生物视角下的机器学习与物联网：数据驱动的生物技术与医疗进展

需积分: 9 75 浏览量更新于2024-07-18 收藏 6.66MB PDF 举报

《机器学习与物联网：生物视角》是一本由Shampa Sen、Leonid Datta和Sayak Mitra合编的著作，由CRC Press和Taylor & Francis Group出版。该书聚焦于当今计算机科学领域内发展迅速的两个关键领域——机器学习和物联网（IoT）。机器学习的核心是赋予计算机通过学习和预测而不依赖明确编程的能力，而物联网则致力于通过传感器将所有事物（包括人）连接起来，实现万物互联。在生物学领域，随着技术的进步，大量数据的生成使得生物技术与医疗保健研究面临海量信息处理和分析的挑战。这本书深入探讨了如何利用机器学习和物联网技术来提升这些领域的研究效率和精确度。生物学家们正在处理的不仅仅是基因测序、蛋白质结构分析等基础数据，还包括实时监控、环境因素影响分析等方面的数据。机器学习可以对这些数据进行模式识别、异常检测和预测，帮助科学家们发现新的生物规律和潜在疾病标志物。另一方面，物联网设备能够收集并传输生物样本的实时数据，如健康监测、生态变化等，为研究提供了实时背景信息。书中可能涵盖了以下知识点： 1. **生物数据挖掘**：如何运用机器学习算法，如聚类、分类和回归，从大量生物数据中提取有价值的信息，例如基因表达谱、蛋白质相互作用网络等。 2. **物联网在生物医学中的应用**：通过植入式传感器或可穿戴设备，实时监测生物信号，如心率、血压、血糖等，以及环境因素对健康的影响。 3. **生物信号处理**：介绍如何利用机器学习处理生物信号，如神经元活动、生理信号分析，以便于理解和诊断疾病。 4. **个性化医疗**：结合机器学习和物联网，开发个性化医疗方案，根据个体的生物特征和行为数据提供定制化医疗服务。 5. **生物安全与隐私保护**：讨论如何在利用生物数据的同时，确保数据的安全性和个人隐私。 6. **伦理与法规**：书中可能还会涉及生物数据使用的伦理问题，以及如何在遵守相关法律法规的前提下进行跨学科研究。《机器学习与物联网：生物视角》这本书不仅提供了理论框架，还可能包含实际案例研究和未来趋势分析，帮助读者理解这两个技术如何重塑生物科学领域的研究方法和实践。通过结合这两个领域的优势，本书为生物学家、医疗专业人士和工程师们提供了一种全新的工具，以推动生物领域科研的前沿发展。

Machine Learning

A Powerful Tool

for Biologists

Mohd Zafar, Ramkumar Lakshmi Narayanan,

Saroj K. Meher, and Shishir K. Behera

CONTENTS

1.1 Machine Learning (ML): An Overview ...........................................................2

1.1.1 What is ML? .........................................................................................2

1.1.2 ML versus Other Computing Environments ........................................2

1.1.2.1 ML versus Deep Learning ..................................................... 2

1.1.2.2 ML versus Statistics ...............................................................2

1.2 Various Approaches to ML ...............................................................................3

1.2.1 Decision Tree (DT) ...............................................................................3

1.2.2 Articial Neural Networks ...................................................................4

1.2.3 Deep Learning ...................................................................................... 8

1.2.3.1 Deep Learning Algorithm .....................................................9

1.2.4 Support Vector Machines .....................................................................9

1.2.5 Clustering ............................................................................................10

1.2.6 Bayesian Networks ............................................................................. 10

1.2.7 Mining Methods ................................................................................. 12

1.3 Data Mining ....................................................................................................13

1.3.1 Knowledge Discovery Process ........................................................... 14

1.3.2 Classication .......................................................................................15

1.3.3 Clustering ............................................................................................15

1.3.4 Regression ........................................................................................... 16

1.3.5 Outlier Analysis .................................................................................. 16

1.4 Big Data .......................................................................................................... 17

1.4.1 Why Big Data? .................................................................................... 17

1.4.2 Big Data Implementation Framework ................................................ 17

1.4.3 MapReduce ......................................................................................... 18

1.4.4 Big Data and Biology ..........................................................................19

1.4.4.1 Cluster Computing in System Biology .................................20

1.4.4.2 Grid Computing in System Biology .....................................20

1.4.4.3 General Purpose Graphic Processing Units (GPGPUs)

Computing in System Biology .............................................20

1.4.4.4 Cloud Computing in System Biology ..................................20

References ................................................................................................................22

2 Machine Learning and IoT

1.1 MACHINE LEARNING (ML): AN OVERVIEW

1.1.1 What is ML?

ML is a method of data analysis dealing with the construction and evaluation of

algorithms. It is the science that gives computers and computing machines the ability

to act without being explicitly being programmed. It is dened by the ability to

choose effective features for pattern recognition, classication, and prediction based

on the models derived from existing data (Tarca etal. 2007). A suitably programmed

computing machine is required to perform the aforementioned tasks with the help

of an efcient data analysis algorithm in such a way that the classier itself is highly

mechanized without the involvement of human input.

