我正在尝试通过命令创建一个带有页码的 para id 作为单独的文件
\def\paraopenid#1{\label{#1}\immediate\write\paraID{<ParaID="#1" StartPageNumber="\getpagerefnumber{#1}"}
当我使用或获得正确的输出时，我面临的问题是\label出现在页面顶部时，\getpagerefnumber会产生错误的页码（例如，MWE 中的最后一段） 。\clearpage\pagebreak
如何通过自动化方式获取正确的页码而无需提供\clearpage/pagebreak？
平均能量损失
\documentclass{book}
\usepackage{showframe}
\usepackage{hyperref}
\newwrite\paraID
\AtBeginDocument{\immediate\openout\paraID\jobname.paraID}
\def\paraopenid#1{\label{#1}\immediate\write\paraID
{<ParaID="#1" StartPageNumber="\getpagerefnumber{#1}"}}
\def\paracloseid#1{\label{#1}\immediate\write\paraID
{EndPageNumber="\getpagerefnumber{#1}"/>}}
\begin{document}
\paraopenid{p1}Smart health care assistants are designed to improve the
comfort of the patient where smart refers to the ability to imitate the
human intelligence to facilitate his life without, or with limited, human
intervention. As a part of this, we are proposing a new Intelligent
Communication Assistant capable of detecting physiological needs by
following a new efficient Inverse Reinforcement learning algorithm designed
to be able to deal with new time-recorded states. The latter processes the
patient's environment data, learns from the patient previous choices and
becomes capable of suggesting the right action at the right time. In this
paper, we took the case study of Locked-in Syndrome patients, studied their
actual communication methods and tried to enhance the existing solutions by
adding an intelligent layer. We showed that by using Deep Inverse
Reinforcement Learning using Maximum Entropy, we can learn how to regress
the reward amount of new states from the ambient environment recorded
states. After that, we can suggest the highly rewarded need to the target
patient. Also, we proposed a full architecture of the system by describing
the pipeline of the information from the ambient environment to the
different actors.\paracloseid{p1-end}
\paraopenid{p2}For LIS patients, there is only one \textit
{Alternative Communication} method that is eye communication. Patients count
on their caregiver to understand them. The latter relies on eye-control
methods which are \textit{eye blinking} and \textit{eye gaze}. There are two
critical steps in the eye-tracking process: eye detection that consists in
localizing the eye position in the captured image, and gaze estimation
(also called gaze detection) that consists in estimating where the user is
looking (Tatler, Hansen, and Pelz, 2019). Gaze tracking techniques are
integrated into Alternative and Augmentative Communication (AAC) systems.
They permit the user to communicate with others without talking. That is why
such systems in persons with speaking disabilities' life represent a vital
need (Light et al., 2019). AAC research field is concerned with assisted
communication for people that have a speech impairment. One of AAC
research's main aims is to investigate the possibilities of improving
communication skills for non-speakers through the use of communication There
are several AAC systems that consider eye-tracking techniques. For instance,
FEMA (Chareonsuk et al., 2016), which is a software designed for Amyotrophic
Lateral Sclerosis (ALS) patients, enables them to use the computer as
ordinary people using face and eye movements. They proved that the accuracy
rate of left-click, right-click, double-click, scroll down, left movement
and right movement is more than 80\%. Loja et al. proposed an AAC
architecture that helps researchers develop an adaptive communication
environment at a low cost (Loja et al., 2015).\paracloseid{p2-end}
\paraopenid{p3}Besides, there are some commercial tools that are designed for
similar type of users, and some of them have a patent that gives them the
authority conferring their license or title for a set period like Tobii
which is a leading eye-tracking technology by gaining many
patents.\paracloseid{p3-end}
\paraopenid{p4}Our system's intervention commune in facilitating the patient's
life by analyzing and trying to learn from the history of his behavior. The
contribution consists in adding an intelligent layer to the AAC systems that
are actually used by the target patients. It records, analyzes, and predicts
the need depending on the history, knowing that when the patient approves
the proposed suggestion, the system records the time and the state of the
environment. To do so, we studied the two sides of RL. On the first side, RL
seeks to learn the optimal behavior based on experiences, and IRL seeks to
best understand and represent the observed behavior by learning the
corresponding reward function. IRL introduces a new way of learning policies
by deriving expert's intentions, in contrast to directly learning policies,
which can be redundant and have poor generalization ability.\paracloseid
{p4-end}
\paraopenid{p5}To choose the appropriate method for this problem, we reffered
to (Coronato et al., 2020) who proposed a useful guideline for the
application of RL to the healthcare problems. They provide some indications
to the designer in order to help him in choosing the appropriate RL method.
