Kakao Reveals AI Technology Sessions at @if(kakao)2021
Key point
Kakao will reveal eight AI technology sessions at the @if(kakao)2021 conference, including inference 10 times faster than TensorFlow Serving and BERT training optimization.
Details
Kakao has revealed AI-related technology presentation sessions at the @if(kakao)2021 conference. Sessions for developers can be found under the Day2 tab, and a limited edition emoticon is given upon pre-registration.
Key Technology Sessions
The presentations cover practical optimizations and application cases such as the following.
- Serving 10 times faster than TensorFlow Serving: Shares the process of improving inference server performance by 10 times through optimizations from various perspectives, including code, domain, and Layer Fusion.
- Building models 10 times faster than BERT: Introduces modeling and serving experiences that improved training speed by 10 times compared to existing BERT while also enhancing performance.
- ML-based gift certificate distribution: Explains feature generation and selection, as well as precautions for applying to real services, for distributing gift certificates for free episode viewing on Piccoma.
Service Application and MLOps
The development story of KakaoTalk Smart Message Service covers data processing and ad message transmission technology using Kafka Streams. The Kakao Pay session shares experiences of applying proprietary Object Detection and OCR models for non-face-to-face account opening.
Kakao Mobility shares key points from the early adoption of MLOps and experiences with deep learning-based road speed prediction (Traffic Time/Speed Prediction), while also revealing cases of ML Lifecycle automation in response to changes in the service environment.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.