Topic List

In Progress

FPGA Edge AI – AOI Defect Classification

The Automated Optical Inspection, or AOI, is a fast and accurate optical inspection system based on machine vision technology. The advanced inspection method has multiple advantages over traditional inspections executed manually.AOI technique can be applied to business in multiple areas such as R&D in high tech, manufacturing, defense, healthcare, environmental protection, and electric utility with great potential to increase productivity with reduced costs and time.ITRI has dedicated in the R&D of flexible displays for decades and aims to improve accuracy of AOI systems to elevate product quality.We would like to invite data scientists from all different fields to join this competition. With the provided AOI image data to build a model to identify the correct defect type and strengthen the effectiveness of AOI inspection through data science.FPGAs (Field Programmable Gate Array) are often used to maximize application performance as workloads and algorithms evolve through reconfigurable fabric. In this topic, the participants are challenged to demostrate their ability to optimize machine learning model for FPGA data center accelerator cards (Xilinx® Alveo™ U50LV). Apart from accuracy of the model, time of model inference will also be taken into consideration. You have to upload the trained model to Aldea. The platform will load your model to FPGA accelerator card, where the model inference will be performed and scored. Aside from being familiar with AI algorithms, the ability to optimize models for AI accelerator devices such as FPGA cards is equally important in the flourishing future of Edge AI. Come accept the challenge and join the competition now! Participation FlowRegister and download training dataPrepare development environmentTrain float modelQuantizationCompilationPackage your model in a zip file. Upload it to Aldea for model inference and scoringFor detailed description, please refer to the Tutorial and Sample Code. Discussions and questions on the forum of this topic are welcomed as well. You may also send email to support@e-elements.com.tw for further questions regarding this competition.You may rest assured that your submitted model will not be used for any purpose other than the evaluation of this contest.

2021-08-26T16:00:00+00:00 ~ 2021-11-17T15:59:59+00:00
Closed

Edge AI Competition – AOI Defect Classification

The Automated Optical Inspection, or AOI, is a fast and accurate optical inspection system based on machine vision technology. The advanced inspection method has multiple advantages over traditional inspections executed manually.AOI technique can be applied to business in multiple areas such as R&D in high tech, manufacturing, defense, healthcare, environmental protection, and electric utility with great potential to increase productivity with reduced costs and time.ITRI has dedicated in the R&D of flexible displays for decades and aims to improve accuracy of AOI systems to elevate product quality.We would like to invite data scientists from all different fields to join this competition. With the provided AOI image data to build a model to identify the correct defect type and strengthen the effectiveness of AOI inspection through data science. Reference[1] https://en.wikipedia.org/wiki/Automated_optical_inspection CompetitionThis topic is mainly for testing your ability of applying NVIDIA Jetson Nano, an embedded platform. Apart from accuracy of the model, time of loading model as well as time of model inference, will also be taken into consideration. You may need to upload the trained model to Aldea. The platform will load the model to Jetson Nano, where the model inference will be done, and the score will be calculated. Instructions:Register the contest on Aldea.Download training set and train your modelPackage your model in a zip file. Upload it to Aldea for model inference and scoring. The maximum memory usage of the program on the platform is 3GB, and the maximum execution time is 20 minutes. If any limit is exceeded, it will be considered as a failure.Aside from being familiar with AI algorithms, the ability to optimize models for edge services is equally important in the flourishing future of Edge AI. Come accept the challenge now!For a more detailed description, please refer to the description file in the downloaded file. Discussions and questions on the forum of this topic are welcomed as well.description file You may rest assured that all the programs as well as models, will not be used for any purpose other than the evaluation of this contest. 

2021-05-18T16:00:00+00:00 ~ 2021-08-18T15:59:59+00:00
Closed

Employee Turnover Prediction

Employees are the most important asset of a company. For organizations to make steady growth, it is important to detect early signs of employee turnover and retain top talents. The prediction of employee turnover is based on big data and artificial intelligence to analyze possibilities of resignation. Talent retention initiatives can be activated as soon as possible when warnings are identified.Many indices of employee turnover are collected, such as age, performance records, highest academic degree, number of business trips and number of leaves, and etc. HR professionals need to evaluate the tendency of employees leaving by referring to past experiences and working conditions. In this topic, participants are asked to build models to predict whether the employee will resign by adopting the method of machine learning. PrizeThe following awards will be given to the top three participants on the Private Leaderboard whose score is higher than the baseline 0.18 and the report is submitted before 11/4.First place: 150,000 discount points of hicloud + NTD 15,000 (Tax included)Second place: 100,000 discount points of hicloud + NTD 10,000 (Tax included)Third place: 50,000 discount points of hicloud + NTD 5,000 (Tax included)Honorable mention: 50,000 discount points of hicloud (multiple winners, depends on the final result)* Hicloud discount points were provided by Chunghwa TelecomNotes:1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom reserves the right to make changes to the terms and conditions herein.3. The example of discount points calculation can be referred to https://aidea-web.tw/computing 

