overview_english_Chowa Giken

CHOWA GIKEN Corporation © CHOWA GIKEN Corporation 2 Catch the forefront of evolving AI and open up the future with ideas We are a startup company born from a university laboratory on a mission to apply AI research into practical use for our society. For over ten years of experience, we are now making the most of AI knowledge and challenging to be at the forefront of the times together with our customers in various industries. 2 Copyright © CHOWA GIKEN Corporation All rights reserved. 3 Copyright © CHOWA GIKEN Corporation All rights reserved. Company Overview Company Name Chowa Giken Corporation Location Sapporo, Hokkaido (Head office) Tokyo (Branch) Establishment 4 November 2009 Number of employees 30 Capital 24,500,000 Yen Business AI Research and Development Chowa-AI Technology License AI Expert Education Representative Director Nakamura Takuya Outside Director Suzuki Keiji Outside Director Kawamura Hidenori November 2009 Established Chowa Giken Corporation July 2017 Moved office to Hokkaido University North Campus “Hokudai Business Spring Bldg.” October 2017 Authorized as Hokkaido University Venture by Hokkaido University May 2018 Established Tokyo Office January 2019 Established Bangladesh and Thailand promotion division Our Clients 4 Copyright © CHOWA GIKEN Corporation All rights reserved. Key Members 11 doctoral degree holders Hidetoshi Nonaka Doctor of Engineering Former associate professor at Hokkaido University Mathematical engineering, Human interface Outside director Kawamura Hidenori Doctor of engineering Professor at Hokkaido University Artificial intelligence, Tourism Information, Multi-Agent System, Learning / Adaptation, Optimization, Complex System CEO Nakamura Takuya Keio University Faculty of Commerce Yohko Konno Doctor of Engineering Hokkaido University Researcher Optimization, Data mining, Natural language processing Kin Shofuku Researcher Machine learning, Image analysis, Image recognition, Image classification, Framework for deep learning, Web application vulnerability diagnosis Outside Director Suzuki Keiji Doctor of engineering Professor at Hakodate Future University Artificial intelligence, Multi-agent system, Distributed autonomous robotic systems Yamashita Tomohisa Doctor of Engineering Associate Professor at Hokkaido University Service engineering, People Flow analysis, Artificial Intelligence Yokoyama Soichiro Doctor of Engineering Assistant Professor at Hokkaido University Scheduling, Optimization, Artificial Intelligence Yamashita Akihiro Doctor of Engineering Professor at National Institute of Technology (Tokyo College), Intelligent system, Embedded system 5 Copyright © CHOWA GIKEN Corporation All rights reserved. Sapporo, Hokkaido Tokyo Where we are in Japan? 6 Copyright © CHOWA GIKEN Corporation All rights reserved. Collaboration with academic research institutes Advanced information technology research Needs Hokkaido Univ. Company/Administration Hakodate Future University Hokkaido Joho Univ. Chitose Institute of Science and Technology Hokkai-Gakuen Univ. Implementation of research results Since its establishment, We have continued transmitting “Practical AI” to meet the needs of the times. By collaborating research and development with academic research institutes as well as holding a regular technology sharing meeting, we can keep updated on the latest AI research trend. 7 Copyright © CHOWA GIKEN Corporation All rights reserved. Research Environment 8 Copyright © CHOWA GIKEN Corporation All rights reserved. Our competitive advantages 1: Cutting-edge AI Technology 2: Ten years of experience and achievement 3: AI Expert and Intellectually curious people By collaborating with universities and research institutes, we can keep updated with AI logic and technology to solve any specialized challenging task. According to the lab-like environment, there are a large number of intellectually curious people gathering for working with us. After the establishment as a university based venture, we have been experienced practical research and achievement for over ten years. Copyright © CHOWA GIKEN Corporation All rights reserved. Solutions 10 Copyright © CHOWA GIKEN Corporation All rights reserved. AI Engines we have been cooperating with research institutes and universities to develop and utilize versatile AI engines. Therefore the engines are constantly updated and adopted the latest technology. By customizing these engines according to your business challenges, we can build the powerful AI that serves your specific needs and delivers a superior result. Language engine group Image engine group Numerical engine group 11 Copyright © CHOWA GIKEN Corporation All rights reserved. AI engines Sentence classification engine An engine that classifies texts into an appropriate category based upon their context. We can extract feature quantities and classify documents appropriately by using N-Gram, Word2Vec, Doc2Vec, Deep Learning, etc. Emotion Recognition engine An engine that detects and recognizes types of feelings such as joy and sadness from the expression of texts. Considering the whole sentences, RNN will learn the context and analyze the emotion. Sentence summarization engine An engine that creates a text summary by understanding and evaluating the importance of words base on morphological analysis, syntax