量化自我：机器学习在感官数据中的艺术与行为洞察

需积分: 9 97 浏览量更新于2024-07-19 收藏 13.46MB PDF 举报

"《机器学习在量化自我中的艺术：从感官数据中学习》是一本属于Cognitive Systems Monographs系列的学术著作，该系列旨在快速、非正式但高质量地探讨认知系统研究的新发展。本书由Mark Hoogendoorn、Burkhardt Funk等人编著，涵盖了自我追踪作为现代生活的一部分，特别是在健康领域的广泛应用。随着可穿戴设备和智能手机的普及，它们轻松支持自我追踪，产生了海量的数据，这些数据的分析和利用成为了一个挑战。在当今社会，人们通过智能设备如健身追踪器、心率监测器等实时记录自己的生理指标、行为习惯等数据，形成了所谓的“量化自我”。然而，面对如此庞大的数据量，如何从中提取有价值的信息并转化为实际行动的洞察力，是机器学习在这个领域的重要课题。机器学习技术，特别是深度学习和数据分析算法，被用来处理这些复杂的数据流，识别模式，预测趋势，甚至个性化推荐健康管理方案。本书深入探讨了机器学习在量化自我方面的应用策略，包括但不限于数据预处理、特征工程、模型选择和优化，以及如何通过模型解释性来理解算法的决策过程。它还涉及了跨学科的研究，将认知科学、生物学与工程学相结合，强调了人机交互和智能系统的智能化设计在提升生活质量中的作用。此外，书中可能还会介绍一些实际案例，展示如何通过机器学习技术解决诸如运动分析、情绪识别、睡眠质量评估等问题，以及如何在保障用户隐私的同时，实现个性化健康管理和预防医学。《机器学习for the Quantified Self》是一本既理论深入又实践导向的著作，对于那些对自我追踪和机器学习感兴趣的读者，无论是科研人员、工程师还是健康管理专业人士，都具有很高的参考价值。"

2 1 Introduction

can lead to severe mental health problems. The app allows Bruce to pick up early

signals on a pending mood swing and to make changes to avoid relapsing into a

depression.

While Arnold and Bruce might be two rather extreme examples, they do illustrate

the developments within the area of measurement devices: more and more devices

are becoming available that measure an increasing part of our daily lives and well-

being. Performing such measurements around one’s self, quantifying one’s current

state, is referred to as the quantiﬁed self, which we will deﬁne more formally in

the next section. This book aims to show how machine learning, also deﬁned more

precisely in this chapter, can be applied i n a quantiﬁed self setting.

1.1 The Quantiﬁed Self

The term quantiﬁed self does not originate from academia, but was (to the best of

our knowledge) coined by Gary Wolf and Kevin Kelly in Wired Magazine in 2007.

MelanieSwan[114] deﬁnes it as follows:

Deﬁnition 1.1 The quantiﬁed self is any individual engaged in the self-tracking of

any kind of biological, physical, behavioral, or environmental information. There is

a proactive stance toward obtaining information and acting on it.

When considering our two example individuals, Arnold would certainly be a

quantiﬁed self. Bruce however, is not necessarily driven by a desire to obtain infor-

mation, more by a better way of managing his diseases. Throughout this book we are

not interested in this proactive stance, but i n people that perform self-tracking with

a certain goal in mind. We therefore deviate slightly from the deﬁnition provided

before:

Deﬁnition 1.2 The quantiﬁed self is any individual engaged in the self-tracking of

any kind of biological, physical, behavioral, or environmental information. The self-

tracking is driven by a certain goal of the individual with a desire to act upon the

collected information.

What data precisely falls under the label quantiﬁed self is highly dependent on

the rapid development of novel measurement devices. An overview provided by

Augemberg [9] demonstrates the wealth of possibilities (Table 1.1). To what extent

people track themselves varies massively, from monitoring the personal weight once

a week to extremes that are inspired by projects such as the DARPA’s LifeLog. For

example, in 2004 Alberto Frigo started to take photos of everything he has used

with his right hand, captured his dreams, songs he listened to, or people who he has

met—the website 2004–2040.com is the mind-boggling representation of this effort.

