Reading Guide: Artificial Intelligence (CICAI 2021, Part I)

In June 2021 the Chinese Association for Artificial Intelligence (CAAI) held its first International Conference on Artificial Intelligence, CICAI 2021, in Hangzhou, China, partly online because of the COVID-19 pandemic. Springer published the selected papers as a two-volume set in the Lecture Notes in Artificial Intelligence subseries of Lecture Notes in Computer Science (LNCS 13069 and 13070). This guide covers Part I (LNCS 13069), edited by Lu Fang, Yiran Chen, Guangtao Zhai, Jane Wang, Ruiping Wang and Weisheng Dong. Across both volumes, 105 papers were selected from 307 qualified submissions, giving a broad cross-section of AI research at the start of the 2020s.

Get the book: View CICAI 2021 Proceedings, Part I on Amazon

What this book is (and who it's for)

This is a peer-reviewed conference proceedings volume of around 800 pages. Each chapter is a self-contained research paper with its own problem, method, experiments and references. It is aimed at readers who already know the basics of AI and machine learning and want to see how researchers were applying and extending them.

Across the two-volume set, papers are organized into topical sections: applications of AI; computer vision; data mining; explainability, understandability and verifiability of AI; machine learning; natural language processing; robotics; and other AI-related topics. Part I opens with the applications section.

It suits:

• Graduate students looking for research directions, related work and examples of how papers are structured.
• Researchers and engineers in computer vision, recommendation and applied machine learning.
• Advanced undergraduates who have finished an introductory machine learning or deep learning course and want to read real research.

Before you start

You will get the most from this volume if you are comfortable with:

• Core machine learning concepts: training and test data, loss functions, overfitting and evaluation metrics.
• Deep learning basics, especially convolutional networks for images, and some familiarity with graph neural networks for the recommendation papers.
• Reading research papers: knowing how to move between abstract, method, experiments and related work.

If you want to experiment with ideas from the papers, a Python environment with a deep learning framework is the usual setup. Check each paper for links to code or public datasets.

How to read it

Do not read a proceedings volume front to back. Choose a theme, read a few papers carefully, and use their references to go further.

Start with the applications of AI section

The applications section is a good way into the volume because the problems are concrete. Its papers include, for example:

• "Comparative Sharpness Evaluation for Mobile Phone Photos," a nice entry point for image-quality assessment.
• "DiffGNN: Capturing Different Behaviors in Multiplex Heterogeneous Networks for Recommendation," for readers interested in graph-based recommender systems.
• "Graph-Based Exercise- and Knowledge-Aware Learning Network for Student Performance Prediction," which applies AI to education.
• "Odds Estimating with Opponent Hand Belief for Texas Hold'em Poker Agents," for those interested in game-playing agents and reasoning under uncertainty.
• "Object Goal Visual Navigation Using Semantic Spatial Relationships," which links vision and embodied agents.
• "Classification of COVID-19 in CT Scans Using Image Smoothing and Improved Deep Residual Network," a medical imaging application.

Suggested paths by reader

• Computer vision readers: begin with the image-quality, long-tailed visual recognition and medical imaging papers, then move to the computer vision section of the set.
• Recommender-system readers: start with DiffGNN and the remote sensing image recommendation paper, then look for related data mining work.
• Agents and robotics readers: start with the poker and visual navigation papers.
• Education and applied AI readers: start with the student performance prediction paper.

For any paper, read the abstract and conclusion first, then the experiments. Skim heavy derivations on a first pass and return to them only if you plan to build on the work.

Key ideas to take away

• AI research is broad and applied. The same deep learning toolkit appears in photography, education, medicine, games and recommendation.
• Graphs are a powerful structure. Several papers model relationships between users, items, exercises or concepts as graphs.
• Real-world data is messy. Problems like long-tailed class distributions push researchers to design new sampling and training strategies.
• Trust matters. The inclusion of a section on explainability and verifiability shows how central trustworthy AI had become.
• Proceedings capture a moment. Reading them shows what the field cared about in 2021 and helps you trace how ideas developed since.

Practice and discussion questions

1. Pick one applications paper. What problem does it solve, what data does it use and how is performance measured?
2. Why is long-tailed visual recognition difficult, and what does increasing oversampling diversity aim to fix?
3. How does modeling a recommendation problem as a multiplex heterogeneous network differ from a simple user-item matrix?
4. What makes poker harder for AI agents than games where all information is visible?
5. For the medical imaging paper, what extra validation would be needed before clinical use?
6. Choose any paper and describe one way its approach could be made more explainable.
7. Search for later work that cites one of these papers. How has the idea evolved?

Pair it with

• Part II of the set. CICAI 2021 Part II (LNCS 13070) completes the proceedings; pairing the two gives you the full range of topical sections.
• Deep learning foundations. A solid course or textbook on convolutional and graph neural networks will make most papers far easier to follow.
• Other AI proceedings in the store. The BooksPlusAi shelf also carries Advances in Soft Computing (MICAI 2021, Part II), another 2021 Springer LNAI volume, for comparing research trends from the same year.

Keep going

The best way to use this book is to pick one paper that excites you, read it closely and try to reproduce or extend a small part of it. That single project will teach you more than skimming dozens of abstracts, and the references will open up the next steps. Get CICAI 2021 Proceedings, Part I on Amazon, or browse more titles on the BooksPlusAi shelf.

As an Amazon Associate, BooksPlusAi may earn from qualifying purchases.

Sources

• Springer: Artificial Intelligence, CICAI 2021, Part I
• Springer product page (softcover)
• Technical University of Munich library catalog record