2020-08-04T16:00:00+00:00 ~ 2020-10-28T15:59:59+00:00
Closed

Speech Classification

Making machines understand the meanings of human speech has always been a common goal for industries and the academia. The techniques of speech recognition has been studied for a few decades, and has kept gaining momentum as the innovation in artificial intelligence grew significantly. Will there be more breakthroughs in speech recognition of Mandarin Chinese? Come participate in this challenge and show that you are an expert in speech recognition of Mandarin Chinese! The data set of this topic is selected from “artificial intelligent voice data” released by the Ministry of Science and Technology (MOST). It contains four Classic Chinese Novels: Dream of the Red Chamber, Romance of the Three Kingdoms, Journey to the West and Water Margin, as well as traffic reports by the Police Broadcasting Service (Pbs), the news broadcasted by National Education Radio, etc. There are 1,751 audio files in total and are divided into 8 categories, the participants should utilize speech recognition techniques to determine the category of each file. Source of the materials: Artificial intelligent voice data【Procedure】RegisterDownload data, Build modelUpload result (a maximum of three times per day)Select result (please click the “result select tab” on the right side of the upload page, a maximum of three files can be selected)Awards announcementNote: If no results are chosen before the upload deadline, the participant will be disqualified and will not be able to enter the final ranking stage.

2020-01-15T16:00:00+00:00 ~ 2020-04-15T15:59:59+00:00
Closed

Characteristics Analysis of Exhibition Visitors

This topic refers to the 2018 Marathon Expo, which was held in Taipei World Trade Center Exhibition Hall and lasted for three days, data contains 38,000 images taken from the camera set at the entrance.The location and gender (male, female, and others[note]) of the visitors in the image can be distinguished rapidly by adopting the object detection technique in machine learning, so gender distribution can be known after a preliminary analysis.For practical application, visitors’ behaviors and preferences can be analyzed by incorporating data from other devices (i.e. mobile phone). This information can be the reference of route arrangements and booth setup, exhibitors can also plan stands location and activities based on the preference of visitors.This topic adopts a new competition mode: Participants will upload the analysis program to Aldea, the program will be executed by the platform; original data will only be kept on the platform, and the contest is going to be proceeded with no raw data released.The purpose of this competition mode is to protect sensitive data, no data will be released on the platform, instead, the data will be read by the uploaded analysis platform. Participants can download container images and develop programs inside the container to ensure the program compatibility.Notes: “Others” in gender category refers to cases that gender cannot be judged by their appearances, such as children and infants. Competition Mode【Remote execution environment】Execution: Aldea will execute the program after analysis program is uploaded to Aldea by the participants. The contest is going to be proceeded with no raw data released.Remote execution environment is a computing environment, which contains two parts: environment for program executing (Python is supported), and data that needs to be analyzed (training sets as well as test sets). After participants upload their analysis program to Aldea, the program will be executed in the computing environment and the result (csv) will be sent to the module to calculate the scores, which will be shown on Aldea. Compared to traditional challenges (participants download the data, perform the analysis, build the model and then upload), this competition mode can protect sensitive data since all the data will only be kept on the platform.This computing environment uses the techniques of containers. Participants can download container image and develop programs inside the container.The file of container image is the same as the file in the remote execute environment so that program compatibility is ensured.Participants develop analysis programs based on sample images, which contains images and labels.Participants can improve and adjust their algorithms of analysis program referring to their score on the platform.After uploading the program, the maximum time of program execution is 4 hours per submission. 【Enrollment】There is a limit on the number of enrollments given that the competition mode and function of this topic are still in a test phase. For those who wants to enroll in this contest, please send us an email. A reply will be given after the evaluation of your related information, if you are determined eligible, then you are welcomed to join the contest.Please complete account registration on AIdea and then send an email to AIdea(admin.AIdea@itri.org.tw) with a subject of [“Characteristics analysis of exhibition visitors” enrollment], and the information below should be included: platform account, email.Related experiments of image analysis (such as other topics on Aldea or other similar contests) 【Procedure】For the enrollment, please refer to the description mentioned above.Prepare a Docker environment, and download container image (docker pull nvaitc/ai-lab:19.08), which is the program development environment.Download sample images, and develop analysis program based on the images.Name the analysis program “main.py,” and pack it into a ZIP file (i.e. prog.zip). If there are other files, pack them in the ZIP file together.The analysis program “main.py” should produce “result.csv” as output (please refer to the “Upload Format Description” for format requirements). If there is no “result.csv” files output, it will be considered as format error.AIdea’s computing section will perform score computation once “prog.zip” was uploaded to the platform, including training and test datasets.The score will be shown on Aldea after the computation completes successfully. Participants can adjust their algorithms of the analysis program referring to it. PrizeThe following awards will be given to the top three participants whose score on the Private Leaderboard meets the baseline(mAP > 0.50):First place: 150,000 discount points of hicloud + NTD 50,000 (Tax included) Second place: 100,000 discount points of hicloud Third place: 50,000 discount points of hicloud Honorable mention: 50,000 discount points of hicloud (multiple winners, depends on the final result)Gift Redemption: A specially made coaster will be given to the participants who submit their results.Gift supplies are limited, redemption of coaster is available on a first-come first-served basis.Each account can redeem 1 coaster only. ※ Participants are welcomed to test the smoothness or other related operating conditions of the platform, and are also encouraged to give feedbacks through forum discussion or admin mailbox (the forum shall have the priority). The one who gives the most valuable and the largest number of feedbacks can receive the honorable mention award! Hicloud discount points were provided by Chunghwa TelecomNotes:1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.3. The example of discount points calculation can be referred to https://aidea-web.tw/computing 

