What is AI
AI stands for artificial intelligence, which refers to a technology that enables machines to perform tasks that require intelligence through computer science and technology. These tasks include learning, reasoning, problem solving, and perception, and are often manifestations of human intelligence. The goal of artificial intelligence is to enable computer systems to perform tasks that require human intelligence without the need for human intervention.
The emergence of artificial intelligence can be traced back to the 1950s. In 1956, the concept of "artificial intelligence" was first proposed at a two-month seminar held at Dartmouth University in the United States. In the following decades, the development of artificial intelligence technology experienced several ups and downs, but ultimately failed to enter the public eye. Artificial intelligence has become completely popular and has received widespread attention. In 2016, AlphaGo defeated the world Go champion. This competition made artificial intelligence a topic of discussion in thousands of households, and also set off a widespread learning craze for deep learning.
Related research directions in the field of artificial intelligence
Artificial intelligence can be divided into two types: weak artificial intelligence and strong artificial intelligence. Weak AI refers to systems focusing on one specific task, while strong AI involves having general intelligence similar to humans and being able to perform multiple complex tasks.
Narrow/Weak AI
natural language processing
- What is natural language processing?
Natural Language Processing (NLP) is a branch of the field of artificial intelligence, involving multiple disciplines such as computer science, artificial intelligence, and linguistics. Its main goal is to enable computers to understand, interpret, generate and interact with human natural language.
- What are the main research directions?
Text and language understanding: Enables computers to understand the meaning of human language, including syntax, semantics, and context.
Text generation: Generate natural language text that conforms to grammatical and semantic rules through computers, such as automatic summary, text translation, etc.
Speech recognition: Converting human speech into text form so that computers can understand and process spoken information.
Information retrieval: Retrieve and extract relevant information from large-scale text data to meet user query needs.
Sentiment analysis: Analyze the emotional color in the text and identify the emotional tendency in the text, such as positive, negative or neutral.
Machine Translation: Automatic translation between different languages, enabling computers to translate text from one language to another.
Question Answering Systems: Developing systems that can answer questions posed by users, often requiring understanding the semantics of the questions and extracting answers from relevant text.
- Summarize
The integration of unsupervised models and Sequence to Sequence tasks is a very important progress and development direction. For example, the paper "Unsupervised Machine Translation Using Monolingual Corpora Only" submitted at ICLR 2018 is a representative technical idea, which uses a non-aligned bilingual training corpus collection. Train the machine translation system and achieve better results. This technical idea is essentially very similar to CycleGAN. I believe that this unsupervised model idea will have a lot of follow-up research in 2018. Secondly, how to combine popular technologies such as reinforcement learning and GAN with NLP in the past two years and really play a role is a promising direction. In the past year, exploration in this area has begun and some progress has been made, but it is obvious that this The road has not yet been cleared, and this area deserves further exploration. Thirdly, the Attention mechanism is further widely used and more variants are introduced, such as Self Attention and hierarchical Attention. This trend can be clearly understood from the technical ideas of Google's new machine translation paper "Attention is all you need". In addition, how to integrate some prior knowledge or linguistics-related domain knowledge with neural networks is a popular research trend, such as explicitly introducing information such as the syntactic structure of sentences into the Sequence to Sequence framework. In addition, the interpretability of neural networks is also a research hotspot, but this is not only limited to the field of NLP, but is also a research trend of great concern in the entire field of deep learning.
The rapid development of natural language processing technology enables computers to understand, process and generate human language, bringing important innovations to the fields of human-computer interaction, information retrieval and language understanding. The advancement of this technology provides powerful support for applications such as smart assistants, speech recognition, and automatic translation, allowing computers to communicate with humans more naturally. As more and more researchers devote themselves to the field of natural language processing, innovative products and applications continue to emerge, ranging from social media analysis to intelligent customer service, demonstrating the wide application of this technology. For students who are interested in "making computers understand and use natural language", choosing the direction of natural language processing will be a research field full of challenges and opportunities.
- Laboratory recommendation
Social Computing and Information Retrieval Research Center of Harbin Institute of Technology:
- Website: http://ir.hit.edu.cn/
- Description: The research directions of the Social Computing and Information Retrieval Research Center of Harbin Institute of Technology include five aspects: text retrieval, text mining, language analysis, and cross-language retrieval. The research center uses cognitive psychology and machine learning as its theories and language analysis as its basic research. It takes information extraction, text retrieval, and cross-language/media retrieval as application research, and uses precision search and mining systems as application system platforms. The research center has completed and is currently undertaking more than 40 projects from the National Natural Science Foundation, National 863, international cooperation, and enterprise cooperation.