analysis, LexRank, etc. Feature word extraction engine An engine that extracts the important phrase by using TFIDF and AIC to evaluate how important a word is to a document. Conversation engine An interactive response engine that understands the conversation content and makes a response appropriately. RNN will learn conversation data and make a response by considering the context. Object recognition engine An engine that detects the position of various objects in an image and recognizes what the image contains, such as things and faces. The combination of YOLO algorithm and classifiers such as R-CNN and SVM enables high-speed object recognition. Image classification engine An engine that classifies images into appropriate category base on their features. With CNN incorporating the latest research such as ResNet and Inception, we can analyze the image with high accuracy . Prediction engine By using Bayesian estimation, neural networks, and random forests, this engine can analyze previous and current situation to predict the likelihood in the future, such as the number of sales and visitors. Optimization engine Among the conditions, this engine searches for the best way and creates efficient schedule proposals by using optimization algorithms such as genetic algorithms and metaheuristics. Recommendation engine Based on user behavior, this engine can determine hobbies and thoughts and make a recommendation such as a coupon or event information. This recommendation engine is performed by using a memory-based method or a model-based method such as Bayesian networks. An engine that detects anomalies based on accumulated data in the past and sends us the notification. This engine performs a density ratio estimation and One-Class SVM to detect values outside the distribution of data. Abnormality detection engine An engine that converts an ordinary photo into a masterpiece as well as generates a real-image like from a drawing. Generator algorithm typified by VAE and GAN is applied. Image generation engine Copyright © CHOWA GIKEN Corporation All rights reserved. Case study Case Study 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 13 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 AILL: AI Love Navigator Conversation Navigation Favorable Navigation We have been experimenting on how "love navigation” works by learning how love develops. By navigating the progress of the relationship between men and women, the result lowers the psychological hurdles of confusing love and creates an environment where love can be grown without being hurt. https://aill-navi.jp/ Topic analysis Word hierarchy relation analysis Logistic regression Deep Learning AI will assist you and the other party chat by advising on a topic to talk and the right timing to invite him/her to a date. Therefore, you can know the other party better before you meet. AI will support you by visualizing his/her preference for you. According to that, you can decide how to approach him/her efficiently. 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 14 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Store Development by Demand Forecasting By identifying factors, such as weather or campaign, that affect store sales, we can predict the sales tendency and use the result for inventory optimization and human resources management. A big retailing company in Japan now testing this demand forecasting. Predict sales and visitor numbers from POS data using autoregressive models, random forests, neural networks, and deep learning. Store section Factor Quantification POS DATA Event Calendar (Event ・ Ad) External environment Store Section Sales forecast Store product Sales forecast Inventory / delivery schedule management / order number calculation Store section Customer number forecast Task shift management / Human allocation Model evaluation feedback Store ordering support Human resources support Latest forecast: For inventory Interannual prediction: for people and time Model evaluation feedback Business style and sales floor strategy Advertising strategy Product combination analysis Store · EC Customer analysis Store development life cycle Household information  Demand forecast  Investigate customer needs  Store development Evaluation feedback 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 15 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Dynamic pricing by Demand Forecasting Current recommended price 240[Yen/60mins] Revenue simulation Revenue increased by 22% compared to fixed price Implementation example of dynamic pricing for coin parking It is a service that predicts future demand based on factors, such as weather, day of the week and time of the day, then calculates and show the price that maximizes KPI. Forecast demand by neural network. We train the machine in advance that “to earn a profit, how much should be set for the price according to the demand.” Usage data chronological order the weather event Step1 Predict future "demand" Step2 Calculate "price" suitable for demand 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 16 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Automatic Shift work Creating for the Bus driver This is a demonstration experiment for automatic shift work creating for a bus driver. By considering complex constraints and conditions, this will help human to work more effectively. Currently, we are experimenting and developing this service with our partner company to make it available for bus companies throughout Japan. Request from staffs Staffing constraints ・ Legal restrictions ・ Restrictions on service rules ・ Expertise and technical level ・ Vacation and business trip / training ・ Constraint by personal convenience Use genetic algorithm (GA) to solve the problem of shift creation that meets all constraints Shift work optimization engine Timetable information It requires time for people to consider every constraints. Automatic shift work compared between skilled driver and inferior. 