Click the blue text above to follow us
Professor Zhu Zhiting was born in August 1949 in Quzhou, Zhejiang Province. He graduated from the Mathematics Department of East China Normal University in 1974 and is currently a tenured professor and doctoral supervisor in Educational Technology at East China Normal University; he is also the Dean of the School of Online Education and the Director of the Research Center for Educational Informatization Systems Engineering at the same university.
Abstract: Human deep learning is a tangible development of the 21st-century learning framework, revealing a high degree of alignment between deep learning and smart education. Based on this, a capability iceberg model of deep learning in the field of smart education is constructed from the dimensions of cognition, self, and interpersonal relationships: from cognition to self, the capabilities gradually internalize; from cognition to interpersonal relationships, the capabilities gradually aggregate. On this basis, development strategies for human deep learning are formulated: supporting deep learning with a smart learning environment, empowering deep learning with cultural wisdom, data wisdom, and teaching wisdom, guiding deep learning with smart teachers and adaptive prescriptions, and verifying the effectiveness of deep learning with smart assessments, aiming to pave the way for the joint development of deep learning and smart education.
Authors | Zhu Zhiting, Peng Hongchao
Source | “Chinese Journal of Education” 2017, Issue 5 Note | This article has been abridged from the original
Since 1970, with the economic changes brought about by technology and globalization, the demand for employees with conventional skills has sharply declined, while the demand for talents with complex thinking and communication skills has surged. University teachers have found that freshmen indeed lack critical thinking and problem-solving abilities upon enrollment. The gap between social demand and talent capabilities has prompted the rise of 21st-century learning. As an advanced development of 21st-century learning, deep learning has garnered attention from governments, schools, social institutions, and the media. The 2014 Horizon Report pointed out that “more and more school leaders are beginning to realize its value in formal learning environments,” and “the pursuit of deep learning” has become a recent trend driving the application of educational technology. Currently, deep learning has shown a new normal in teaching and learning. In the technical field, machine deep learning has also become a buzzword in recent years. In this regard, this study provides a deeper interpretation of deep learning in the fields of technology and education, constructs the capability iceberg model of deep learning in the realm of smart education, and formulates development strategies for human deep learning under the concept of smart education, aiming to position deep learning as the core pillar and new direction of smart education, better serving the cultivation of intelligent talents.
01Deep Learning in the Technical Field(1) The Relationship Between Artificial Intelligence, Machine Learning, and Deep LearningArtificial intelligence is the intelligence exhibited by artificial products, aiming to enable machines to demonstrate human-like intelligence, ultimately allowing machines to achieve self-awareness and possess perception, thinking, and action capabilities comparable to or surpassing those of humans. Learning is an important intelligent behavior of humans, while machine learning is a technology that simulates or implements human learning activities using machines, serving as a pathway to artificial intelligence. Through machine learning, machines can acquire new knowledge or skills, thereby continuously improving their performance and achieving artificial intelligence. Machine learning encompasses numerous methods, among which artificial neural networks (Artificial Neural Network, ANN) are quite popular. Deep learning is a new domain of machine learning, named for the high number of hidden layers in artificial neural networks; it is an efficient technology for implementing machine learning. In contrast to deep learning is (machine) shallow learning (Shallow Learning), which often involves machine learning in neural networks containing only 1-2 hidden layers. (2) The Concept of Deep Learning The idea of machine learning is to acquire data through sensors, followed by preprocessing, feature extraction, feature selection, and then reasoning, prediction, or recognition. The three intermediate steps are feature representation, which is key to the effectiveness of machine learning and has been a persistent challenge in the field of artificial intelligence. Neurobiologists David H. Hubel and Torsten N. Wiesel found that the signal processing of the visual system is hierarchical, providing a biological basis for the layered structure of deep learning neural networks. In such a hierarchical neural network, each hidden layer gradually extracts and abstracts features from the data received from the input layer (the output of this layer becomes the input for the next layer), thereby obtaining high-level semantics. Therefore, deep learning is a process from the concrete to the abstract (Figure 1).
