CYCOGS® brand of AI ML consulting
CYCOGS® Artificial Intelligence Machine Learning (ML)
implementation, design and consulting.
experienced team is available for consultation in many
Artificial Intelligence (AI) / Machine Learning (ML)
Planning on using an AI / ML system in the future? Get your data collection going now.
Your first step is collecting the appropriate quantity and quality of data and insuring its prepared for its current purpose, and what you may need for the future.
Data, from your data, to any available data, is a resource and a tool waiting to be used.
A ML systems performance must be evaluated on reproducing known knowledge from data. Collect data now, as future uses have yet to be needed or discovered.
But data collected improperly becomes garbage. No amount of data preprocessing can fix some problems.
With AI / ML systems, garbage in may lead to garbage out, no matter how much data mining is performed trying to pull usable, actionable information out.
No preparation, possibly no future, but we can help you with your data and AI / ML.
If you do have a current need, the CYCOGS®
company can help you prepare to use AI / ML technology for now and for the future.
Machine Learning (ML)
is a subset in the field of Artificial Intelligence (AI), sometimes called Data Science / Statistical Learning / Probability Theory.
Machine Learning (ML) focuses on predictions
using known properties or discovery of unknown properties.
Most Machine Learning uses statistical techniques running on computing systems such as your corporate network to process data and non-programically learn from the data.
Using back propagation can help verify the Machine Learning (ML) models and data. Algorithms and Modeling can be used and developed to enable data driven decisions and predictions.
Machine Learning (ML) can break the process of programing algorithms and the extensive debug and testing cycle with more of a black box approach.
The Machine Learning predictive ability using computers relies on mathematical optimization, computational statistics, predictive analytics and data analysis.
This is a small contrast to Expert Systems
where the Expert System is preprogrammed with “Expert” knowledge and decision making abilities.
We at the CYCOGS®
company feel the difference between the Machine Learning (ML) ability to reproduce known knowledge verses the focus of unknown knowledge in
Knowledge Discovery and Data Mining (KDD)
is too small a distinction to worry about.
The use of Machine Learning (ML) is not easy or fool proof and problems can arise. Such problems include Data Bias
(sometimes referred to as Variance Decomposition or Quantify Generalization errors),
using training data that is not fully representative, over optimized, wrong model, etc., can lead to results in error or skewed.
The improper optimization and not minimizing the loss function must be addressed.
Human judgement may ignore or simply undervalue proper data representation and the results could be useless in some or all situations.
Other concerns are not fully validating the classification learning model, poor training data selection and faulty data interpretation.
A weak model may not yield sensitivity, give false positives and experience other reliability issues.
Learning models can also be overdone, called “over fitting the model”
. Lastly, the collection, storage and use of the pre and post data sets is an ongoing concern.
For example, if you collected personably identifiable data 10 years ago, and a security breach liberated such confidential information,
or if the conclusions were manipulated, ethical questions and issues will arise.
company is not responsible nor held liable for any such breech, data manipulation or erroneous results.
The Machine Learning (ML) approach uses many techniques and methods, depending on the need and target application.
ML uses classification schemes
to sort through the data, called Learning Classification Systems (LCS) which identifies,
learns or resolves rules, also called Rule-Based Machine Learning
company can help you determine if your data or elements can be classified using binary classification into single or parallel trees
(to avoid over specialization) in a hierarchy progression and use Classification Tree / Decision Tree.
If your data or elements can be classified into different categories using a distribution function and estimating possibilities, use the Logistic Regression
If your data or elements can be classified using the Bayes’ theorem for a set of completely independent variables, use the Naïve Bayes
See also Bayesian Networks
If your data or elements can be classified or split into two categories to determine a mathematical function including classification and regression for the data,
use the Support Vector Machine
If your data or elements, (possibly non-linear and or complex) can be analyzed similar to the function of a biological brain consisting of one or more layers of a Neural Network
(NN) / Artificial Neural Network
(ANN), based on Perceptron’s and Probability Reasoning, use the Deep Learning
If your data or elements can be placed into similar subsets, where these subsets can be called clusters.
Such clusters can be analyzed based on cluster features such as similarity between cluster data and variations between different clusters.
Use the unsupervised Learning method of Clustering
If your data or elements can be pre-programmed using encoded background knowledge as logical facts, use the Inductive Logic Programming
company can help with these
Machine Learning (ML)
learning task categories which can be based on several learning types.
One Learning Type is the Reinforcement Learning
type in the form of punishment and reward as a feedback mechanism.
Another type is Active Learning
labeling over a restricted instance set.
The Un-Supervised Learning
type of possibly unknown features and patterns is similar to the Semi-Supervised Learning type based on a training set which is missing some output targets.
The Supervised Learning
type is where examples of input and outputs are taught in order to form general rules covering input to output relationships.
A look at Dimensionality Estimation / Reduction
type can map out distributions in one or more spaces.
The type of Developmental Learning (Robot Learning) cumulatively generates skills and solutions through Autonomous Experiences and Learning which may use Genetic Algorithms
Genetic Algorithms (GA) are closely related to Evolutionary Algorithms for solution determination and is a search heuristic.
In Machine Learning (ML) data or elements need some preprocessing before being sent to the classification or prediction routines.
One learning algorithm used for this preprocessing is the Representation Learning Algorithm
In addition, data noise may be reduced using space matrix representation from the Sparse Dictionary
method, which is very useful in image analysis.
can help you in these and other examples:
- Develop a ML plan for your organization.
- Help you select data and prepare it for use now and in the future.
- Help you find value in the data and select data sources for the future.
- Help setup ML systems and its training and verification, including choosing a classification type, preprocessing and analysis.
In some ways, Machine Learning (ML) seems to be a lot of terms, methods and different ways to reach an outcome.
We at the CYCOGS®
company think analyzing data using Machine Learning (ML) is a bit more involved than using an Excel spreadsheet or a fancy SQL query.
The first step may be recognizing what to apply. If you are having trouble connecting some data to a result, let us help.
The AI Machine Learning space is evolving and complex, please contact us to help you prepare now.
Contact information such as your name, email address, title, company, and telephone number.
Contact: Send questions and comments about this web site to the CYCOGS® Contact.
Sales E-mail: sales@CYCOGS.com