DeepRecSys, лекция 6: Нейросетевое ранжирование II скачать в хорошем качестве

DeepRecSys, лекция 6: Нейросетевое ранжирование II

Lecture: Kirill Khrylchenko Seminar: Artem Matveev This week we continue our exploration of deep learning–based ranking models. During the lecture, we discuss: 1. Feature interaction modeling — from linear models and factorization techniques to modern architectures such as DCN-v2 and transformer-based interaction models 2. Deep layers — what architectures are actually used in practice instead of plain MLPs (spoiler: ResNets and DenseNets) 3. Multi-task learning for ranking — including the Multi-gate Mixture-of-Experts (MMoE) architecture and the Entire Space Multi-Task Model (ESMM) 4. Positional bias and the main techniques used to mitigate it 5. Knowledge distillation and how it is applied in large-scale ranking systems During the seminar, we have an in-depth discussion of the DCN-v2 paper - DCN V2: Improved Deep & Cross Network and Practical Lessons for Web-scale Learning to Rank Systems.