2019-12-24T16:00:00+00:00 ~ 2020-04-29T15:59:59+00:00
Closed

Marine Litter Prediction in Taiwan

Marine litter has become a critical pollution issue which is worthy of global attention. Scientific studies show that artificial waste has posed a tremendous and irreversible threat to our environment and economy.Compared to air and water pollution, the geographic scale of marine waste is far more larger, which results in difficulties for finding the connection between the pollution source and polluted area.There are high discrepancies in artificial waste in appearances, size, weights and ingredients, which makes it challenging to identify these pollutants with single detector or utilizing methods commonly used in examining heavy metals and pesticide.Although the reduction of artificial waste is the fundamental way to solve the pollution issue, however, it’s challenging and time-consuming. Thus, terminal solutions such as beach cleanup have played an important role in alleviating this crisis.Beach Litter Rapid Assessment allows researchers to conduct sampling survey in a wide range of locations in a short period of time. It is one of the examination methods that helps quantify the marine wastes and assist scientists decide the location where beach cleanup should be held. The sampling survey is conducted by each observation station every ten kilometers of the shoreline. Taiwan has 121 observation stations in the 2210-kilometer coastline.This topic focuses on exploiting the data collected by one observatory to predict the sampling survey results of nearby observatories. This method is expected to lower the amount of observation station and the involvement of human labor. Reference[1] https://www.sow.org.tw/sites/sow/files/hai_an_fei_qi_wu_kuai_shai_diao_cha_zhi_yin_190802.pdf PrizeFirst place: 150,000 discount points of hicloudSecond place: 100,000 discount points of hicloud Third place: 50,000 discount points of hicloud Honorable mention: 50,000 discount points of hicloud (multiple winners, depends on the final result) Awards were provided by Chunghwa Telecom Notes:1.Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.3. The example of discount points calculation can be referred to https://aidea-web.tw/computing 