Figure 1 The Concept of Deep LearningBig data itself has no inherent value; its value is excavated. In deep learning, this excavation process is a journey from the concrete to the abstract. As shown at the bottom of Figure 1, during the training phase of image recognition, the data input from the training set consists of pixel data from images; at this point, it is challenging to perform image recognition, necessitating a gradual abstraction to transform pixels into local edges or contours within image fragments. As the level of abstraction increases, local edges combine into local shapes, followed by parts of the image, and finally abstracting into the entire image. In this gradual abstraction process, the lower-level concrete features combine to form higher-level abstract features; the higher the level of abstraction, the clearer the features and inherent semantics of the image, ultimately increasing the probability of successful image recognition. This process from concrete to abstract enables the extraction of image features and their inherent semantics (data value). (3) Educational Insights from Machine Deep Learning From the perspective of teaching and learning, the conversion process from the training set in the human world to the feature representation in the data world is an internalization process, while the transformation from the feature representation in the data world to the results in the human world is an externalization process. Ideally, the training set and the resulting output are equivalent, i.e., result = f(training set), so that every layer except the input layer is a different representation of the original information, thereby achieving layered representation of the original information through deep learning. However, the concept of “information loss at each layer” in information theory indicates that this is difficult to achieve, which aligns with the idea in educational communication theory that “information is distorted during transmission,” suggesting that machine learning and human learning share many similarities. Machine deep learning constructs deep neural networks within machines, enabling them to possess the conditions for abstract thinking. By training the neural networks, they exhibit human-like intelligence. Undoubtedly, the human neural network is far more complex than the machine’s artificial neural network, with a significantly greater number of hidden layers (depth). Therefore, humans are better equipped for deeper learning, which is also a condition for the development of smart education. The goal of machine deep learning is to achieve artificial intelligence through machine learning, thereby better assisting humans in solving real-world problems. Thus, from the perspective of teaching and learning, the purpose of machine learning is knowledge transfer. Therefore, when we mention artificial intelligence in the educational context, we reflect: since humans can teach machines deep learning, why can’t we teach children deep learning in schools?02The Rise of Deep Learning in Education(1) The Wave of Deep Learning Movement in Education We are pleased to see that, similar to the field of artificial intelligence, there has also been a wave of deep learning (Deeper Learning, Deep Learning) in the field of education internationally. In 2010, the William and Flora Hewlett Foundation launched a strategic plan for deep learning. The long-term goal of this plan is that by 2025, 80% of American students will be engaged in deep learning; in the short term, the plan aims to ensure that 8 million students (approximately 15% of the K-12 public school population) are taught deep learning skills by 2017. Additionally, organizations such as the Asia Society and other social institutions have collaborated to promote deep learning experimental schools across the United States (now in 41 states), with over 500 schools, more than 10,000 teachers, and over 220,000 students engaged in deep learning. In 2015, the National Association of State Boards of Education released a document designating deep learning as a national policy for 21st-century education in the United States. (2) Definition and Capability Framework of Deep LearningWhat does deep learning really mean? What is its significance for the reform and development of 21st-century education? The National Research Council (NRC) of the United States organized a group of prominent experts for in-depth discussions, resulting in the 2012 report “Learning for Life and Work: Developing Transferable Knowledge and Skills for the 21st Century,” which set the tone for deep learning, defining it as a learning process that enables students to apply what they have learned in one context to new contexts (i.e., transfer). The product of deep learning is transferable knowledge, including content knowledge in a specific field, as well as the knowledge of how, why, and when to apply this knowledge to answer questions and solve problems. Specifically, deep learning encompasses three main areas and six competencies (Table 1, integrating the perspectives of NRC and the Hewlett Foundation). Among them, the ability to master core academic content refers to the ability of students to apply knowledge in