Robotic Intelligence

Phillip C-Y Sheu

• Publisher: WSPC
• Year: 2019
• ISBN:
  9789811203497
• Topic: Computer Science
• Subtopic: Computer Vision & Pattern Recognition
• Index: Computer Science

This volume aims to provide a reference to the development of robotic intelligence, built upon Semantic Computing, in terms of 'action' to realize the 'context' and 'intention' formulated by Semantics Computing during the 'thinking' or reasoning process. It addresses three core areas:

Contents:

• Foreword
• Part 1: Understanding Semantics:
  • Goodness of Machine Learning Models (J R Barr)
• Part 2: Data Science:
  • Goodness-of-Fit of Statistical Distributions (J R Barr and S Zacks)
  • Forensics: Assessing Model Goodness: A Machine Learning View (J R Barr and J Cavanaugh)
  • Diffusion Analysis (M Kim and M Hayakawa)
• Part 3: Data Integration:
  • Network Analysis and GOLAP (J Jin and M Hayakawa)
• Part 4: Applications:
  • Automatic Analysis of Microblogging Data to Aid in Emergency Management (S Manna)
  • Applications of Natural Language Processing (NLP) for Improving Classroom Learning Experiences Using Student Surveys (P Kuiper and K Hood)
  • A Semantic Recommendation System for Cancer-Related Articles (C C N Wang, Y-L Chung, I-S Chang and J J P Tsai)
  • Rapid Qualification of Mereotopological Relationships Using Signed Distance Fields (R Schubotz, C Vogelgesang and Dmitri Rubinstein)
• Part 5: Robotic Intelligence:
  • A Learning from Demonstration Framework for Implementation of a Feeding Task (N Ettehadi and A Behal)
  • Cooperative Obstacle Avoidance for Heterogeneous Unmanned Systems During Search Mission (K Harikumar, T Bera, R Bardhan and S Sundaram)
  • Improving Code Quality in ROS Packages Using a Temporal Extension of First-Order Logic (D Come, J Brunel and D Doose)
  • Integrating Planning and Reactive Behavior by Using Semantically Annotated Robot Tasks (A Schierl, A Hoffmann, L Nägele and W Reif)
  • Portmanteau Word-Play for Vocabulary Enhancement with Humanoid Robot Support (D Schicchi and G Pilato)
  • Enhanced Navigation Using Computer Vision-Based Steering Angle Calculation for Autonomous Vehicles (M R Rochan, K A Alagammai and J Sujatha)
  • Efficient Resource Allocation for Decentralized Heterogeneous System Using Density Estimation Approach (J Senthilnath, K Harikumar and S Suresh)
  • Robotic Control for Cognitive UWB Radar (S Brüggenwirth and F Rial)
  • A Mixed Real-Time Robot Hardware Abstraction Layer (R-HAL) (Giuseppe F Rigano, Luca Muratorez, Arturo Laurenzi, Enrico M Hoffman and Nikos G Tsagarakis)
  • Robotic Intelligence and Computational Creativity (A Augello, I Infantino, U Maniscalco, G Pilato, R Rizzo and F Vella)

Readership: For students and researchers in computer science.Robotics;Artificial Intelligence; Semantic Computing;Thinking Machine0 Key Features:

• Encyclopedia with Semantic Computing and Robotic Intelligence is the most up-to-date on the subject that are available today
• Supported by sister e-journal available on http://www.worldscientific.com/worldscinet/escri, where new articles are released online on the regular basis in a monthly basis
• Once every 6-month, the articles published in the e-journal will be compiled into volume as part of the Encyclopedia with Semantic Computing and Robotic Intelligence, as published as ESCRI collection

Part 1

Understanding Semantics

Goodness of machine learning models

Joseph R. Barr

Trust Science, Irvine, CA 92618, USA
[email protected]

In this context, a model is an algorithm or a procedure that applies to data resulting in a functional relation τ between “input space”

and “output space”

. In this short paper, we will delineate objective criteria which help to disambiguate and rate models’ credibility. We will define pertinent concepts and will voice an opinion on the matter of good versus bad versus so–so models.

Keywords: Machine learning; cross-validation; complexity; parsimony; variable selection; AIC; cost function; deviance.

1.Introduction

A model is a mathematical abstraction of a physical phenomenon. A statistical, or machine learning, model is one which learns “patterns” in the data and is able to predict new instances. The term data science is sometime used to describe that domain.

A model is inherently inaccurate because any procedure used to produce a model is based on heuristics and short of an exhaustive search through an “infinite set,” it (an optimal model) could never be found. Thus a somewhat redundant observation in the spirit of Turing’s “Halting Problem” is that there is no effective computational method to determine whether a model is optimal. This however does not abrogate one from being able to assess a model.

It is generally accepted that a model is good if it has some application or serves some purpose. There is no consensus, however, as to what it means for a model to be correct.

To paraphrase George Box’s pronouncement that a model is only as good as its utility, practicality often helps distinguish goodness or badness of a model, therefore, a good model is one that is able to predict accurately enough. For example, Kepler’s laws of planetary motion are good not because they give the right answer like the shape of the orbit of earth around the sun. Kepler’s laws are good because they are rather accurate as laws of classical mechanics which are able to predict celestial events “accurately enough.” There are application, however that require better accuracy. At times Kepler’s equations are not “accurate enough” because the equations of Newtonian mechanics do not tell the whole story. A case in point, the Global Positioning System (GPS) uses relativistic corrections to “fix” for gravitational “distortions” of time and space.

2.Data and Predictive Algorithms

At the risk of oversimplifying, statisticians, [sic] “machine learners” do one of the two things: They classify items as one of two types, say 0 or 1, and they associate a numerical (or vector) value y to an input X, a vector. Often there are additional layers associated with a model which further result in additional insights.

Lying at the basis of the pyramid is the data, a set of vectors (X, y) =

In machine learning the data is also referred to as the examples. Clearly not every dataset has “neat” form. In fact, production data like log files, surveys, financial transactions, etc. contain disparate forms of data including alphanumeric, textual, dates, IP addresses, images, voice and so on. Practitioners are deft at transforming production data into sets of vectors whose entries are numerical. Numerical values may be real (continuous), Boolean (0/1), categorical (many-value ordinal). Each vector (X, y) is called a record, or in the context of statistical modeling/machine learning, an example. The two archetypical cases are

and

= {0, 1} where

represents the set of all possible output values. The former is related to regression and the latter to classification. In the presence of three classes

= {0,1,2}, it is common to classify every (x₁, x₂, . . . , x_p) as y = 0 or as y = 1 and the further classify the “1 ” class (as class 1 or 2). Whether a model is statistical (like generalized linear models) or strictly “machine learning” (like “deep” neural networks), or a hybrid, the person who builds a model shall be henceforth known as a statistical modeler, or sometimes just modeler.

3.Models

Over the past century and a half, since Sir Francis Galton introduced regression as a predictive procedure in the 1880s, a plethora of modeling techniques were developed to address diverse prediction problems. A statistical modeler has to straddle ...

Table of contents

• Cover
• Halfitle
• Series Editors
• Title
• Copyright
• Contents
• Foreword
• Part 1: Understanding Semantics
• Part 2: Data Science
• Part 3: Data Integration
• Part 4: Applications
• Part 5: Robotic Intelligence