Tsinghua University Interactive Artificial Intelligence Research Group:
- Website: https://deny.tsinghua.edu.cn/
- Description: The Natural Language Processing and Social Humanities Computing Laboratory (THUNLP) of the Department of Computer Science at Tsinghua University was established in the late 1970s. It is the earliest and most influential scientific research unit in China to carry out natural language processing research. Focusing on natural language processing with Chinese as the core, the laboratory has carried out systematic and in-depth research in large-scale pre-training models, Chinese information processing, machine translation, social computing, smart education and knowledge graphs.
The subject leader of the laboratory is Professor Sun Maosong, and the teaching team includes Professor Liu Yang and Associate Professor Liu Zhiyuan. In recent years, the laboratory has undertaken a series of important research tasks such as national key R&D projects and major projects of the National Social Science Fund, and has established close academic cooperation with companies such as Tencent, Huawei, Alibaba, and Meituan. The Github open source toolkit generated by related results has received tens of thousands of stars.
Fudan University Natural Language Processing Laboratory:
- Website: https://nlp.fudan.edu.cn/
- Description: The Natural Language Processing Laboratory of Fudan University was founded by Mr. Wu Lide, chief professor of Fudan University. It is one of the earliest laboratories in China to conduct research on natural language processing and information retrieval.
Machine Intelligence and Translation Research Laboratory, Language Technology Research Center, Harbin Institute of Technology:
- Website: https://mitlab.hit.edu.cn/
- Description: The Machine Intelligence & Translation Laboratory (MI&T Lab) of the School of Computer Science, Harbin Institute of Technology has been committed to machine translation research and system development since 1985.
Stanford NLP Group:
- Website: Stanford NLP Group
- Description: The NLP group at Stanford University has achieved outstanding research results on various natural language processing tasks, including semantic analysis, question answering systems, machine translation, etc.
MIT Computer Science and Artificial Intelligence Laboratory (MIT CSAIL):
- Website: MIT CSAIL
- Description: MIT CSAIL is a laboratory covering multiple directions of artificial intelligence, including natural language processing. They have extensive research in language understanding, text generation, etc.
Google Research:
- Website: Google Research
- Description: Google's research lab has conducted in-depth research in natural language processing, machine learning and other fields, and has launched a series of influential projects.
Carnegie Mellon University Language Technologies Institute (CMU):
- Website: CMU LTI
- Description: CMU's Language Technology Research Center has a high reputation in the fields of natural language processing, speech recognition and machine translation.
Oxford University Natural Language Processing Research Group (Oxford NLP):
- Website: Oxford NLP
- Description: The NLP research group at Oxford University has conducted in-depth research in language modeling, text mining, etc.
computer vision
Computer vision systems are capable of processing image and video information, but they are often specialized for specific tasks, such as image recognition, object detection, rather than having general visual understanding capabilities.
- What are the main research directions?
Object recognition and detection : For input pictures, the computer can automatically identify the objects in the picture through algorithms, classify them and frame the location of the objects. Technologies that have already been implemented include facial recognition, vehicle detection, etc.
Semantic segmentation : To segment images based on semantics, the computer needs to classify each pixel on the input image to segment out the different objects present in the image. Currently, semantic segmentation plays an important role in fields such as autonomous vehicle driving and medical image analysis.
Motion and tracking : In a given video, the first frame gives the position and size of the tracked object. The tracking algorithm needs to find the position of the tracked object in subsequent videos and should have a certain degree of robustness. Adapt to changes in lighting and other conditions in the video.
Visual question and answer : The user asks questions based on the input image, and the algorithm automatically answers the question based on the content; or the computer automatically generates a text describing the image based on the input image, and no question and answer is performed.
- Summarize
New technologies such as reinforcement learning and GAN have begun to be tried to solve problems in many other image processing fields and have made certain progress, such as Image-Caption, super-resolution, 3D reconstruction and other fields, and attempts have been made to introduce these new technologies. In addition, how to integrate their respective advantages and deeply integrate deep learning and traditional methods has also been the direction of visual processing in the past year. Deep learning technology has advantages such as excellent performance, but it also has shortcomings such as uninterpretable black boxes and weak theoretical foundations. Traditional methods have It is important to combine the two to give full play to their respective advantages and overcome their own shortcomings. Thirdly, weak supervision, self-supervision or unsupervised methods are becoming more and more important in various fields. This is a practical demand. Although deep learning is effective, it requires a large amount of labeled training data, which in turn requires a large amount of The labeling cost is often not feasible in reality. Exploring weak supervision, self-supervision and even unsupervised methods will help to promote the rapid development of research in various fields.
- Laboratory recommendation
Stanford Artificial Intelligence Lab (SAIL)
- Description: Located at Stanford University in the United States, it is an important artificial intelligence research center covering many fields such as computer vision.