different contexts based on their understanding of subject knowledge; critical thinking and complex problem-solving ability refers to the ability to use tools and technologies to collect core knowledge and information to formulate and solve problems; learning to learn ability refers to the ability to monitor and guide one’s own learning; the ability to develop and maintain academic intent refers to the ability to cultivate a positive attitude and belief, thereby enhancing academic resilience, facilitating effective learning, overcoming difficulties, and ultimately achieving goals; collaborative working ability refers to the ability to work with others to identify and create solutions for academic, social, professional, and personal challenges; effective communication ability refers to the ability to clearly organize and express one’s data, discoveries, and ideas. These competencies are key “abilities” for students to succeed in rapidly changing work and civic life. Their effective combination, when applied to mastering core content, will greatly enhance the achievement of expected learning outcomes. Table 1 Deep Learning Capability Framework
(3) Analysis of Deep Learning and Related Standards The Partnership for 21st Century Learning (P21) launched the most renowned “21st Century Learning Framework” (referred to as the “Framework”) in 2007. The Framework outlines the blueprint for 21st-century learning: in terms of learning outcomes, it depicts the core knowledge and skills (21st-century competencies) needed for success in life and work; in terms of support systems, it describes the characteristics of key systems that ensure students master 21st-century competencies. As a guiding document for 21st-century learning, the Framework has a profound impact on subsequent educational reforms and provides a foundation for elevating 21st-century learning to deep learning. Most of the competencies identified in the Framework (such as critical thinking and problem solving, collaboration and communication, creativity and innovation, etc.) are indeed deep learning competencies. The supporting systems outlined in the Framework should “emphasize deep understanding of knowledge, encouraging students to actively engage in solving meaningful problems”; curricula and instruction should “teach 21st-century competencies in the context of core subjects and 21st-century interdisciplinary themes; support innovative learning methods that integrate technology, inquiry-based approaches, problem-based approaches, and higher-order thinking skills” which have also been absorbed by deep learning. The success of workers in modern workplaces depends on a solid scientific foundation. To prepare students for the modern workforce, 23 states in the U.S. took joint action in 2013 to release the “Next Generation Science Standards” (NGSS). NGSS focuses on the content and practices of science, aiming to promote deep learning of scientific knowledge and skills so that students can succeed after graduating from 12th grade. NGSS aligns closely with CCSS in terms of core competencies (Figure 2), thus serving as an effective measure for assessing deep learning competencies. 
Figure 2 Alignment Map of NGSS and CCSSIn summary, the Framework serves as the top-level design for CCSS, NGSS, and deep learning, with CCSS and NGSS being highly aligned. CCSS can serve as the standards for deep learning in the fields of English Language Arts and Mathematics, while NGSS can serve as the standards for deep learning in the field of Science; deep learning is an effective process for achieving CCSS and NGSS. 03The Theory and Practice of Deep Learning in Education(1) Analysis of the Core Concepts of Deep LearningDeep learning involves three aspects of “depth”.First, the depth of learning outcomes, manifested as high-level abilities in cognition, self, and interpersonal relationships, which are the reserves of capabilities that students need to successfully solve problems in higher education, life, and work. To cultivate these abilities and ensure effective transfer, compatible learning methods are required. Second, the depth of learning methods, manifested as the resolution of complex problems (rather than mere knowledge transmission); deep learning methods are diverse, such as inquiry-based learning and project-based learning, but regardless of the method, they are all problem-solving oriented. To facilitate students’ successful completion of deep learning, they need to actively engage in the process. Third, the depth of learning participation, which is the foundation of deep learning. The human neural network can also be artificially divided into input layer, hidden layers, and output layer. Deep participation can encourage more hidden layers to engage in “training,” thereby achieving higher-level abstraction and uncovering deeper meanings (externally manifested as students’ enhancement from memorization and understanding to thinking and creation). From this perspective, deep learning in education shares common concepts with deep learning in technology (Figure 3). However, the training set in the former consists of exercises and projects containing knowledge, and the ultimate goal is to develop human wisdom. 
