DataTalks HFA#19: Advanced Methods in Time Series Forecasting and Classification

📢 Join us for the next Datahack Haifa meetup, hosted at Augury's offices in the city’s downtown! Dive deep into advanced machine learning techniques bridging academia and industry. This meetup features two talks: an industrial application of time series forecasting for reinforcement learning, and a novel semi-supervised approach to risk control, demonstrated through early time series classification.

• 17:00 MPB (mingling, pizza, beer)
• 17:30 Bat-Sheva Einbinder
• 18:15 Zacharie Cohen

♦ Location: Augury offices, 39 Haatzmaut Street, 1st Floor, Haifa
♦ Language: The event will be held in Hebrew
♦ Background: Basic knowledge in data science and machine learning is advised
------------------------------------------------------------------------------------------------
Abstracts:
📈 Supercharging Time-Series Forecasting for Offline RL Using Teacher Forcing
Offline reinforcement learning provides data-driven recommendations for optimizing manufacturing plant operations by leveraging historical data to guide decision-making in complex environments. To accurately evaluate and refine these policies, we employ a time series-based simulator trained via teacher forcing, ensuring more reliable future state predictions and enhanced policy performance.
Zacharie Cohen is a data scientist at Augury

📊 Semi-Supervised Risk Control via Prediction-Powered Inference
The risk-controlling prediction sets (RCPS) framework transforms machine learning model outputs into predictive rules with rigorous error rate control but relies on labeled calibration data, which can limit accuracy with small datasets. To address this, we propose a semi-supervised calibration method using unlabeled data to tune hyper-parameters while maintaining statistical validity. We demonstrate the method through experiments in few-shot image classification and early time series classification.
Bat-Sheva Einbinder is a PhD candidate at the Electrical & Computer Engineering Faculty of the Technion