[20220330] Boostcamp-AI-Casual-Talk - 2회차

[ili0820]

News

• [GTC 2022] BMW, 가상 공장용 오픈소스 이미지 데이터셋 'SORDI' 공개
  • [GTC 2022] KT, GPT-3 뛰어넘는 언어모델 만든다

Arxiv

• Recommendation as Language Processing (RLP):A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5)

• Make-A-Scene: Scene-Based Text-to-Image Generation with Human Priors

• The Little Red Boat Story (Make-A-Scene)

• Privacy-Preserving Personalized Fitness Recommender System: A Multi-level Deep Learning Approach

• Automatic Song Translation for Tonal Languages

[hawe66]

News

GTC 2022 review: Chelsea Finn

• Scaling Data-driven Robot Learning [S42252]
  

Paper

Using deep reinforcement learning to reveal how the brain encodes abstract state-space representations in high-dimensional environments

• Human-level control through deep reinforcement learning
  • By training DQN, the AI agent could perform a human-level gameplay

1. Train DQN from human gameplay data
2. Linear Predictive Model:
   • input: (from neural activation of) DQN
   • output: (fMRI timecourse data of) human brain activation

[ijkimmy]

News

• KISTI, 연구데이터 개방·활용 위한 세계 최대 규모 '국제데이터위크(IDW 2022)' 성공 개최 위한 업무협약
• UNIST 연구팀, 소리 알아 듣는 인공피부 기술 개발...소음 환경에서도 AI 통해 95% 정확도로 사람 목소리 인식
• Ukraine uses facial recognition to identify dead Russian soldiers, minister says

Arxiv

Revisiting Neighborhood-based Link Prediction for Collaborative Filtering

• snapchat & cmu에서 진행한 논문
• introduce LINKPROP & LINKPROP-MULTI
  • a new linkage (connectivity) score for bipartite graphs (of user and item as node), generalizing multiple standard link prediction methods
• an iterative degree update process in the user-item interaction bipartite graph to exploit local graph structures without any node modeling
• non-deep learning model with only six learnable parameters
• outperforming existing SOTA GNN-based CF on four widely used benchmarks

[naem1023]

CLIP Guidance Diffusion model

Diffusion Models Beat GANs on Image Synthesis

diffusion model 노션 정리

이미지의 노이즈를 복원하는 task로 학습.

GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models

glide 노션 정리

clip으로 diffusion model의 loss를 guide. (벡터합)

novel-illustration-disco-diffusion

github

[ili0820]

고생 하셨습니다~