1.1.2 ML versus Other COMputing envirOnMents

Data mining is an advanced technique used for data exploration in these domains

and is broadly categorized into descriptive and predictive approaches. Under the

descriptive approaches, the interesting patterns (i.e., relations) in the dataset can be

identied and clustered into meaningful groups. The predictive approaches refer to

supervised learning which establishes the relationship between input (independent)

variables and target (dependent) variables through a structural presentation, that is, a

model. Among different ML algorithms, a supervised learning algorithm includes a

set of mathematical instructions which can derive a typical, high-dimensional model

in the form of input-output relationships. The developed supervised learning model

help in identication of mapping of input variables to output variables based on a given

example of joint observations of the values of these variables (Geurtsetal.2009).

Data science compares and overlaps with many related elds such as ML,

deep learning, articial intelligence, statistics, operations research, and applied

mathematics. ML is largely a hybrid eld, taking its motivation and practices from

both computational science and mathematics. It is the core part of both data mining

and predictive analytics.

1.1.2.1 ML versus Deep Learning

The two main supervised models of ML are classication and regression. The

regression model is used to predict the demand of a given product in relation to

its characteristics. However, the classication model maps the input variables into

predened classes. The widely used classiers under ML approaches are support

vector machine (SVM), articial neural network (ANN), and decision trees (DT).

Deep learning is referred to as the intersection between ML and articial intelligence.

It is a subeld of ML focusing narrowly on a subset of ML tools and techniques

supported by neural network inspired algorithms.

1.1.2.2 ML versus Statistics

The main focus of ML is the study and design of systems which can learn from

data. On the other hand, statistical modeling represents the mathematical equation

revealing the relationships between variables of a data set. ML is all about predictions,

classication, regression, forecasting, clustering, and association models, whereas

3Machine Learning

statistics is concerned about sample, population, and hypothesis testing. ML

techniques and approaches completely rely on computing power. On the other hand,

statistical techniques can be developed where computing power is not visualized

asan option.

1.2 VARIOUS APPROACHES TO ML

1.2.1 DeCisiOn tree (Dt)

The application of DTs is very popular in biological data mining because of its

simplicity and transparency (Table 1.1). The constructed DT is self-explanatory in

nature and can efciently explain and interpretate the output result without the help

of a data mining expert.

The structure of a DT has much resemblance in real life, as well as in articial

intelligence life. It is an intersect between articial intelligence and statistics,

which can be used to build the predictive model from the observation of a system

(Geurts etal. 2009). In its structure, the input and output variables are known as

attributes or features, while joint observations of these values are called objects.

The given example of objects, which is supposed to be used as a model, is known

as a learning sample.

A DT is made up of various nodes, which can appear in the form of rooted tree

having no incoming edges. Besides, the other nodes comprise exactly one incoming

edge. A node with outgoing edges is called an “internal” or “test” node, whereas

all other nodes are known as “leaves” (also called “terminal” and “decision” nodes)

(Figure 1.1). As per the specic discrete function of the input variables, each internal

node can split the case space into two or more subspaces.

TABLE 1.1

Application of DT in Biological Science

Type of DT Study Focus References

Genetic algorithm based

DT and neural network

Prediction of benthic macroinvertebrates D’heygere etal. (2003);

D’heygere etal. (2006)

Genetic algorithm

basedDTs

Prediction of microRNA target genes Behzad etal. (2015)

Wavelets and genetic

algorithm

Extraction of cancer data classication

using mass spectrometry

Nguyen etal. (2015)

DT DT for binding of dipeptides to the

thermally uctuating surface of

cathepsin K

Nishiyama (2016)

Genetic algorithm

basedDTs

Ensemble model (iACP-GAEnsC)

proposed for classication of anticancer

peptides

Akbar etal. (2017)

Discrete bacterial

algorithm based DTs

Selection of feature in classication of

microarray gene expression cancer data

Wang etal. (2017b)

4 Machine Learning and IoT

The complexity of a DT depends on available variables in data and the total

number of nodes, leaves, and tree-depth. It can be controlled by using a suitable

pruning method and stopping criteria (Geurts etal. 2009).

