Course Objective & Scope
Pretrained foundation models are no longer confined to text and vision: zero-shot and in-context learning models for tabular data and time series have entered practical use without hyperparameter tuning. This course examines TabPFN v2 and tabular transformers fitted on synthetic priors alongside time-series foundation families including Amazon Chronos, Google TimesFM, Salesforce Moirai and Lag-Llama/TimeGPT.
The focus is sober engineering evaluation: sample and feature-count limits, exogenous covariate support, probabilistic calibration, GPU inference cost and rigorous head-to-head benchmarks against LightGBM and statsforecast baselines on enterprise datasets.
Who Should Attend?
- Senior data scientists and ML engineers
- Forecasting, risk, pricing and analytics platform teams
- R&D and AI leads evaluating foundation models against their current ML stack
- Prerequisite: Python, GBDT/time-series experience and basic PyTorch/Hugging Face familiarity
Key Takeaways
- Run strong in-context classification and regression on small-to-medium tabular datasets with TabPFN v2
- Generate zero-shot point and quantile forecasts across diverse series using Chronos, TimesFM and Moirai
- Combine exogenous drivers (price, promotions, calendar) with foundation forecasts via residual/hybrid architectures
- Evaluate when parameter-efficient fine-tuning on domain series justifies the compute cost
- Build an adoption decision matrix comparing foundation models with GBDT and statistical baselines on accuracy, latency and cost
Curriculum
Tabular Foundation Models and TabPFN
- Prior-Data Fitted Networks (PFNs) and in-context Bayesian inference
- Using TabPFN v2, row/column scaling limits and subsampling ensembles
Time-Series Foundation Architectures
- Quantisation/tokenisation, patching and probabilistic output heads
- Comparing Chronos, TimesFM, Moirai and Lag-Llama
Zero-Shot Forecasting, Covariates and Hybrid Pipelines
- Cold-start series forecasting and context-window trade-offs
- Pairing foundation forecasts with a LightGBM exogenous residual model
Fine-Tuning and Probabilistic Calibration
- Fine-tuning on domain series while avoiding overfitting
- Conformal calibration of quantile and sample-path forecasts
Benchmarking, Cost and Production Architecture
- Fair backtesting against LightGBM, CatBoost and AutoARIMA/AutoETS
- GPU batch inference, caching and an enterprise adoption scorecard
Other Courses in This Category
-
Supervised and unsupervised learning, train/test splits, overfitting, core metrics and first models in scikit-learn, so technical teams can frame a business problem as an ML task and build a sound baseline.
- Live Course
- 5 modules
-
Gradient boosting, ensembles and stacking, hyperparameter optimisation with Optuna, imbalanced data, probability calibration and custom loss functions for experienced data scientists building production-grade models.
- Live Course
- 5 modules
-
Takes analysts beyond classic reporting with segmentation, driver analysis, anomaly flagging and simple predictive models in Python, scikit-learn and SHAP, producing analyses that tie insight to business decisions.
- Live Course
- 5 modules
-
Categorical encoding, time and window-based features, target encoding, interaction and text features, feature selection and feature stores: a systematic approach to the step that most improves model performance on tabular data.
- Live Course
- 5 modules
-
Choosing the right metrics, evaluation tied to business KPIs, error analysis, fairness checks and model explanation with SHAP, LIME and PDP, so models can be presented to stakeholders reliably and understandably.
- Live Course
- 5 modules
-
Independent model validation: data and conceptual soundness checks, backtesting, stability (PSI), challenger models, stress tests and findings reporting, producing a validation report aligned with a model risk management framework.
- Live Course
- 5 modules
Get in touch
