Those who want to know about the Japan Society for Artificial Intelligence (JSAI) 2019
Data Scientist, Machine Learning Engineer
How long does it take to finish reading this article
5 minutes
Introduction
Long time no see. My name is Tsuchiya from Macnica AI Research Center.
Actually, I participated in the JSAI 2019 National Conference of the Japanese Society for Artificial Intelligence (JSAI) held from June 4th to June 7th, 2019. I'm a little late, but From a researcher's point of view, I will summarize the state-of-the-art of Japanese AI technology.
Is there such an analysis this time? I will write a blog while mixing interesting stories that I thought, A must have for anyone interested in data science. By the way, I tried to make the content enjoyable to read while avoiding mathematical formulas.
By all means, I would like you to read it to the end (^^)
Overview of JSAI 2019
JSAI is one of the largest AI-related academic societies in Japan. It is held for a few days from May to June every year, but it was held in Aichi two years ago, Kagoshima last year, and Niigata this year!
In the past, it seems that there was a lot of theoretical content, but it seems that applied content has increased in recent years, probably because Mr. Naohiko Uramoto, who came from industry, became the president of the Japanese Society for Artificial Intelligence. Certainly, there were many sessions on applied technology development, and the question corner after the session was not a scramble from the professors, but a lot of questions from people working in the industry. It was impressive.
By the way, the number of participants has exceeded 2000 this year, and it seems that the number of participants is increasing year by year. Tickets for academic conferences overseas sell out quickly, but there seems to be a lot of demand in Japan due to the recent AI boom.
AI attracting industry attention
What I thought after participating in JSAI 2019 this time, He said that the JSAI National Conference is very effective as one of the methods to learn about the cutting-edge information of AI. Since the chairman of JSAI has an industrial background, research content that is in high demand in the industry has been selected.
In particular, research on "XAI (Explainable AI: explainability of deep learning)" attracted a lot of attention at this conference. In fact, it was impressive that "XAI" always came up in all the keynote speeches and invited speeches.
For those of you who wondered, "Explainability of deep learning?", let me explain briefly.
For example, there are some reasons why people judge a cat to be a cat. The reason we don't judge cats as dogs when we see them is because humans perceive characteristics that are different from dogs. For example, having square ears, having a long beard, or having perfectly round eyes. In recent years, AI has also been actively researched to explain "why AI made that decision" as described above.
産業では、XAIの研究のニーズも多いので、私も日々論文でキャッチアップし、実装をしております。
Also, the enthusiasm for reinforcement learning does not cool down at academic conferences. Personally, I think the need for reinforcement learning will continue to grow. I expect a little.
Attracted sessions at JSAI 2019
This time, there were a number of sessions that attracted me, but there was also a problem with the amount of text, so here I would like to summarize the two papers that were presented. The following two papers are presented.
Customer Classification by Time-Series Order Data Analysis at a High-end Yakiniku Restaurant
良品率予測と装置組合せ最適化による生産性向上
The first thing was that I found it interesting. Second, the data analysis method itself was namahage.
Customer Classification by Time-Series Order Data Analysis at a High-end Yakiniku Restaurant
Clarifying order patterns and classifying customers by analyzing order data at high-end yakiniku restaurants
[ What is amazing compared to previous research ]
There are many studies on the labor shortage problem in the restaurant industry, but there are few studies that focus on the most important "customer composition" in restaurant management.
[Where is the key to technology and methods?]
Fundamental, but in-depth analysis is key
570 products classified into 57 are divided into 12 clusters
It can be seen that the data can be explained up to the seventh principal component by PCA for 12 classified products.
Differences in principal component scores between lunch and dinner
Equipment combination optimization that simultaneously improves the non-defective product rate and shortens the make span when there is little actual data
[ What is amazing compared to previous research ]
実用実績のない装置組み合わせにおける良品率予測のためのアルゴリズム精度向上
Proposal of dispatching rules that consider not only the improvement of make span but also the rate of non-defective products
[Where is the key to technology and methods?]
In order to predict the non-defective product rate with equipment combinations that have no practical track record, we developed Field-aware Factorization Machines that express the relationships between input variables after dividing the input variables into several fields.
Highest Expected Yield Rate, Allocating jobs to equipment with the highest expected good yield rate, Sorting all jobs in order of earliest possible processing time, and determining the job processing equipment and processing order so as to maximize throughput for each process Throughput Multiple Insertion (TMI) and hybrid rules were developed by combining the left
Summary
This time, I summarized the 2019 Annual Conference of the Japanese Society for Artificial Intelligence (JSAI) from the perspective of a researcher. What did you think?
JSAI also has theoretical content and practical content that will make you think "I see!", so please take a look at the original text.
Latest Information
Case Study
Blog
Document List