MIT Computer Science and Artificial Intelligence Lab (CSAIL)
- Description: MIT's CSAIL is one of the world's largest computer science laboratories, dedicated to advancing cutting-edge research in artificial intelligence and computer vision.
Oxford University Computer Vision and Robotics Group
- Description: Located at the University of Oxford in the UK, it focuses on research in computer vision, machine learning and robotics.
Computer Vision Laboratory, ETH Zurich
- Description: Located in Zurich, Switzerland, it is known for its research excellence in the fields of computer vision and graphics.
UC Berkeley Computer Vision Group (Berkeley Computer Vision Group)
- Description: As part of the University of California, Berkeley, this laboratory conducts in-depth research in computer vision, machine learning, and pattern recognition.
Computer Vision Laboratory, Institute of Automation, Chinese Academy of Sciences
- Description: Located in Beijing, China, this laboratory has achieved significant research results in areas such as computer vision and pattern recognition.
Max Planck Institute for Informatics, Germany
- Description: The institute has world-leading research teams in computer vision, machine learning and computer graphics.
HKUST Department of Computer Science and Engineering
- Description: Located in Hong Kong, this department excels in computer vision and artificial intelligence research.
Strong artificial intelligence (General/Strong AI)
Machine Learning and Deep Learning
What are machine learning and deep learning?
- Machine learning: Let machines have the ability to learn like humans . It specializes in studying how computers can simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their own performance. It is artificial intelligence The core of intelligence.
- Deep Learning: These fields focus on getting computer systems to learn patterns from data, and their wide range of applications and capabilities bring them closer to strong artificial intelligence.
What are the main research directions?
machine learning
- Supervised Learning: Includes classification and regression problems where the model learns based on labeled training data.
- Unsupervised Learning: It mainly studies tasks such as clustering, dimensionality reduction, and association rules, and does not rely on labeled training data.
- Semi-Supervised Learning: Uses partially labeled data and unlabeled data for learning.
- Reinforcement Learning: The study of how an agent learns optimal strategies through rewards and punishments in its interaction with the environment.
- Transfer Learning: Apply knowledge learned on one task to different but related tasks.
- Reinforcement learning (Meta-Learning): enables the model to learn quickly when facing new tasks.
- Ensemble Learning: Integrate multiple models to improve overall performance, such as random forests and gradient boosting trees.
- Kernel Methods: Use kernel functions to map nonlinear problems into high-dimensional space to solve complex pattern recognition problems.
deep learning
- Neural network structure and architecture: Design new neural network structures, such as convolutional neural network (CNN), recurrent neural network (RNN) and transformer (Transformer).
- Deep learning optimization algorithm: Research optimization algorithms such as gradient descent and stochastic gradient descent to improve training efficiency and model performance.
- Application of transfer learning in deep learning: Use the knowledge learned by the deep learning model on one task and apply it to new tasks.
- Generative Adversarial Networks (GANs): Study generative models that generate realistic data samples through adversarial learning.
- Self-supervised learning: Developing models that automatically learn useful representations from unlabeled data.
- Deep reinforcement learning: Combining deep learning with reinforcement learning enables agents to process high-dimensional, complex inputs.
- Interpretability and Robustness: Improve the interpretability of deep learning models, making their decision-making processes easier to understand and improving their robustness to adversarial attacks.
- Automated Machine Learning (AutoML): Study how to automate machine learning processes, including feature selection, model selection, and hyperparameter optimization.
- Multimodal Learning: Processing multiple types of data input, such as images, text, and speech.
Summarize
Machine learning and deep learning are two cutting-edge and practical disciplines. They process data in an intelligent way, providing huge impetus for real life and industrial applications. Learning these two technologies can help you discover patterns in complex data, predict trends, and provide strong support for decision-making.
Laboratory recommendation
Machine Learning Lab Recommendations :
- Stanford Machine Learning Group
Located at Stanford University in the United States, it is committed to promoting cutting-edge research in machine learning and artificial intelligence.
- MIT Machine Learning Group
The MIT Machine Learning Laboratory covers multiple directions of machine learning, including supervised learning, reinforcement learning, etc.
- AMPLab (Algorithms, Machines, and People Lab), University of California, Berkeley
Focus on large-scale machine learning and distributed computing, especially in projects such as Apache Spark.
- Cambridge Machine Learning Group, University of Cambridge, UK
This group has achieved remarkable results in deep learning, probabilistic graphical models, etc., and is one of the centers of European machine learning research.
Deep Learning Lab Recommendations :
- University of Toronto Deep Learning Group
The group is famous for its pioneering research in the field of deep learning, involving computer vision, natural language processing and other directions.
- Stanford Artificial Intelligence Lab (SAIL)
In addition to machine learning, SAIL is also one of the leaders in deep learning research.
- MILA (Montreal Institute for Learning Algorithms), University of Montreal
Located in Canada, it is famous for its research in deep learning and neural networks and is an important research center in the field of deep learning.
