智能系统中的AI与算法：2018年第七届在线计算机科学会议

需积分: 10 87 浏览量更新于2024-07-18 1 收藏 44.23MB PDF 举报

"Artificial Intelligence and Algorithms in Intelligent Systems Conference 2018 II" 是一个会议论文集，属于"Advances in Intelligent Systems and Computing"系列的第764卷，由Radek Silhavy编辑。这个会议是第7届在线计算机科学会议的一部分，主要关注人工智能和算法在智能系统中的应用。此资源提供了英文无水印原版的PDF格式，可在多种PDF阅读器和浏览器中打开。内容涵盖人工智能领域的理论、应用和设计方法，涉及到工程、自然科学、计算机与信息科学等多个学科。在"Artificial Intelligence and Algorithms in Intelligent Systems"这一主题中，研究者们探讨了以下关键知识点： 1. 人工智能：人工智能是模拟、扩展和增强人类智能的科学与技术，包括机器学习、深度学习、自然语言处理、计算机视觉等子领域。这些技术在自动驾驶、智能家居、智能助手等领域有着广泛的应用。 2. 算法：算法是解决特定问题的明确步骤，对于人工智能至关重要。会议可能涵盖了各种优化算法（如遗传算法、粒子群优化）、机器学习算法（如支持向量机、决策树、神经网络）以及数据挖掘和分析算法。 3. 智能系统：智能系统是能够自主学习、适应和决策的系统，可以模仿人类智能的行为。这类系统通常包括嵌入式系统、分布式系统、自适应系统等，它们结合了硬件和软件来实现复杂任务。 4. 计算智能：计算智能包括模糊系统、神经网络、进化计算等软计算方法。这些方法在处理不确定性和非线性问题时特别有效。 5. 社交智能：研究如何让计算机理解和模拟人类的社交行为，包括协作、沟通和决策制定。 6. 环境智能：环境智能涉及使设备能感知并适应其周围环境，常用于智能家居和智能城市项目。 7. 计算神经科学：研究大脑的计算特性，以发展更高效的计算机模型。 8. 人工生命：模拟生物系统和进化过程，以理解生命的本质，并应用于系统设计。 9. 虚拟世界和社会：探讨虚拟现实、增强现实以及在线社区的构建和影响。 10. 认知科学与系统：研究人类认知过程的计算机模型，以便构建更智能的系统。 11. 感知：关于机器如何获取和理解来自外部世界的感官输入，如图像识别和声音识别。通过这个会议，参与者有机会了解最新的研究成果，交流思想，并推动人工智能和算法在智能系统领域的创新和发展。对于研究人员、学者和行业从业者来说，这是一个宝贵的资源，可以帮助他们跟上这一快速发展的领域的步伐。

which integrate the situational context of an event (see [4]). The dimensional models

suggest that emotion can be explained by a dimensional space, such as a 2D space

spanning valence and arousal. Valence describes the extent to which an emotion is

positive or negative, whereas arousal refers to emotional intensity, i.e., the strength of

the associated emotional state.

Since advertising seeks to in ﬂuence the user’s mental functions - ﬁrst of all, per-

ception, behavior, thinking and emotions, the interest in the role of affective factors on

the perception of advertising is steadily growing, and so there is a growing number of

studies that help to understand how advertising is most affected by emotion [5, 6].

The task is made more difﬁcult by known fact that the content can not be applied

indiscriminately, e.g., if one ad engages children, it will not necessarily be suitable for

adults. Moreover, there is a dual perception of the emotions visible in the image [7]as

follows:

(1) Syncretic cognition, which is occurring instinctively, spontaneously and broadly

subjectively. The image is perceived by what it is visually recognizable and

known from before, it causes relevant emotions (e.g., joy, sorrow, fear, etc.); and

(2) Analytical cognition, which is slower. The image is perceived by thinking about

the visible characters, analyzing their meanings. The analytical perception is less

subjective, since other well-known descriptions of the image are used.

While advertisers are often trying to disturb people emotions and thus force them to

engage in an advertised product or product, a user who ﬁrst encounters a positive

impression on a product becomes more inﬂuenced by emotions (syncretic perception)

than rational decisions (analytical perception).