2019-11-19T16:00:00+00:00 ~ 2020-03-04T15:59:59+00:00
Closed

Prediction of Milk yield of Taiwan Ranches

According to statistics, the expected revenue of global dairy is more than one hundred million US dollars[1], and in accordance with the data provided by DATA.COA[2], there are 557 dairy farms, 115,685 cows in Q2 2019, and that’s an average of 207 cows per farm. Date back to 2017, the annual report[3] showed that there were only 61,976 cows being bred and 419,342 tons of milk being produced. While digital transformation has became a trend, agriculture and cattle industry also start to embrace new technologies. Under the circumstance of downsizing, the average milk yield per cow in Taiwan has surpassed the average milk yield per cow in Australia, Germany and China...etc, and it is still increasing. The overall Taiwan dairy-strength are catching up the strength of the countries which have a highly developed dairy industry. (such as America and Israel… etc.)The goal of this issue is to predict the milk yield of Taiwan ranches in different regions by referring to the Dairy Herd Improvement, DHI dataset which is provided by Dairy Association Taiwan,R.O.C, hoping that these data can help grasping the key of milk yield forecasting, and thus benefits the dairy in Taiwan in the aspect of  Intellectual Management and pricing. AwardsThe first place 100,000 discount points of hicloudThe second place 50,000 discount points of hicloud The third place 50,000 discount points of hicloud honorable mention 50,000 discount points of hicloud (multiple winners, depends on the final result)Awards were provided by Chunghwa TelecomNote: 1.Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2.Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.The example of calculating points can be referred to https://aidea-web.tw/computing

2019-09-30T16:00:00+00:00 ~ 2019-12-18T15:59:59+00:00
Closed

Prediction of Marathon Expo Visiting Route

Crowd route is one of the most considerable part in the study of crowd behavior. It is not only the main reference on marketing events, but also a long-term concern of shops, department stores, exhibitions, and all kinds of campaigns.By collecting and analyzing crowd routes, we can understand the preference and the staying time of the visitors, then make the best use of limited space and provide a more proper content for everyone. Based on the analysis result, we can optimize and customize campaigns by considering the preferences of certain target audiences. In practical applications scenarios, crowd route analysis can also be used to identify potential visitors and send push notifications.The data of this topic is a random sampling of 2018 Marathon Expo, each sample had been classified into five different visiting route groups. The focus of this challenge is to train and build a favorable strategic model to correctly label visitor type by using such a great amount of route combinations. PrizeFirst place: 100,000 discount points of hicloudSecond place: 50,000 discount points of hicloud Third place: 50,000 discount points of hicloud Honorable mention: 50,000 discount points of hicloud (multiple winners, depends on the final result) * Rewards were provided by Chunghwa TelecomNotes:1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.3. The example of discount points calculation can be referred to https://aidea-web.tw/computing 

2019-07-15T16:00:00+00:00 ~ 2019-09-30T15:59:59+00:00
Closed

Nickel Price Forecast

This topic is provided by Walsin Lihwa. Walsin Lihwa Corporation is one of the leading manufacturers of communication cables and specialty steel in greater China area, dedicating in the development and manufacturing of copper wires, cables and specialty steel.Nickel is one of the most important materials of stainless steel, which is massively used in diverse industries such as automobiles, remote communication, engineering construction, chemical engineering, food processing, aviation and aerospace industry. The price of nickel fluctuates wildly between US$ 13,000 to US$ 52,300 per ton. One of the key factors to control the manufacturing cost of stainless steel is to precisely predict the price of nickel.Records such as multiple futures trading prices, industry articles and news related to the topic are provided to the participants, hoping that these data can help grasping the key of Nickel price forecasting.PrizeFirst place: 100,000 discount points of hicloudSecond place: 50,000 discount points of hicloud Third place: 50,000 discount points of hicloud Honorable mention: 50,000 discount points of hicloud (multiple winners, depends on the final result) Awards were provided by Chunghwa Telecom Notes:1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein. 

2019-07-09T16:00:00+00:00 ~ 2019-10-22T15:59:59+00:00
academic only
Closed

Trend forecast of Second-line chemical reactivity

This topic is provided by China Petrochemical Development Corporation (CPDC), whose core business is in the production of petrochemical intermediates and related engineering plastics, synthetic resins, raw materials for chemical fiber and other derivative products. The data of this topic is the production record of products, and this product is produced by adopting chemical catalytic reactions. The production record includes operating parameter (A1~A7)and forecast of product conversion(B1~B12), B4 is the main product and B5,B7 is the by-product. Personnels can determine the changes of future product conversion by referring the known operating parameter, and thus can adjust the operating parameter in order to keep B4- the main product, in the best condition. By doing so, we can reduce the usage of raw materials and improve the productivity of B4. Everyday we will go through a complete analysis assay of chemical catalytic reactions during Monday to Friday, and we will adjust the parameters refer to the product conversion of the day. By observing the product conversion daily, we can make sure that B4 can remain its best condition. On this issue, we would like to forecast the product conversion by referring the time trend, mainly the product conversion of B4, B5 and B7, so that we are more likely to adjust the parameter beforehand, to make sure the product conversion can go above the standard. reward The first place 100,000 discount points of hicloud The second place 50,000 discount points of hicloud The third place 50,000 discount points of hicloud honorable mention 50,000 discount points of hicloud( multiple winners, depends on the final result) Rewards were provided by Chunghwa Telecom Note: 1.Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically. 2.Chunghwa Telecom deserves the right to make changes to the terms and conditions herein. The example of calculating points can be referred to https://aidea-web.tw/computing

