CL兩個(gè)塔是同一個(gè)item做兩次數(shù)據(jù)增強(qiáng)，論文里數(shù)據(jù)增強(qiáng)是同時(shí)做了mask與dropout，也可以選擇其它增強(qiáng)方式。
SSL, 20, Google自監(jiān)督學(xué)習(xí)推薦算法Self-supervised Learning for Large-scale Item Recommendations 1. Motivation 在一些item特別多的...
Roberta: A robustly optimized bert pretraining approachCitation: 1669 (2021-09-09) 1. M...
Transformer-xl: Attentive language models beyond a fixed-length contextCitation: 1326 (...
Bert: Pre-training of deep bidirectional transformers for language understanding 1. Mot...
Attention Is All You NeedCitation: 26532 (2021-09-04) 1. Motivation 重讀經(jīng)典，一個(gè)重要的起點(diǎn)。 在作者寫作...
Multi-interest network with dynamic routing for recommendation at TmallCitation: 52 (20...
Rapid learning or feature reuse? towards understanding the effectiveness of maml. Citat...
Meta-Learning in Neural Networks: A SurveyCitation: 236 (2021-08-29) 1. Proposed Taxono...
Meta-Learning in Neural Networks: A SurveyCitation: 236 (2021-08-29) 1. Motivation 一個(gè)典型...
DRN: A Deep Reinforcement Learning Framework for News Recommendation Citation: 232 (202...
Relation-aware Meta-learning for Market Segment Demand Prediction with Limited Records ...
Exploring Simple Siamese Representation Learning 1. Motivation Kaiming He[1]大神的又一力作，證明在...
SERank: Optimize Sequencewise Learning to Rank Using Squeeze-and-Excitation Network 1. ...
Deep Interest Network for Click-Through Rate Prediction 1. Motivation 本文是阿里媽媽發(fā)表在KDD18上的...
1. 期望 定義：假設(shè)離散型隨機(jī)變量的分布律為： 如果級數(shù) 絕對收斂，則稱級數(shù)的和為隨機(jī)變量的數(shù)學(xué)期望，記為。即， 設(shè)連續(xù)型隨機(jī)變量的概率密度為，如果積分 絕對收斂，則稱積分...
One Model to Serve All: Star Topology Adaptive Recommender for Multi-Domain CTR Predict...
Meta-Learned Specific Scenario Interest Network for User Preference Prediction 1. Motiv...
R-Drop: Regularized Dropout for Neural Networks 1. Motivation 想法很直接，對模型做兩次Dropout得到兩個(gè)不同...