Recently, the researchers began evaluating the neurological correlates of consumer

behavior using the electroencephalography (EEG), electromyography (EMG), galvanic

skin response (GSR) and gaze tracking technology in order to understand

marketing-relevant human behavior [8]. The quantitative data collected and properly

analyzed provides an advantage over the commonly used surveys and quest ionnaires,

which rely on the ability and willingness of the participants to honestly and accurately

report their attitudes and behaviors.

Eye movements, gaze landing sites and areas of interest are know n to be good

predictors of customer behavi or. People spend more time for examining (i.e., ﬁxating

their gaze on) options that they will eventually choose [9]. For example, consumers

spent 54% more time looking at the ads of businesses that they ended up choosing [10].

Unusual brand extension cause disruption of visual processing and result in longer

visual ﬁxation times as evidenced in [11]. The emotional state can be recognized using

gaze tracking technology [12], while different factors such as the emotional state and

gender have inﬂuence on visual processing and understanding [13].

The aim of the paper is to analyze gender and affect related differences in the

perception of visual advertising. The results of experiments using gaze tracking are

presented and analyzed, and the implications for neuromarketing are discussed.

The structure of the remaining parts of the paper is as follows. Section 2 describes

the experiments. Section 3 evaluates the results and concludes.

2 G. Liaudanskaitė et al.

2.3 Methodology

The experiment was performed on the basis of a mixed quantitative and qualitative

research met hodology. To achieve the objective, an eye tracking device and special

software were used. Research was conducted using the methodol ogy presented in [15],

which sought to ﬁnd out how people react to smiling faces in promotional imagery. The

aim is to ﬁnd out whether two groups of subjects of similar composition, which are

shown the same adv ertisements in which only the expression of a person’s face differs,

can notice the difference in the intensity of the reaction.

For each participant the gaze-tracking device was adjusted and calibrated. Each

experiment consisted of 3 stages, while calibration was repeated at the beginning of

each stage. Following the experiments, the raw data was pre-processed and eye ﬁxa-

tions shorter than 60 ms were discarded because of contamination from equipment

measurement noise and blinking effects [16]. The upper threshold of ﬁxations has been

set to 600 ms as recommended in [17].

2.4 Procedure

Before the experiment, the intensity of human emotions seen in each advert isement was

measured using FaceReader software (Noldus, the Netherlands). The program can

recognize the basic facia l expressions: happiness, surprise, sadness, anger, fear, con-

tempt, disgust, and neutrality. The intensity of each emotion is evaluated on a scale

from 0 to 1, where zero means that emotion is not recognized, and one - the emotion is

very strong. The results of emotions measurements are shown in Table 1.

The participants were divided into two equal groups with an equal number of

women and men of the same age, emotional advertisements were displayed, or mod-

iﬁed with the use of additional photos for retouching. One ad was displayed for a

maximum of 30 s, but each respondent was able to switch the ad on when she did not

want to watch it for longer. Numeric data collected during the experiment was the

timeframe for ﬁxation of sight of the software (start time and duration in ms).

Table 1. Face emotion measurements with FaceReader

No. Ad Variant Recognized

emotion

Value Assigned

emotion

1 “Trump for President

2016”

Original Neutral 0.4716 Happy

Modiﬁed Anger 0.2055 Sad

Sad 0.2124

Neutral 0.5408

2 “Charming Chocolate” Original Joy 0.9965 Happy

Modiﬁed Joy 0.1504 Sad

Neutral 0.7603

3 “I am Muslim” Original Neutral 0.9996 Sad

Modiﬁed Joy 0.9999 Happy

4 G. Liaudanskaitė et al.

2.5 Statistical Analysis

Usually, power laws are used to model human visual perception. The distribution of

eye ﬁxation times can be adequately characterized using a two-parameter Weibull

probability distribution [18, 19], and its performance on the eye-gaze data is shown to

be better than other power law based approaches. The probability density function of a

Weibull random variable is:

pdf xðÞ¼



k1

 x=kðÞ

ð1Þ

where k > 0 is the shape param eter and k > 0 is the scale parameter of the distribution.

A value of k > 1 indicates that the failure rate increases with time. This happens if there

is an aging or fatigue [20] process.