Let us focus a bit on how widespread the quantiﬁed self is in society. Fox and

Duggan [47] report that two thirds of US citizens keep track of at least one health

indicator. Thus, following our deﬁnition, a large fraction of the US adult population

1.1 The Quantiﬁed Self 3

Table 1.1 Examples of quantiﬁed self data (cf. Augemberg [9], taken from Swan [114])

Type of measurement Examples

Physical activities miles, steps, calories, repetitions, sets, METs (metabolic

equivalents)

Diet calories consumed, carbs, fat, protein, speciﬁc

ingredients, glycemic index, satiety, portions,

supplement doses, tastiness, cost, location

Psychological states and traits mood, happiness, irritation, emotions, anxiety,

self-esteem, depression, conﬁdence

Mental and cognitive states and traits IQ, alertness, focus, selective/sustained/divided

attention, reaction, memory, verbal ﬂuency, patience,

creativity, reasoning, psychomotor vigilance

Environmental variables location, architecture, weather, noise, pollution, clutter,

light, season

Situational variables context, situation, gratiﬁcation of situation, time of day,

day of week

Social variables inﬂuence, trust, charisma, karma, current role/status in

the group or social network

belongs to the group of quantiﬁed selves. Even if we restrict our deﬁnition to those

who use online or mobile applications or wearables for self tracking, the number of

users is high: An international consumer survey by GfK [50] in 16 countries states

that 33% of the participants (older than 15 years) monitor their health by electronic

means, China being in the lead with 45%. There are many indicators that the group

of quantiﬁed selves will continue to grow, one is, the number of wearables that is

expected to increase from 325 million in 2016 to more than 800 million in 2020 [110].

What drives these quantiﬁed selves to gather all this information? Choe et al. [38]

interviewed 52 enthusiastic quantiﬁed selves and identiﬁed three broad categories

of purposes, namely to improve health (e.g. cure or manage a condition, achieve

a goal, execute a treatment plan), to enhance other aspects of life (maximize work

performance, be mindful), and to ﬁnd new life experiences (e.g. learn to increasingly

enjoy activities, learn new things). A similar type of survey is presented in [51] and

considers self-healing (help yourself to become healthy), self-discipline (like the

rewarding aspects of the quantiﬁed self), self-design (control and optimize yourself

using the data), self-association (enjoying being part of a community and to relate

yourself to the community), and self-entertainment (enjoying the entertainment value

of the self-tracking) as important motivational factors for quantiﬁed selves. They refer

to these factors as “Five-Factor-Framework of Self-Tracking Motivations”.

While Gimple et al. [51] study the goals behind the quantiﬁed self, Lupton [83]

focus on what she calls modes of self-tracking and distinguishes between private and

pushed self-tracking, the latter referring to situations in which the incentive to engage

in self-tracking does not come from the user himself but another party. This being

said, not only users themselves are interested in the data generated within the context

of the quantiﬁed self. Health and life insurances come to one’s mind immediately,

4 1 Introduction

they love to know as much as possible about the current health status and lifestyle

of a potential customer before underwriting an insurance contract. For insurance

companies, leveraging self-tracking data for personalized offerings is a natural next

step to questionnaire based assessments that currently employed. Insurers do not

have to force their customers to share their data, but can set ﬁnancial incentives

to do so. Besides insurances and health providers, other companies are also keen

to tap into this data source. Companies, e.g. from the recreation industry, like to

understand user behavior and location to target their offerings. Only recently, “the

workplace has become a key site of pushed self-tracking, where ﬁnancial incentives

or the importance of contributing to team spirit and productivity may be offered for

participating” [83].

Since self-tracking data can be misused or used in a way that is not fully in the

interest of a person, it is not surprising that users state the loss of privacy as their

main concern in this context. For example, in 2013 it was reported that a supermarket

chain in the UK used wearables to monitor their employees who in return (and again

not surprising) felt a lot of pressure. As said before, user proﬁling with respect to

health and ﬁtness behavior will help companies to personalize their offerings. For

some users this might be beneﬁcial, others might be excluded as customers, as is

obvious in the insurance and ﬁnancial industry. Another very sensitive piece of the

quantiﬁed self data is location that can be abused for criminal purposes but also to

increase control by public authorities.

