Course Objective & Scope
When catalogues reach millions of items and user intent shifts click by click, classical matrix factorisation hits a ceiling. This advanced programme covers the end-to-end deep learning stack of modern industrial recommenders: Two-Tower retrieval trained with in-batch negatives and served via ANN indices (FAISS/Qdrant), sequential session modelling with SASRec and BERT4Rec, and feature-crossing ranking networks including DeepFM, DCN-v2 and DLRM.
Multi-task optimisation balancing clicks, conversions and engagement via Multi-gate Mixture-of-Experts (MMoE/PLE) and injecting multimodal text/image embeddings to eliminate item cold start are practised hands-on. Participants build a two-stage deep recommender prototype in PyTorch.
Who Should Attend?
- Data scientists and ML engineers building recommendation and personalisation in e-commerce, media, marketplaces and banking
- Search, ranking and AdTech machine learning engineers
- Teams upgrading classical recommenders to real-time deep retrieval and ranking
- Prerequisite: Python, PyTorch and foundational recommendation-system knowledge
Key Takeaways
- Train Two-Tower user/item encoders with hard and in-batch negatives and serve them via FAISS/Qdrant ANN search
- Model in-session clickstreams with SASRec and BERT4Rec to predict next-item interactions
- Combine high-cardinality sparse IDs and dense features in DCN-v2 and DeepFM ranking networks
- Optimize competing objectives (CTR, CVR, dwell time) jointly using MMoE while correcting position bias
- Eliminate item cold start by fusing LLM/vision content embeddings into retrieval and ranking towers
Curriculum
Industrial Multi-Stage Deep Recommender Architecture
- The Retrieval -> Pre-ranking -> Ranking -> Re-ranking funnel and latency budgets
- High-cardinality ID embedding tables, feature hashing and memory layout
Two-Tower Candidate Retrieval and ANN Search
- User and item towers, InfoNCE, in-batch negatives and logQ frequency correction
- Sub-millisecond approximate nearest neighbour search with FAISS/Qdrant (HNSW, IVF-PQ)
Sequential and Session-Based Recommendation
- GRU4Rec, SASRec (causal self-attention) and BERT4Rec (masked item modelling)
- Combining short-term session intent with long-term user preferences
Deep Ranking Networks: DeepFM, DCN-v2 and DLRM
- Explicit and implicit feature crosses (Cross Network v2, Factorization Machines)
- Target attention (DIN) weighting user history relative to the candidate item
Multi-Task Ranking (MMoE), Cold Start and Debiasing
- Joint CTR + CVR + engagement optimisation with Shared-Bottom, MMoE and PLE
- Multimodal content embeddings for item cold start and position-bias correction
Other Courses in This Category
-
Text preprocessing, Turkish morphology, regex, tokenisation, TF-IDF, word/sentence embeddings and building first text classification, NER and sentiment pipelines with spaCy, scikit-learn and Hugging Face.
- Live Course
- 5 modules
-
Advanced NLP under scarce labels: domain-adaptive pretraining (MLM), DeBERTa/RoBERTa fine-tuning, LoRA/PEFT, few-shot learning with SetFit, weak supervision and label cleaning (Cleanlab), long documents and cross-lingual transfer.
- Live Course
- 5 modules
-
End-to-end neural NLP covering subword tokenisation, BiLSTM-CRF and Transformer (BERT, DeBERTa, T5) architectures, with hands-on sequence tagging, seq2seq summarisation and question answering in PyTorch and Hugging Face.
- Live Course
- 5 modules
-
Running NLP models in production at low latency and cost: ONNX/Optimum quantisation, Triton/FastAPI serving, dynamic batching, PII redaction (Presidio), behavioural testing (CheckList) and text drift monitoring.
- Live Course
- 5 modules
-
Turning unstructured documents into structured tables and JSON: hierarchical and multi-label classification, GLiNER and Transformer NER, relation and event extraction, layout-aware PDF parsing and schema-validated extraction.
- Live Course
- 5 modules
-
Enterprise search beyond keyword matching: Sentence-Transformers (bi-encoders), multilingual embeddings, BM25/SPLADE + dense hybrid retrieval, Reciprocal Rank Fusion, Cross-Encoder reranking, Qdrant/pgvector and NDCG@k evaluation.
- Live Course
- 5 modules
Get in touch