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 17 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Customer Inquiry Search Engine Usually, to respond to inquiries from a customer about products , we have to contact many departments to get the answer. This system is a game changer that helps us find the answer instantly. We are currently collaborating this project with a major manufacturer. Search for similar cases ： TF - IDF 、 Word2Vec 、 Doc2Vec 、 SCDV 、 FastText Case summary ： LexRank 、 Word weighted summarization Messaging app Web Chat Q. A. Front end Server side Azure Bot Framework Response generation Scenario Analysis Existing service Individual development Existing service 1. Model 2. Phenomenon 3. etc. API Natural language analysis Similar question search Skype for Business, etc. Azure Bot Service Countermeasure content search model Case Study DB Dialogue history User Model improvement by feedback Search for appropriate measures based on the information revealed in the dialog with the user Interact with the user according to the scenario and collect the information necessary to search for countermeasure contents 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 18 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Automatic target data extraction from receipts It automatically reads the accounting information from the image of the receipt and sorts it into structured data. An specially designed character recognition (OCR) system extracts transaction contents and accounts information. It is free from the trouble of manual entry, and it can be expected to save labor by improving efficiency and reducing typing errors. We are considering to implement it for more complex tax calculation so that it can automatically distinguish appropriate tax type that needs to be applied without any human interaction. Word2Vec SCDV (Sparse Composite Document Vectors) Original receipt Scan Identification engine image data Read the necessary information from the image Structured data Issuer 丸善 丸の内 Date 2018/07/08 Total 4860 Consum p. Tax 360 Title Food 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 19 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Bone area recognition in fresh food processing machine In bone removal process, there are several difficulties that can not be solved using only parameter adjustment in traditional image recognition. Specially, low recognition accuracy is a great problem. But, using deep learning we solved the the biggest difficulty in extraction process of the bone area, and improved the accuracy dramatically from about 70% to 92-96%. Hybrid of semantic segmentation and traditional image processing Image of bone extraction area from X-ray image In actual image, the tip of the bone area is blurred, and has many unexpected conditions, for example, a fracture are occasionally seen. Performing bone extraction process from information of rib area 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 20 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Transportation route optimization (VRP) Truck transportation route planning takes a lot of time with conventional methods. We simulated efficient transportation routes to determine which vehicles would be most efficient when traveling in which order. By optimizing the transportation route of the truck, we can expect not only the cost reductions such as labor cost and gasoline cost, but also the environmental impact reduction such as CO2 emissions. Use guided local search (GLS) to find low cost routes while avoiding local minima. Geographic data (such as your location and route information) Optimized transportation route 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 21 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Operation route optimization in the warehouse Constrained optimization problem Meta-Heuristics Genetic algorithm In the logistics industry, the growing demand due to the spread of the Internet mail order has become an issue how to efficiently load luggage in the warehouse as the base. Upon request from the manufacturer of the load carrying vehicle in the warehouse, we have developed an algorithm to minimize the patrol route while meeting the requirements such as the maximum loading capacity of the vehicle and the loading order of the safe load. Product 2 Until YY Time Shelf: DD Productｎ Shelf：PP Product 1 Large size Shelf: XX Shipping list and constraints Optimization algorithm Visit order in the warehouse Vehicle 1 Depart 9:00 AA → XX … Shipping Vehicle 2 DD → PP … Vehicle 3 XX → ZZ … 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 22 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 B U R A N D O B U R A N D O B U R A N D O Estimation engine