We are aware that an intensive, open, and broad discourse on self-tracking is

needed that puts the interest of individuals ﬁrst. However, to discuss these risks,

personal concerns, and also the opportunities that come with the quantiﬁed self for

individuals and companies is far beyond the more technical and methodological

perspective of our book. A good starting point for this discussion is the book by Neff

and Nafus [89].

1.2 The Goal of this Book

Now that we know more about the quantiﬁed self, what do we seek to achieve with

this book? As you might have noticed, the quantiﬁed self can and will most likely

result in a huge amount of data being collected about individuals. An immediate

question that pops up is how to make sense of this data. Even enthusiasts such as

Arnold will not be able to oversee it all, and might miss valuable information. This

is where machine learning comes into play. Many deﬁnitions of machine learning

exist. In our case, we deﬁne machine learning as follows:

Deﬁnition 1.3 Machine learning is to automatically identify patterns from data.

This book aims at showing how machine learning can be applied to quantiﬁed self

data; speciﬁcally to automatically extract patterns from collected data and to enable

a user to act upon insights effectively, which in turn contributes to the goal of the

1.2 The Goal of this Book 5

user. Let us make this a bit more concrete for our two fellows Arnold and Bruce by

illustrating potential situations and questions:

• Advising the training to make most progress towards a certain goal based on past

training outcomes

• Forecasting when a speciﬁc running distance will be feasible based on the progress

made so far and the training schedule

• Predict the next blood glucose level based on past measurements and activity levels

• Determine when and how to intervene when the mood is going down to avoid a

spell of depression

• Finding clusters of locations that appear to elevate one’s mood

All these questions could be answered by extracting patterns from historical data.

An observant reader might ask at this point whether this is yet another book

in the area of machine learning among many others. The data from the quantiﬁed

self does however pose its own challenges, which requires dedicated algorithms and

data preparation steps. We will precisely focus on this area and take a more applied

stance. For more theoretical underpinning of algorithms the reader will be referred

to fundamental machine learning books such as Hastie et al. [57] and Bishop [18].

So what are the unique characteristics of machine learning in the quantiﬁed self

context? We identify ﬁve of them: ( 1) sensory data is noisy, (2) many measurements

are missing, (3) the data has a highly temporal nature, (4) algorithms should enable

the support of and interaction with users without a long learning period, and (5) we

collect multiple datasets (one per user) and can learn across them. Each of these

issues will be treated in this book. Note that the approaches we introduce here are

not limited to the development of applications for quantiﬁed selves, but that are

also relevant for a broader category of applications, such as predictive modeling for

electronic medical record data (think of a patient lying at the ICU for example).

1.3 Basic Terminology

Before explaining the formal notation used throughout this book, we will introduce

some terminology ﬁrst. This is by no means meant to be complete, but will provide

a basic vocabulary that we can build upon. We will start with the introduction of

basic terms to describe aspects of data, followed by some basic machine learning

terminology.

1.3.1 Data Terminology

Datasets encompass different attributes such as the heart rate of a person or the num-

ber of steps per day. The most elementary part of data is in our case a measurement,

which is deﬁned as follows:

剩余238页未读，继续阅读

WebDynpro

粉丝: 2
资源: 13

量化自我：机器学习在感官数据中的艺术与行为洞察

system-level-RUL uncertainty

Quantified AG公司是干什么的

LQT问题的强化学习方法

weixin286基于SSM框架的童装购买平台微信小程序+ssm.rar

基于Spring框架的Java Web文章CRUD应用.zip

springboot基于Android的酒店预订系统App毕业论文.docx

weixin271松江大学城就餐推荐系统设计与实现+ssm.rar

基于java的保密信息学科平台系统设计与实现.docx

基于java的图书电子商务网站的设计与实现.docx

最新资源