Course Objective & Scope
Hand-crafted velocity aggregates and tree models score individual transactions well, yet they miss subtle in-session clickstream patterns, long-horizon account drift and coordinated fraud rings across linked accounts and devices. This advanced course covers Transformer and GRU sequence encoders for transaction and session streams, behavioural representation learning, deep/variational autoencoders for zero-day fraud patterns and Graph Neural Networks (GraphSAGE/GAT) for relational payment and mule networks.
Production constraints stay front and centre: participants build hybrid architectures that fuse deep embeddings with GBDT scorers, train with focal and asymmetric losses, and design low-latency online inference pipelines.
Who Should Attend?
- Data scientists in banking, fintech, payments, e-commerce and insurance
- ML engineers augmenting existing GBDT/rule fraud engines with sequence and graph signals
- Account takeover (ATO), card fraud and AML analytics teams
- Prerequisite: Python, basic PyTorch and prior classification/fraud modelling experience
Key Takeaways
- Represent irregular transaction histories and session clickstreams as sequence tensors
- Train Transformer and GRU encoders to produce rich user and card behavioural embeddings
- Detect emerging fraud patterns with deep autoencoders and contrastive representation learning
- Uncover coordinated fraud rings across shared devices, IPs and transfers using GNNs
- Combine cached deep embeddings with real-time LightGBM scoring within strict latency budgets
Curriculum
Limits of Tabular Features and Deep Fraud Architectures
- What hand-crafted aggregates lose in ordering, timing and topology
- Multimodal fraud architectures combining sequence, graph and tabular signals
Transaction and Session Sequence Modelling
- Encoding card and session histories with GRU and Transformer encoders
- Time-delta embeddings for irregular events and Account Takeover (ATO) detection
Representation Learning and Deep Anomaly Detection
- Self-supervised and contrastive pretraining for customer and merchant embeddings
- Denoising/Variational Autoencoders and Deep SVDD for zero-day fraud patterns
Fraud Rings with Graph Neural Networks
- GraphSAGE and GAT on heterogeneous account-device-IP-merchant graphs
- Detecting money-mule networks and collusive fraud clusters
Hybrid GBDT+DL Pipelines and Low-Latency Serving
- Focal loss, PR-AUC/Recall@FPR and dollar-weighted evaluation
- Nearline embedding refresh paired with millisecond online GBDT scoring
Other Courses in This Category
-
Deep defect and fault detection across inspection images, acoustic/vibration signals and system logs using PatchCore/PaDiM, Anomalib, YOLO/segmentation, 1D-CNNs/autoencoders and normal-only training.
- Live Course
- 5 modules
-
Multivariate, long-horizon and probabilistic time-series forecasting with N-BEATS, N-HiTS, Temporal Fusion Transformer (TFT), PatchTST, DeepAR and TCN using NeuralForecast and PyTorch Forecasting.
- Live Course
- 5 modules
-
Large-scale deep recommendation architectures: Two-Tower retrieval with ANN (FAISS/Qdrant), sequential session recommenders (SASRec/BERT4Rec), DeepFM/DCN-v2/DLRM ranking and multi-task MMoE learning.
- Live Course
- 5 modules
-
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
Get in touch