Figure 3 Deep Learning in Technology and Education In summary, the core concept of deep learning is “to promote deep participation, cultivate high-level abilities, and learn for transfer.” Its distinction from (educational) shallow learning (Shallow Learning, Surface Learning) is shown in Table 2. Table 2 Comparison of Deep Learning and Shallow Learning
(2) The Theoretical Foundation of Deep Learning Although deep learning is a relatively new term, its core concepts are not. In other words, deep learning has a theoretical foundation. Machine deep learning is a stage product of the development of computer science, while human deep learning can be seen as a stage product of the development of learning sciences. One of the missions of learning sciences is to identify and promote deep learning, and related research is among the most prevalent and rigorous. It is an interdisciplinary research field dedicated to scientifically understanding learning, designing and implementing learning innovations, and improving teaching methods. A major theme in learning sciences research is the social context, which posits that knowledge can only be transferred across different contexts when it is immersed in complex, authentic social contexts, which is a prerequisite for deep learning. [23] Following this concept, deep learning emphasizes that it occurs within complex social contexts. The modern theory of deep learning can be traced back to John Dewey’s educational philosophy, which asserts that schools are not only places to acquire content knowledge but also places to learn how to live. The best education is “learning from life, learning from experience,” and the teaching process is a process of “doing” that connects the learning of knowledge and abilities with activities in life. Students will inevitably thrive in an environment that allows them to experience and interact with the curriculum, and all students should have the opportunity to participate in their learning. Dewey’s idea of “learning by doing” serves as the basis for the design of deep learning activities, linking students with social contexts and connecting the outputs and inputs of their neural networks, thereby forming a “closed loop” (Figure 4). It is within this iterative cycle of the “closed loop” that students abstract layer by layer, forming deep learning capabilities. 
Figure 4 The Theoretical Foundation of Learning by Doing in Deep Learning Internally, in addition to the previously mentioned neural network mechanism (from concrete to abstract), the theoretical foundation of deep learning also includes the Five Key Principles of Brain-Friendly Rehearsals. First, what is done is what is learned: students remember 14% of what they hear, but can remember 92% of what they teach to others (i.e., what is done). Second, the brain likes to make connections through diverse and novel things: unexpected or unusual events or activities can stimulate and connect synapses, so deep learning should be creative rather than routine. Third, action involves more brain regions (50% of brain cells), which can internalize deep learning: any kinesthetic connection can enhance the brain; action is the foundation of intellectual understanding, consolidating learning and facilitating working memory or procedural memory to enter long-term memory. Fourth, emotional stimulation is a prerequisite for deep learning: one of the main functions of the brain is to discard useless information, and emotional stimulation lets the brain know that what is being done is valuable. Fifth, an appropriate risk of failure can enhance brain engagement and deep learning: deep learning requires creating a safe, supportive environment where the risk of failure is acknowledged and becomes part of the learning process, thus enhancing motivation and memory through an appropriate level of “concern (beneficial anxiety).” The theory of deep learning can be traced back to classical Chinese educational wisdom, such as the saying in the “Book of Rites: Doctrine of the Mean,” Chapter 19: “Extensively learning, critically questioning, cautiously thinking, clearly distinguishing, and earnestly practicing.” In Xunzi’s “Ruxiao Pian, Chapter 8”: “Not hearing is not as good as hearing; hearing is not as good as seeing; seeing is not as good as knowing; knowing is not as good as doing.” In Wang Yangming’s “Record of Teaching,” Volume One: “Extensively learning, critically questioning, cautiously thinking, clearly distinguishing, and earnestly practicing are all ways to seek precision and unity.” How to carry forward the wisdom of classical Chinese education in the new era is an important mission for contemporary educators. (3) Practical Exploration of Deep Learning Action theory posits that if a culture of “mutual trust and respect between educators and students; and being accountable for each other’s success as learners” is established, coupled with teachers as professionals in collaborative communities, then teachers can design or adjust meaningful learning experiences for students, leading students to intentionally practice deep learning skills, systematically acquire and apply knowledge and abilities. This will enable students to graduate with the capability to know how, why, and when to apply content knowledge, possessing a set of non-cognitive skills to solve challenging problems in college, careers, and life. The four elements of action theory are shown in Figure 5, and these four elements are indispensable and must be established sequentially. 
