LinkedIn has developed and deployed CADET, a decoder-only transformer model for predicting ad click-through rates (CTR). This new model significantly outperforms their previous LiRank baseline, achieving an 11.04% CTR lift in A/B testing. CADET incorporates several innovations, including a context-conditioned decoding architecture, a self-gated attention mechanism, and timestamp-based positional embeddings, all engineered for efficient industrial-scale deployment. AI
IMPACT This deployment demonstrates the effectiveness of transformer architectures in specialized domains like ad CTR prediction, potentially influencing future recommender system designs.
RANK_REASON Deployment of a new model within an existing product by a major tech company.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