We concluded that we should adopt the Inverse Reinforcement Learning
(IRL) paradigm because of the following five facts: (1) The problem involves
a multi-step decision process. Patient's needs are expressed sequentially.
There is no complete and accurate model of the environment. Patient's
environment is difficult to present. (2) The actions and states can be
presented as arrays. (3) We cannot let the agent interact with the real
environment because we cannot propose all the possible needs at each time
step, and let the patient answer through trial-and-error. Also, we cannot
conclude his preference since he can change them depending on his mind or
mood. (4) We are not able to model the reward function <italic>R</italic>
for every possible state features. After studying the IRL variants, we
decided to adopt the Deep Inverse Reinforcement Learning using Maximum
Entropy since multiple reward functions can explain the patient's behavior,
and Maximum Entropy variants are the best to represent the problem when
there are multiple reward functions that can explain the expert's behavior
(Arora and Doshi, 2021).\paracloseid{p5-end}
\paraopenid{p6}Maximum Entropy methods are useful when multiple reward
functions can explain the expert's behavior (Arora and Doshi, 2021).
Consequently, we decided to adopt it because multiple reward functions can
explain the patient's behavior.\paracloseid{p6-end}
\paraopenid{p7}To enhance patient care, we suppose that the patient follows
his healing process in a smart environment that employs ambient
intelligence (AmI) technologies. For instance, (Kartakis et al., 2012)
designed a smart patient room with a user interface development to assist
both patients and medical staff in order to enhance Health Care Delivery
through Ambient Intelligence Applications. Importantly, AmI supports the
pervasive diffusion of intelligence in the patient environment thanks to
wireless technologies (e.g., Zigbee, blacktooth, RF, WiFi, and intelligent
sensors). Many other research works focused on enhancing health care
services through AmI applications. We took the problem from the data
scientist's perspective. What caught our attention is data delivered from
those intelligent environments and how we can use it to ameliorate the
patient life. Smart patient rooms can record data related to the patient
(e.g., recording the heart beating, blood pressure, pulse oximetry, etc.)
and related to his environment (e.g., temperature, bed, tv, window, door,
visitor detecting, etc.).In our work, we consider six features that describe
the patient's room in order to catch the environment's state, as mentioned
in\paracloseid{p7-end}
\paraopenid{p8}A good deal of psychological research criticized Maslow's
classification of needs. Some of them reported that needs doesn't really
follow a hierarchy (Wahba and Bridwell, 1976). On the other hand, some other
researchers were influenced by Maslow's contributions and considered them as
a big shift in psychology. For instance, (Tay and Diener, 2011) put the
hierarchy to test and discovered that the fulfillment of the needs was
strongly correlated with happiness. We concluded that satisfying human needs
is not an easy task because they differ from one person to another. On the
first side, we have needs that depend on the studied person, his background,
center of interest, hobbies, relationships, etc. On the other side, we have
LIS patients who only want to accomplish the basic survival needs, and to
balance between those two sides, we can treat the basic survival needs that
belong to the lowest level by focusing on the \textit
{physiological needs}\paracloseid{p8-end}
\end{document}