Figure 5 The Four Elements of Action Theory in Deep Learning Our team, while researching smart teaching models, borrowed the concept of flipped classrooms to propose the Flipped Classroom 2.0 model, practicing the creativity-driven learning philosophy and methods (Figure 6), which aligns closely with deep learning. Creativity-driven learning shifts from a conventional approach of “memorization” to one centered around “creation.” In this deep learning method, the starting point and destination are both “creation”; students learn to achieve a specific creative task, and in the process of creation, they learn what they need, engage in necessary activities, and avoid the pitfalls of remaining at low cognitive levels. In the creativity-driven learning method, teachers need to assume four roles: “facilitator,” “promoter,” “guide,” and “evaluator.” In the early stages of learning, the first two roles are primary, while the latter two take precedence in the later stages. Currently, this method has been implemented in experimental schools for this project through the flipped classroom format, aiding in the cultivation of intelligent talents. 
Figure 6 Creativity-Driven Learning04Deep Learning in the Context of Smart Education(1) The Alignment Between Deep Learning and Smart Education As an enhancement of 21st-century learning, deep learning not only requires mastering core subject content but also emphasizes the cultivation of higher-order abilities and qualities such as critical thinking, problem-solving, and academic intent. Smart education (Smarter Education, SerE) cultivates intelligent talents who are adept at learning, collaboration, communication, judgment, creativity, and solving complex problems; they not only master foundational knowledge but also possess practical skills, good character, and pragmatic creativity. Therefore, deep learning and smart education are highly aligned in talent cultivation, both aiming to nurture new talents that develop knowledge, skills, abilities, and qualities in a balanced manner (see Table 3). The ultimate goal of deep learning is to prepare students for survival and development in college, life, and the workplace (belonging to the capability of doing). Therefore, deep learning places great emphasis on cultivating high-level capabilities and their transferability needed in these three areas. Smart education not only requires students to have the ability to survive and develop but also demands that they possess good thinking qualities and deeper creative potential (belonging to the capability of doing well). Thus, the capability level of intelligent talents shown in Table 3 is higher than that of deep learning capabilities, making the cultivation of intelligent talents even more challenging. The concept of deep learning “cultivating high-level capabilities and learning for transfer” is precisely an effective pathway and essential need for cultivating intelligent talents. It is precisely the high alignment between the two and the aforementioned essential needs that make deep learning a core pillar of smart education, supporting the dream of cultivating intelligent talents.Table 3 Alignment Between Deep Learning Capabilities and Intelligent Talent Capabilities
(2) The Deep Learning Capability Model in Smart EducationAs mentioned above, deep learning can serve as the core pillar of smart education, but smart education also imposes higher demands on it, and current society has new requirements for intelligent talents (such as humanistic awareness, etc.). In response, the author revised the framework for intelligent talents and proposed it as the deep learning capability model in smart education (Figure 7). This model is an iceberg model, where the deeper the capability, the more difficult it is to monitor and cultivate. From a neuroscience perspective, this is because the deeper the capability, the greater the number of hidden layers in the neural network required for higher-level abstraction and forming higher-level meanings. Additionally, the model distributes different capabilities across three dimensions: cognitive, self, and interpersonal, arranged in a gradually increasing order from top to bottom. From cognition to self, capabilities gradually internalize; from cognition to interpersonal, capabilities gradually aggregate.
