Course Objective & Scope
Across hundreds of related series, high-frequency sensor or energy streams, long horizons and rich static/known-future covariates, deep time-series architectures offer strong capabilities. This course covers pure decomposition networks (N-BEATS, N-HiTS), autoregressive and convolutional models (DeepAR, TCN), the interpretable Temporal Fusion Transformer (TFT) and patch/variate-centric Transformers (PatchTST, iTransformer).
Using NeuralForecast and PyTorch Forecasting, participants work hands-on with quantile and distributional losses, Reversible Instance Normalisation (RevIN) and cold-start series, benchmarking every neural architecture against statistical and LightGBM baselines on CRPS, MASE and compute cost.
Who Should Attend?
- Data scientists and ML engineers working on demand, energy, financial or IoT forecasting
- Analytics teams building multi-horizon global forecasting pipelines
- Practitioners extending statistical and GBDT forecasting stacks with neural models
- Prerequisite: Python, basic PyTorch and classical time-series concepts
Key Takeaways
- Structure static metadata, observed past inputs and known future covariates for deep sequence models
- Train global N-BEATS, N-HiTS, PatchTST and TFT architectures across panel series with NeuralForecast
- Produce calibrated probabilistic distributions using quantile loss and DeepAR likelihood heads
- Handle non-stationarity and scale variation across series with RevIN and window normalisation
- Benchmark neural forecasters against GBDT and statistical baselines on CRPS, WAPE and runtime
Curriculum
The Global Deep Time-Series Paradigm
- Lookback/horizon windowing and direct MIMO vs autoregressive decoding
- Static attributes, observed history, known future inputs and RevIN scaling
MLP, Convolutional and Recurrent Models: N-BEATS, N-HiTS, DeepAR
- Basis expansion and multi-rate pooling in N-BEATS and N-HiTS
- Dilated TCNs and probabilistic autoregressive forecasting with DeepAR
Transformers for Time Series: TFT, PatchTST and iTransformer
- Temporal Fusion Transformer (TFT): variable selection networks and attention
- Sub-series patching (PatchTST), channel independence and iTransformer
Probabilistic Losses, Hierarchies and Cold Start
- Pinball (quantile) loss, MQLoss, CRPS and conformal calibration
- Zero/few-shot forecasting on cold-start series via global neural weights
Hands-On with NeuralForecast and Baseline Benchmarking
- Rolling-origin backtesting against LightGBM and AutoETS
- Neural + GBDT ensembling and an enterprise evaluation report
Other Courses in This Category
-
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
-
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
Get in touch
