Ilho Ahn Email & Phone Number

Seoul, South Korea

Data Science works at AI Platform Team.

Seoul, South Korea

Fashion style classification model- Development of a style classification model for clothes for recommendation and search- Designed a multi-mode model combining a text encoder and a image encoder- Designing a solution for training/inference strategy in a noisy dataset of subjective judgment styles that vary from person to personSearch purpose estimation model- Direct mapping of desired targets such as target brand, category, color, etc. from the user's search… Show more Fashion style classification model- Development of a style classification model for clothes for recommendation and search- Designed a multi-mode model combining a text encoder and a image encoder- Designing a solution for training/inference strategy in a noisy dataset of subjective judgment styles that vary from person to personSearch purpose estimation model- Direct mapping of desired targets such as target brand, category, color, etc. from the user's search keyword- Estimation using search logs using learn to click method- Helps search engine results produce more appropriate resultsNatural language modeling for search- To enable flexible natural language models to be used in search engines, a model is learned so that search engines can use natural language model embeddings for products and keywords.- Joint learning of search keyword encoder and document encoder for similarity. Show less

Seoul, South Korea

Data science works/ Technical leading at Lighthouse Center- http://kpmgkr.notion.site - Lighthouse center is dedicated to solving problems using AI.- We are currently focusing on complex natural language processing for the financial area.Language models for finance domain specific.- To accomplish NLP works on finance field, Pre-training a Masked Language Model(BERT, ELECTRA, DEBERTA) with specific domain data- Designing downstream tasks with the Language Model like… Show more Data science works/ Technical leading at Lighthouse Center- http://kpmgkr.notion.site - Lighthouse center is dedicated to solving problems using AI.- We are currently focusing on complex natural language processing for the financial area.Language models for finance domain specific.- To accomplish NLP works on finance field, Pre-training a Masked Language Model(BERT, ELECTRA, DEBERTA) with specific domain data- Designing downstream tasks with the Language Model like Machine Reading Comprehension.- Development document data augmentation methods for small size training dataset.- Reading order detection for organizing text from OCR results of form documents- Building a model training experiments management system that making automatically reports, visualizations and a experiments db.Vision models for finance domain specific.- Object detection for document elements- Development table/cell object detection data augmentation methods with various diversity.Building generalized document AI solution.- Auto modeling for unstructured document value extraction by stacking/designing complex inference pipeline with various language models.- Semi-structured form document data extraction Show less

Korea

Online game abuser detection modeling/serving- Machine learning/Deep learning based anomaly detecting and reporting system- Deep autoencoding gaussian mixture modeling/ Self-similarity- Ad-hoc analysis for abusing issues on live games- In game chatting advertisement detection by machine learningOnline game in-appropriate chatting(Korean) detection with AI- Adversitement for real money trading / Swear chatting detection modeling/serving using Deep learning-… Show more Online game abuser detection modeling/serving- Machine learning/Deep learning based anomaly detecting and reporting system- Deep autoencoding gaussian mixture modeling/ Self-similarity- Ad-hoc analysis for abusing issues on live games- In game chatting advertisement detection by machine learningOnline game in-appropriate chatting(Korean) detection with AI- Adversitement for real money trading / Swear chatting detection modeling/serving using Deep learning- 1DCNN and AttentionNet- Keras/Pytorch were used by modeling toolsOnline game user retention model- Prediction model for users' retention to give a special event to the users.- Ensemble model with Gradient boosting (Spark-ML) & LightGBM & XGBMPUBG PC/Mobile Analysis- Construction Business Intelligence.- Drill-down Analysis to help making decisions.- Huge user's log data handling.Online PC game platform data analysis system- Collecting game live log streams and construction of Integrated DW for analysis.- Tracking critical pc game platform Issues by reconstructing session logs.En Masse (Bluhole North America : Publisher) B.I- Billing and game log data ETL on Distributed data system(Hadoop, YARN, Spark).- B.I Reports with Tableau.- Ad-hoc B.I analysis (Inspecting game/billing issue, etc)AWS Summit 2019 Seoul- A speaker of "MMORPG ABUSER DETECTION WITH SAGEMAKER"- https://www.slideshare.net/awskorea/mmorpg-abuser-detection-with-sagemaker-krafton-krafton-aws-summit-seoul-2019-141257125Lead of the data engineering team- Leading data engineers to accomplish data analysis and engineering. Show less

Bigdata team, department of Data and search developmentMachine learning (Spark, Scikit-learn)- Personalized shop item recommendation for Coocha slide- Personalized contents recommendation for PikiCastBusiness Intelligence- PikiCast adhoc analysis- PikiCast B.I DashboardHadoop- DW ETL with spark on Hadoop ecosystem- Management Hadoop ecosystem

- Push data service- Chatting search service- Metadata repository System- Event stream service- ETL

- Tourism recommendation system (for Hana-Tour) with Mahout- App API server- Push service- iOS app