The conventional greedy inducers such as ID

, C4.5, and CART (Rokach and

Maimon 2008) are widely used in DT, which help in construction of a DT in recursive

manner, that is, divide and conquer. The greedy nature of DT induction usually fails

to elucidate underlying relationships between the datasets and endpoints (output).

However, using evolutionary programing, the induction of a DT is more signicant

than the induction by recursive portioning. It has been reported that the evolutionary

programming induced DTs to exhibit a signicant accuracy on previously unseen data

(DeLisle and Dixon 2004). They have produced less complex classiers with an increase

in predictive accuracy of 5% to 10% over the traditional method (DeLisle and Dixon

2004). Evolutionary Algorithm (EA) is a search heuristic mimicking the process of

natural biological evolution. During induction using EA, a collective learning process

within the selected population of individuals is randomly initiated. Using the EAs, the

problem of local minima in the search space can be overcome by randomization process

(Zafar etal. 2010). Multiobjective algorithm comprises different objectives in aggregated

form and combined into one objective using a xed weight. Ant colony optimization

(ACO) algorithm is another heuristic-based algorithm successfully applied to extract

classication rules, and DT induction in unexplored research elds. Bacterial-based

optimization algorithms with embedded weighted feature selection strategy has been

developed with the aim to reduce the feature dimension in classication. The feature

selection algorithm-based optimization processes possess low computational complexity

and improve search ability, even in discrete optimization problems (Wang etal. 2017a).

1.2.2 artifiCiaL neuraL netWOrks

The inspiration for computing with ANNs came from various studies on biological

neurons. The major components of a neuron are shown in Figure 1.2. They consists of

Tall

Medium

Short

Branch

Female

Male

Height

Gender

Internal nod

Root node

Leaf node

>1.8 m

>2.0 m

<1.3 m

<1.5 m

FIGURE 1.1 Typical diagram of DT.

5Machine Learning

dendrites, soma, axon, and synapses. While dendrites are the receptive zones (input

surface of the neuron), the axon is the transmission (output) line. Synapses connect

the axon of one neuron to various parts of other neurons. When the input signals come

in contact with these synapses, it results in local changes in the input potential in the

cell body of receiving neurons, which eventually spread through the main body of

the cell, and are then weighted.

ANNs possess neurons joined together by variable strength connections (synaptic

weights) to form a highly interconnected information system, which helps in the

modeling and control of nonlinear systems (Hornik etal. 1989).

The input signals from an articial neuron (Figure 1.3) ow through a gain or

weight called synaptic weight, whose function is analogous to that of the synaptic

junction in a biological neuron. The weights can be positive or negative depending on

the acceleration or inhibition of the ow of signals in a biological cell. The summing

node accumulates all the input weighted signals, adds a bias signal, and then passes

to the output through the activation function. The most commonly used activation

function is the sigmoidal function also called the log-sigmoid or logistic function,

which is dened by:

()=

−

(1.1)

A generalized example of a sigmoid function with inputs ranging between −1

to +1 is shown in Figure 1.4.

The interconnections of these neurons results in an ANN, and its prime objective

is to emulate the human brain and to solve scientic, engineering, and many other

real life problems. Various neural network models have been developed in the

4 parts of a typical nerve cell

Dendrites: Accept inputs

Soma: Process the inputs

Axon: Turn the processed

inputs into outputs

Synapses: e electrochemical contact

between neurons

FIGURE 1.2 Typical biological neuron. (From Rene, E. R., “Removal of volatile organic

compounds using compost bioters.” PhD Thesis, Submitted to the Dept Chem. Eng., IIT

Madras, 2005.)

剩余357页未读，继续阅读

ddalecn

粉丝: 8
资源: 31

生物视角下的机器学习与物联网：数据驱动的生物技术与医疗进展

机器学习与物联网：生物学视角

"异构超密集网络的机器学习技术研究：智能手机、移动应用与物联网驱动的发展

"Espressif IoT SDK V0.9.11 编程手册详解与版本更新

Machine-Learning-for-IoT-master_iot_iot_machinelearning_IOTmatla

Machine Learning and Intelligent Communications

Learning-IoT-CoAP-源码.rar

Machine-Learning-for-IoT-master

Learning-IoT-CoAP:“学习物联网”一书的CoAP章节的源代码-Source code learning

Learning-IoT-UPnP:“学习物联网”一书的UPnP章节的源代码-Source code learning

Learning-IoT-MQTT:《学习物联网》一书的MQTT章节的源代码-Source code learning

最新资源