for branded product information A major brand purchaser is developing an engine that automatically estimates product information in real time from photos to improve work efficiency. The branded product has a lot of information such as brand name, model name and model number. The time to investigate was a big cost by human assessment. Therefore, by combining hierarchical learning methods, product information is estimated successfully and work becomes more efficient. ResNet Hierarchical learning method Object recognition Text extraction Image feature similarity extraction Extract target branded product area and printed branded product name from input image Model number estimation engine Model number similar image search engine Extracted brand name Image feature B U R A N D O ・ brand name ・ Model name ・ Ref number information ・ brand name ・ Model name ・ Ref number information ・ brand name ・ Model name ・ Ref number information Similarity Degree 98% Return the same model image as the input photo and its certainty factor Brand product information estimation engine Brand product area extracted Similar image candidate list Similarity Degree 93% Similarity Degree 87% B U R A N D O 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 23 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Plant fault detection Abnormality detection, change point detection We also perform fault detection using the sensor system of the plant. In the past, if the system was shut down unexpectedly, human had to wait several hours for cooling. The machine learning based method has made it possible to make an anomaly judgment during startup or immediately after stopping, greatly reducing the time required for fault detection. Abnormality detection engine Sensor data ・ Measurement data All 286 items ・ ON / OFF data All 2038 items System anomaly degree From all items Analyze the main cause system Running Phase analysis The system operation status is divided into phases, and an appropriate machine learning anomaly detection model is designed for each phase 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 24 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Ukiyo-e style conversion Image generation engine that converts human face photos into ukiyo-e (a genre of Japanese art). The conventional algorithm has not been able to extract the features of the Ukiyo-e as human figures. In our approach, by training the data set of Ukiyo-e and human face photos, you can transform your photo into a Ukiyo-e character. Generation system framework GAN CycleGAN など 調和技研 Logo × Ukiyo-e Face photo Ukiyo-e × Conventional method approach 概要 ( 顧客・背景・課題・目的 ) 使用したエンジン 25 Copyright © CHOWA GIKEN Corporation All rights reserved. 背景技術 Quality and value collateral application of goods by block chain It is an application that enables security of quality and value by storing information such as the origin of goods, holder's history and authenticity guarantee by block chain technology. By reading the QR code, it is possible to register and check information easily. Blockchain (Hyperledger Fabric) * Commodity code, buyer, seller, amount, status, quality etc *Trading History Anti-tamper seal prevents label replacement and falsification Product QR Block Chain その他 AI Expertise Employment and Education 27 Copyright © CHOWA GIKEN Corporation All rights reserved. How we hire talented people ・ Many skilled person would like to work with us ・ Most important thing in business is Human Resource ➡ Organization that help employee spontaneously looking forward to enhancing their skill. Let’s do what you love! 28 Copyright © CHOWA GIKEN Corporation All rights reserved. Flexibility Policy of employment ■ Work environment that fits diversity ・ Female Researcher ➡ 3 out of 4 female employee are the holder of a Doctorate ・ Synergy from Senior Researcher ➡ Professor at University 、 Researcher of Sony Corporation ・ High skill Global Human Resource Employment ➡ Bangladesh 、 Thailand 、 Germany 、 China 、 Brazil ・ Reemployment ➡ Graduate student from Laboratory 29 Copyright © CHOWA GIKEN Corporation All rights reserved. There is only diversity to correspond to market change. Currently, we have members from 7 countries, couples/parent and child who work together, employees from age 67 to high school student ➡ Bangladesh based office opening plan 30 Copyright © CHOWA GIKEN Corporation All rights reserved. ■ Start from what you love We believe that “Doing what you love” is one of the most important things to encourage you to take action since it requires strong desire and commitment. Because after all, if you don’t act on it, you won’t succeed. ■ Environment To achieve doing what you love, the work environment is important. The work environment that we can talk about dream, discuss and sharing valuable knowledge as well as having communities both inside and outside company that connect and support each other’s growth. ■ Freedom to act and decide When we share the same mindset and company’s core value, we are free to take action without rules. We can decide what is best for the team to achieve the goal. So we believe that this kind of working style will help us make a better performance and enjoy turning a vision into a reality. We provide the best environment for employee Thank you Let's design an exciting future society together ワクワクする未来社会を一緒にデザインしよう！