Figure 7 The Deep Learning Capability Iceberg Model in Smart EducationSpecifically, knowledge and skills primarily target core subject content and basic skills such as reading, writing, and arithmetic, belonging to the “learning to learn” level of capability. Problem-solving ability belongs to the “doing” level, which includes the ability to identify, analyze, and solve problems.At the problem identification level, it not only involves solving existing problems but also possesses the ability to identify potential problems, i.e., insight. At the problem analysis level, based on existing information and materials, it analyzes the original context of the problem, i.e., reasoning ability. At the problem-solving level, based on existing analyses, it makes decisions, formulates solutions, and adjusts and optimizes the implementation of plans based on actual conditions, i.e., execution ability. Problem-solving is a significant indicator of successful transfer, and some scholars have identified problem-solving as the dividing line between deep learning and shallow learning. Communication ability includes two layers: effective communication and social skills. Effective communication refers to the ability to clearly organize and express objective information, subjective thoughts, and ideas verbally, in writing, or through tools, and to understand others’ expressions. Social skills include the ability to read body language, perceive and understand others’ emotions, thoughts, needs, and other inner activities, and to have an engaging demeanor and behavior. Technical literacy aligns with the Framework and includes three layers: media literacy, information literacy, and ICT literacy. The aforementioned four types of abilities are easier to cultivate and monitor, lying above the horizontal line, while the other three abilities can only be partially monitored, with most lying below the horizontal line. Deep thinking belongs to the “thinking” level, intersecting with all other abilities and serving as their foundation, thus occupying a central position. Deep thinking is an advanced enhancement of thinking abilities in shallow learning, encompassing comprehension, analysis, synthesis, generalization, abstraction, reasoning, argumentation, and judgment. In particular, critical thinking (Critical Thinking) has become a recognized core competency across various sectors. Learning ability includes two layers: learning to learn and enjoying learning. Learning to learn belongs to the “learning” level. It has been proven that learning ability has indeed become an essential need for keeping pace with the rapidly changing 21st century. Enjoying learning belongs to the “enjoying learning” level; it is a positive learning attitude and a thirst for knowledge that facilitates self-directed learning and lifelong learning, enabling students to sustain development in life and work. Collaborative and leadership abilities refer to students’ ability to discuss and negotiate team goals, plan, and collaborate, and to demonstrate a certain degree of leadership when necessary. Leadership includes foresight, charisma, influence, control, and decisiveness. Self-awareness mainly relates to self-development, aiming to take care of oneself, including adaptability awareness, self-regulation, safety awareness, health awareness, and self-protection awareness. Imagination and creativity belong to the “creating” level, referring to the ability to have rich and novel imagination, to form innovative ideas, theories, and methods based on these imaginations, and to transform them into valuable spiritual or material products. Such abilities are accompanied by strong perseverance and endurance. (3) Development Strategies for Deep Learning Based on Smart Education1. Support Deep Learning with Smart Learning Environments Smart learning environments possess four key attributes: connectivity, perception, adaptability, and record-keeping. They connect online and offline spaces and individual learning spaces seamlessly, facilitating students’ use of various platforms and devices for deep learning. They can perceive learning scenarios, student locations, and their social relationships, providing technical support for the integration of deep learning with social contexts. They can also adaptively deliver learning materials, services, and tools based on learning goals, individual learner characteristics, and their current learning status, making deep learning activities more personalized. Furthermore, they can record students’ learning data and present it in intuitive and concise forms, enabling precise visualization of students’ deep learning. 2. Empower Deep Learning with Cultural Wisdom, Data Wisdom, and Teaching Wisdom The cultural wisdom of human-machine collaboration is reflected in the use of technological tools (symbolic products), diverse behavioral activities of technological empowerment (collaboration, openness, sharing, interaction), and the resulting changes in ideological values. The themes of the 21st century, as well as the demands of universities, life, and work, are all manifestations of cultural values. Education is essentially about cultural transmission and development, and the capabilities of intelligent talents are merely means of transmission and development; thus, cultural wisdom can guide the sustainable and healthy development of deep learning. Utilizing big data mining and learning analytics technology, recorded learning data can transition from knowing nothing (Know-Nothing, data itself has no inherent value) to knowing what (Know-What), knowing how (Know-How), knowing why (Know-Why), and ultimately achieving the state of knowing best (Know-Best). This is data wisdom. It can support deep learning, providing decision-making basis for teachers and students. With technology taking over rule-based, repetitive, and monotonous tasks, teachers will have more time and energy to focus on emotionally engaging, inspiring, and creative