You are on page 1of 28

1

SOFT COMPUTING APPLICATIONS


Diego Felipe Pez Granados

OUTLINE
Fuzzy Logic Applications Fuzzy Systems Uses Recent Researches Neural Networks Applications Artificial Neural Networks Uses Recent Researches. Neuro-Fuzzy Systems Evolutionary Computing Applications. Summary

Fuzzy Logic Applications


Is derived from fuzzy set theory dealing with reasoning that is approximate rather than precisely deduced from classical predicate logic. It is a concept that could be applied to any logical problem, instead of the digital or crispy logic.

Current uses
Industrial Control Marine Diesel engines
Low overshoot, highly dynamic and uncertain environment Robust and accurate control. Automatic gear selection behavior Tailored to suit the your driving style Adaptive fuzzy system

Volkswagen Direct-shift gearbox

Energy Saving AC Control


Industrial AC systems use fuzzy logic to minimize energy consumption. A complex control strategies optimizes the load set on heater, cooler, and humidifier. Car AC systems use fuzzy logic to estimate the temperatures at the head of the driver from multiple indirect sensors. Home AC systems are much simpler. They use fuzzy logic for robust temperature control.

House goods
Fridges. Washing machines Saves about 20 % in energy and water consumption

Cameras
Anti-shake Canon and Minolta Image processing - Segmentation of histopathology images

Others
Traffic light control system Seattle, USA.

Recent Research
Energy Management in Hybrid Vehicle Using Fuzzy Control
Fuzzy control methods use fuzzy rules instead of clear rules and fuzzy up the battery state of charge SOC and input power required, then the output ratio of engine and electric motor is obtained by fuzzy decision systems.

From: Chaoqun Zhu and Bin Yang, 2012 IEEE International Conference on Information Science and Technology.

Rule Database and Membership Functions

Results

Advantage *Intelligent Control *Little prior knowledge *Real-Time energy management

10

Neural Networks Applications


Financial Stock Market Prediction Credit Worthiness Credit Rating Bankruptcy Prediction Property Appraisal Fraud Detection Price Forecasts Economic Indicator Forecasts Medical Medical Diagnosis Detection and Evaluation of Medical Phenomena Patient's Length of Stay Forecasts Treatment Cost Estimation Industrial Process Control Quality Control Temperature and Force Prediction

11

Recent Researches
In could be divide in two types of research. Development of new type of neural networks
Advanced neural networks Adaptive neural networks Neuro-Fuzzy Systems (ANFIS)

Applications of Neural Networks using current models.


Robotic learning Navigation in mobile robots Human-Machine interaction Data analysis Unsupervised learning

12

Building High-level Features Using Unsupervised learning


Input: 200x200 pixel images Architecture:
Local Filtering Local Pooling Local Contrast normalization

*Model parallelism 1 billion trainable parameters A single instance of the model out across 169 machines (where each machine had 16 CPU cores).
Quoc V. Le et al. Proceedings of the 29 th International Conference on Machine Learning, Edinburgh, Scotland, UK, 2012

13

Test setup
The test set consists of 37,000 images sampled from two datasets. There are 13,026 faces samples. The network was trained on a cluster with 1,000 machines for three days. The best neuron achieves 81.7% accuracy in classifying faces against random distractors.

14

15.8% accuracy for object recognition on ImageNet with 20,000 categories. It is possible to train neurons to be selective for high-level concepts using entirely unlabeled data. Histograms of faces (red) vs. no faces (blue).

15

The Spaun Brain


Semantic Pointer Architecture Unified Network SPAUN, University of Waterloo, Canada It is composed by: Working memory system *Prefrontal cortex Action Selection system *Basal ganglia *Thalamus

It can observe visual images and indicate its responses with a physical model of an arm. Spaun performs all of these tasks based on the activity of 2.5 million simulated neurons.

16

17

The Spaun Brain

18

Neuro-Fuzzy Systems
Neuro-Fuzzy synergisms for modeling and control, allows to fuzzy controllers not only take linguistic information from human experts but also adapt itself and improve the performance using the numerical data.

19

Structure of a Fuzzy-neural system

20

Neuro-Fuzzy controller applications on marine diesel engines


Marine diesel engines operates in highly dynamic and uncertain environments. Type-2 Fuzzy Logic Controllers can handle such uncertainties. FLC involves a computational overhead associated with the computation of the typereduced fuzzy C. Lynch et al. Using Uncertainty Bounds in the Design of an Embedded Real-Time Type-2
Neuro-Fuzzy Speed Controller for Marine Diesel Engines, 2006 IEEE International Conference on Fuzzy Systems. Vancouver, BC, Canada

21

Type-2 Fuzzy Logic system

22

RT2NFC Structure
Layer 1: Input Layer II: Fuzzification layer Layer III: Inference and Rule Base

Layer IV: Approximates the type-reduced set. Layer V: Defuzzification Layer

23

24

Evolutionary Computing Applications


Learning of a Type-2 Fuzzy controller

25

Path Planning for Unmanned Aerial Vehicle using genetic Algorithms

26

27

Summary
Artificial neural networks can feed very good widely range of problems but still is computationally difficult to train because of the long time needed to iterate and converge to a global solution; also the concept it is black box that doesnt show a relationship between input and output that could be modeled.

Fuzzy logic systems allows controlling highly non-linear process and gives robust and efficient responses, how ever are expert based system that doesnt learn so it is limited to the knowledge of people about the problems.
The join of different methods of soft-computing such as fuzzy logic and neural networks, or genetic algorithms and fuzzy controllers is the most promising fields currently in soft computing.

28

THANK YOU