- ETH Zurich Deep Learning Lab, ETH Zurich
The Deep Learning Laboratory at the Swiss Federal Institute of Technology in Zurich has excellent research in computer vision, natural language processing and other directions.
- Tsinghua University Shenzhen Graduate School Deep Learning Laboratory
An important laboratory for deep learning research at Tsinghua University in China, covering fields such as computer vision and natural language processing.
Reinforcement Learning
What is reinforcement learning?
强化学习是一种通过试错和反馈来优化决策的机器学习方法。强化学习涉及智能体通过与环境的交互学习，以最大化预期的奖励，通常与通用的决策能力相关。随着强化学习技术的不断发展，人工智能系统将能够在更广泛的领域中进行自主学习和决策，如自动驾驶、智能机器人等。What are the main research directions?
Basic theories and algorithms : Develop new reinforcement learning algorithms, such as deep reinforcement learning (DRL), model predictive control (MPC), etc., to improve learning efficiency and stability.
Exploration and development : Design effective exploration strategies to enable the agent to better explore the environment during the learning process, thereby learning the optimal strategy faster.
Multi-agent reinforcement learning : Study the learning problem of multiple agents in a collaborative or competitive environment to achieve more complex collaborative decision-making and adversarial learning.
Deep Reinforcement Learning (DRL) : Apply deep learning techniques in reinforcement learning, such as using deep neural networks to represent value functions or policies to handle high-dimensional, complex state spaces.
Model uncertainty : Consider environmental uncertainty, including modeling of dynamic changes in the environment, and modeling of sensor errors and noise.
Transfer learning : Transfer knowledge learned on one task to another related task to improve learning efficiency and generalization performance.
Inverse reinforcement learning : inferring the reward structure of the environment from the behavior of experts, which is of great significance for imitation learning and understanding human behavior.
Practical application : Apply reinforcement learning in actual scenarios, such as autonomous driving, robot control, power management and other fields, to solve complex decision-making problems.
Interpretability and Robustness : Improve the interpretability of reinforcement learning models, make the decision-making process more transparent and understandable, and improve the model's robustness to interference and noise.
Ethical and Social Implications : Explore the ethical and social implications of the application of reinforcement learning in society to ensure the safety and sustainability of the technology.
Summarize
Reinforcement learning (RL), as a cutting-edge technology, has also made significant progress in the field of image processing in recent years. Similar to the successful application of new technologies such as reinforcement learning and generative adversarial networks (GAN) in image processing, reinforcement learning has not only made breakthroughs in traditional fields such as autonomous driving, but also in fields such as image annotation, super-resolution, and 3D reconstruction. It shows a wide range of application prospects.
Data Mining
What is data mining?
Data mining can be used to extract useful information and patterns from large-scale data sets. It has a wide range of applications and can be adapted to a variety of tasks. As computer storage capacity increases, transmission rates increase, and computing speeds increase, the amount of data transmitted on the network is also increasing. The popularity of short videos has caused the amount of data transmitted to increase exponentially. Today's era can be described as the era of big data. Data mining technology is to mine useful information for humans from massive amounts of data in order to guide decision-making in certain aspects.
What are the main research directions?
Basic theoretical research: mainly includes mining, classification, clustering of rules and patterns, etc. The mining of rules and patterns is to mine some meaningful patterns from massive data, such as discovering in supermarket shopping lists that people tend to buy beer and diapers together. The salesperson placed the beer and diapers in adjacent areas, making it easier for people to purchase and promoting sales of both items.
Social network analysis and large-scale graph mining: mainly include graph pattern mining, network relationship mining, network information dissemination, social network applications, etc. Data mining can dig out the network relationships, shopping tendencies, etc. behind people's daily use of social software data, which will help the recommendation system make accurate recommendations.
Big data mining: mainly includes algorithm parallelization, distributed expansion, multi-source heterogeneous data fusion mining, etc.
Summarize
Data mining is a highly practical subject. Learning this technology can clean out the truly useful information from massive and noisy data and use it in daily decision-making. Data mining can help supermarket sales, help merchants make profits, control online public opinion, etc., and plays a huge role in real life. Friends who are interested in the above directions can engage in data mining related research~
Laboratory recommendation
- Team of Professor Zhang Zhihua from Shanghai Jiao Tong University
- Professor Zhou Zhihua’s team from Nanjing University
- Southwest University Machine Learning and Data Analysis Laboratory
- Professor Wu Xindong’s team from Hefei University of Technology
The above is a summary of the promising research directions of AI ~ Happy learning!
By the way, let me make a small advertisement~ Siki Academy has a large number of free tutorials on artificial intelligence! Come and have a look, use your little hands and take a look, just in case these courses are suitable!