teaching work. Additionally, each teacher has their unique expertise, and when teachers with different specialties come together, they can form an excellent teaching team, greatly amplifying teaching wisdom to serve the design of deep learning. It is particularly important to note the issue of creating a learning culture within cultural wisdom. The core concepts and goals of deep learning dictate that it requires the establishment of a learning culture where “educators and students mutually trust and respect, and are accountable for each other’s success as learners”; a culture of “sharing, collaboration, unity, and mutual assistance” needs to be created. To this end, our team has developed a three-layer structure of culture (ideological values, action methods, symbolic products) to guide the establishment of this culture: when creating culture, first identify the core ideology and values (such as trust, sharing, etc.), then determine which behaviors can reflect these ideological values, and finally select the necessary methods of action, producing some products like posters and promotional materials that can induce these behaviors. Once the culture of deep learning is established, cultural wisdom can determine the “direction” of deep learning, data wisdom can determine the “decisions” of deep learning, and teaching wisdom can determine the “actions” of deep learning. 3. Guide Deep Learning with Smart Teachers and Adaptive Prescriptions As seen in Figure 7, most deep learning capabilities lie below the horizontal line, making them difficult to monitor and cultivate. Therefore, teachers need to break away from conventional roles and evolve into smart teachers. Smart teachers primarily include roles such as teaching facilitators, assessment specialists, activity coaches, instructional designers, data analysts, and resource engineers. In deep learning, each teacher only takes on their strongest role and collaborates with other teachers to create a team of advantageous smart teachers. Specifically, smart teachers who directly interact with students need to establish trust relationships with students; they need to implement learning goals, tasks, and success standards; they need to help students cultivate interest and motivation during tasks; and they need to provide high-quality feedback and encouragement… To lead and motivate students to succeed in deep learning, teachers need strong leadership qualities, including foresight, charisma, influence, control, and decisiveness. In terms of teaching strategies, our team has developed the “Personalized Learning Adaptive Prescription Model based on Micro-Cultural Models” (hereinafter referred to as adaptive prescriptions) to provide guidance for deep learning, especially in large classes in China. Adaptive prescriptions establish a pathway from micro-cultural models to class, group, and individual levels. The concept is as follows: first, at the class level, perceive and learn (mainly using differentiated learning) to understand the ideological values contained in the micro-cultural model to solve the common problems faced by 80% of students; then, at the group level (mainly using group research-based learning), address the remaining 10%-20% of common problems faced by students after class-level learning; if individual students still exhibit learning anomalies, then adaptive learning is conducted at the individual level (generally for 5% of students). This adaptive prescription not only establishes a link between micro-cultural models and individual models but also mitigates the drawbacks of purely personalized adaptive learning, which can be costly and energy-intensive. With the guidance of smart teachers and adaptive prescriptions, the likelihood of success in deep learning can be significantly enhanced, addressing the issue of “current project-based learning and inquiry-based learning (both specific methods of deep learning) often being superficial and not achieving expected learning outcomes.” 4. Verify the Effectiveness of Deep Learning with Smart Assessments Smart assessments are data-driven evaluations empowered by technology, characterized by being comprehensive, diverse, multidimensional, subjective, and result visualized, aiming to promote learning and development through assessment. The philosophy of smart assessments is to assess learning, visualize learning, and support learning through assessments. It interprets the substantive meaning represented by measurement variables through construct theory, analyzes the potential value of monitoring data through mathematical statistics, and utilizes scientific technology to automate and streamline these processes, achieving comprehensive assessment, in-depth exploration, and result visualization. Deep learning capabilities are high-order abilities that are difficult to monitor and assess, while smart assessments can serve as a strategy to address this challenge, verifying the effectiveness of deep learning. Specifically, smart assessments for deep learning need to establish a complete assessment continuum, in which formative assessments are continuous, while the proportion of mid-term and summative assessments is significantly reduced. Within the assessment continuum, various predictive ability assessment models are included. One important assessment model is the deep learning ability model (also known as the diligence learning ability model). Regarding deep learning ability, the American Learning Emergence research team analyzed 15 years of relevant data to derive the eight essential elements of learning ability: proactive awareness, meaning construction, creativity, curiosity, sense of belonging, collaboration, expectation and optimism, and openness. This provides a basis for formulating the learning ability assessment model. Honestly, constructing a deep learning smart assessment system is a tremendous challenge and a significant problem that our team needs to tackle in the future. (End) 