2019-06-11T16:00:00+00:00 ~ 2019-09-30T15:59:59+00:00
academic only
Closed

Forecast prediction of elements for maintenance(2)

The issue, forecast preparation of elements in routine maintenance, is to predict daily consumptions of the shared item groups within laptops. Maintenance records of applied materials and sales records of products would be used as the base to predict future consumptions. Considering the time spent from purchasing materials globally to delivering materials to maintenance base, we have daily forecast preparation offer to predict the weekly consumptions in the next 21 weeks, which enalbes customer service unit to purchase and operate transportations of elements. How to predict the consumption precisely has become a vital issue to lower the cost. This issue is for acedemia to participate only, and are welcome to lower the preparation costs by exploring future consumptions of shared items through data science.This is an open-ended problem that can be solved consistently over a long period of time, and has offered milestone awards for participants. the award will be given to the first three participants who reached the Baseline(MAPE<40), and Private Leaderboard will be updated occasionally.After the milestone is ended, date set will be updated(around October), and the scores (includes individual ones and the ranking) will be cleared.Prize InformationMAPE < 20 300,000 discount points of hicloudMAPE < 30 200,000 discount points of hicloudMAPE < 40 100,000 discount points of hicloudRewards were provided by Chunghwa TelecomNote: 1.Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.Note: 2.Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.Note: 3.For those who reached the baseline, the awards and certificates will be given after the models passed the verification.The example of calculating points can be referred to https://aidea-web.tw/computing

2019-04-01T16:00:00+00:00 ~ 2019-09-30T15:59:59+00:00
Closed

Mosquito Breeding Site Hunting for Dengue Fever Control

Dengue fever is an acute infectious disease transmitted by mosquito. The peak time of dengue fever outbreak in Taiwan is usually at summertime. Mild clinical cases of dengue fever may present as symptoms such as fever, headaches, and myalgia while severe cases may have severe fluid leakage, hemorrhagic symptoms, shock, organ failure, coma and even death. The mortality rate can be as high as 20% or more if the patient does not receive proper treatment in time.To effectively prevent dengue fever outbreak, cleaning up the breeding sites of the mosquitos is essential. Possible breeding sites for mosquitos include all containers that hold stagnant water, such as bottles, basins, buckets, cans, cups, bowls, tires, plastic bags, and etc.Every year, the Taiwan Centers for Disease Control collaborate with local health department to examine the communities and to find uncleaned sites with those containers that may hold stagnant water, where may become mosquito breeding sites afterwards. However, the inspection takes tremendous manpower and time. This challenge provides labeled data for the various types of containers, and aims to build an object detection model for possible breeding sites. This way the inspectors can pinpoint the containers which hold stagnant water by digital camera images or live video, and thus improve the effectiveness of inspection and breeding site elimination. This is an open-ended problem that can be solved consistently over a long period of time, and has offered milestone awards for participants. the award will be given to the first three participants who reached the Baseline, and Private Leaderboard will be updated occasionally. Prize InformationThe final score should be uploaded before 23:59:59 on the day of milestone. The top three exceeding Baseline score (0.595474) on Private Leaderboard would be awarded as follows:The first place: 100,000 discount points of hicloudThe second place: 100,000 discount points of hicloudThe third place: 100,000 discount points of hicloudRewards were provided by Chunghwa TelecomNote:1. Hicloud points can redeem service charge. After the discount, the price will be charged at 30% off based on the list price automatically.2. Chunghwa Telecom deserves the right to make changes to the terms and conditions herein.3. A model verification process will be performed to verify eligibility of awardees. Only those who pass model verification process will be awarded a certificate and/or prize.The example of calculating points can be referred to https://aidea-web.tw/computing

2019-01-27T16:00:00+00:00 ~ 2019-11-30T15:59:59+00:00