meets the analysis assumption. • Different form of issue lies in Expert System. • “Aesthetics” is vital, and un-definable.

  11. Paragraph Summary • Problems to be solved: • Formulate/Automate/Reproduce Aesthetics • Engineering/Noise/Quantity Problem

  12. Heuristic Learning • Main Ideas • Roaming • In strong knowledge relations. • Means • Use V-Graph to represent data. • Use Heurons to express “aesthetics” activity.

  13. roaming • Roaming refers to extending target and learning data during training. • Critical to strong-typed heuristic learning, where original target and learning data could not reach a good result. • Implementation in V-Graph: Heurons • Logic roaming(The meta-movement: generate a meta-link)

  14. V-Graph concept • View • (A->B)@C • Un-logic Quantize Relationships. • Flattened Complex Structure • Database Usage • Search Engine, Linguistic corpus…

  15. Heuron • Neutron like information propagation for V-graph • Ruled Meta-link Generation

  16. Software framework summary • Extended Graph • Capable to express more. • Extended Neuron • Based on Strong Relationship, Qualitative descriptor • Allow quantitative description.

  17. heuristic • Main idea: • Take a reasonable guess, rather than simple parameter fitting.

  18. Heuristic learning framework • Extended Strong typed: Type recognition and transformation performed in V-Graph. • A combination and improvement to Cost function and Evaluation function. • Multiple Heuristic approach allows roaming. • Initial roaming generate a basis set of meta-link. • Heuristic learning still apply statistics to input data, however, in more representative form.

  19. Heurons for instance • Spot • Recall • Short-cut • Preference-Intention • Curiosity • Contrast • Trial

  20. Example: Reverse Pendulum • A solved problem. • Heurons involved: • Short-cut(memory), Stimulus, Trial, (I think it’s essential) • Expectation, Preference, Mimicry, Demand, Interest…

  21. Incremental and Interactivity • The framework has room for the feature, by implement specific heurons.

  22. Summary • V-graph and heurons are methodology to realize Roaming. • Roaming shouldn’t necessarily require V-Graph or heurons. However, the choice of V-Graph and Heurons is because…

  23. Roadmap • V-graph Query Language, V-graph Database. • Architecture Expressway: A methodology and toolkits aims to reduce software engineering difficulties. • Use V-Graph to train an eye tracking program.

  24. Summary • Roaming • Engineering

More Related