Weibull distribution parameters, gamma and beta, highly correlate (between 84 and

93%) with neural responses in the early visual system [21]. Sometimes none of the

considered models represents the data appropriately. Thus, it should be tested how well

the integrated Weibull distribution ﬁts the data. The goodness of ﬁt between experi-

mentally established distribution and real data is evaluated using R

value, which is

calculated as follows:

i¼1





xðÞ

i¼1





xðÞ

ð2Þ

here



x is the mean, and

x is the ﬁtted value.

Having the parameters of Weibull distribution, the mean and standard deviation of

a Weibull random variable can be expressed as follows:

EðXÞ¼kC 1 þ



ð3Þ

and

std XðÞ¼k

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

C 1 þ



 C 1 þ



ð4Þ

here C ðÞis the Gamma function:

C zðÞ¼

z1

x

dx ð5Þ

To evaluate if two data samples belong to the same statistical distribution, a number

of statistical tests can be used.

The Kruskal-Wallis method tests if two or more classes have equal median and

gives the value of p. If the value of p is close to 0 it means that the feature contains

discriminative information.

Analysis of Affective and Gender Factors in Image Comprehension 5

The Wilcoxon rank sum test is used to establish if there is a signiﬁcant difference

between two sample groups using their ranks. If the ranks of the sample groups are

signiﬁcantly a ¼ 0:05ðÞseparated, the test statistic identiﬁes signiﬁcant difference.

The third test statistic is obtained by bootstrapping, a way of estimating statistical

parameters from the sample by performing resampling with replacement. For a

two-sample bootstrap, we independently draw bootstrap samples with replacement

from each sample, and compute the Fisher–Pitman test [22] statistic that compares the

samples. The bootstrap distribution is used for conﬁdence intervals and standard errors.

Given the observed value of the difference (D) between the means of the samples

x ¼ x

; x

; ...; x

ðÞand y ¼ y

; y

; ...; y

ðÞdrawn from two probability distributions X

and Y, the p-value is calculated as follows:

value

1 þ

i¼1

 D

½

B þ 1

ð6Þ

here ½is the Iverson bracket operator, and B is the number of bootstrap resamples, here

we set it to 1000 as suggested in [23].

For calculation of conﬁdence, the percentile intervals are used. The sample values

are arranged in ascending order and numbered sequentially with the smallest value

receiving the number 1, obtaining the ranks of the sample values. For 95% conﬁdence

interval, the 2.5 and 97.5 percentiles are used, which corresponds to 0:025n þ 0:5 and

0:975n þ 0:5 ordered values, where n is the sample size [24].

2.6 Hypotheses

H1. There is no difference between genders in eye ﬁxation times. Hypotheses are:

1ðÞ

: l

male

¼ l

female

; H

1ðÞ

: l

male

6¼ l

female

H2. There is no difference between affect of images in eye ﬁxation times.

Hypotheses are:

2ðÞ

: l

happy

¼ l

sad

; H

2ðÞ

: l

happy

6¼ l

sad

To analyze the interaction effects in population groups, the following

sub-hypotheses are formulated:

H1a. There is no difference between genders in eye ﬁxation times of the positively

valenced images. Hypotheses are:

1aðÞ

: l

happy;male

¼ l

happy;female

; H

1aðÞ

: l

happy;male

6¼ l

happy;female

6 G. Liaudanskaitė et al.

剩余514页未读，继续阅读

yinkaisheng-nj

粉丝: 763
资源: 6231

智能系统中的AI与算法：2018年第七届在线计算机科学会议

Intelligent Systems Technologies and Applications-Springer(2018).pdf

Artificial Intelligence and Algorithms in Intelligent Systems

Artificial Intelligence: A Guide to Intelligent Systems 2nd Edition

Feng, L. (2022). Application Analysis of Artificial Intelligence Algorithms in Image Processing. Mathematical Problems in Engineering, 2022. https://doi.org/10.1155/2022/7382938这是什么

The current research topic in the field of artificial intelligence

artificial intelligence and lisp

string1 = “Artificial Intelligence”, string2 = “Artificial Intelligence”,判断它们的id是否相同？

artificial intelligence and soft computing 17th

Can you tell me some high-quality forums or blogs about artificial intelligence, and attach specific websites or access methods

最新资源