Sc-net PDF Print E-mail
Written by Rizki Noor Hidayat Wijayaź   

MPIL is an instance-based learning system (instances are simply viewed as points in n-dimensional real-space with an associated neighborhood), which utilizes two models for creating neighborhoods. The first model (i.e., MPIL-1) places a single neighborhood sphere (based on Euclidean distance measure) around an instance, and is in nature similar to the nearest neighbor classifier, except that it removes redundant instances.

The second model (i.e., MPIL-2) incorporates N radii (one for each input of an instance). This model also supports knowledge acquisition in the form of rule extraction. In a sense, both approaches are similar to neural networks in that they exploit a very similar parallelism. MPIL represents a good alternative in cases were large amounts of data have to be learned and provides good facilities for storage reduction.

Features:

    • Provides symbolic interface which allows the user to create:

      a) Input and output definition files. b) Pattern files.

      c) Help files for objects i.e., inputs, input values, and outputs).

    • Supports categorization of inputs. This allows the user to readily access inputs via a popup menu within the main TDL menu. The hierarchical structure of the popup menu is under the full control of the application developer i.e., user).
    • Symbolic object manipulation tool:

      a) Allows the user to interactively design the input/output structure of an application. The user can create, delete, or modify inputs, outputs, input values, and categories.

      b) Furthermore, inputs and categories can be moved from one location to another.

      c) Receive a quick overview of the hierarchical category structure.

    • Supports Rule representation:

      a) Extends standard Boolean operators i.e., and, or, not to contain several quantifiers i.e., atmost, atleast, exactly, between.

      b) Provides mechanisms for rule revision i.e., refinement and extraction.

      c) Allows partial rule recognition. Supported are first- and best-fit.

    • Allows co-evolution of different subpopulations based on type of transfer function chosen for each subpopulation.
    • Provides three types of crossover operators: simple random, weighted and blocked.
    • Supports both one-shot as well as multi-shot learning. Multi-shot learning allows for the incremental acquisition of different data sets. A single expert network is constructed, capable of recognizing all the data sets supplied during learning. Quick context switching between different domains is possible.
    • Two types of local learning rules are included: perceptron and delta.
    • Implements 5 types of unit transfer functions: simple threshold, sigmoid, sigmoid-squash, n-level threshold, new n-level-threshold.
    • Data sets can contain either binary or continuous inputs and outputs.
    • Automatically constructed networks can be either tested i.e., measure performance accuracy or used for classifying new patterns.
    • Batch training and testing both in one-shot and multi-shot mode is supported.
    • Graphical interface allows user to view the construction of a network over time and view the change in unit activation during testing or classification. Obtain detailed information on individual network units i.e., unit id, weights, connections, transfer functions, etc. In case of n-level threshold or new-n-level threshold functions the user can view the activationoutput function. Using the network dependency option the user can also view subnetworks.
    • Over a dozen statistics are collected during various batch training sessions. These can be viewed using the chart option.
    • On-line help is available.
    • Network and memory resouces can be viewed directly.

      [download: 2,67 MB]